EGE Football — Career Simulation

A football career simulation that follows six players across six seasons — 2018 junior year of high school through their college careers to the 2023 NFL Draft. The site is part record book (schedules, results, box scores) and part game (each player logs in to their own portal and spends offseason workouts to raise their overalls).

Status: in progress. The homepage player select is built; schedules, stats, and the portal are not. This README is the source of truth for what gets built and in what order.

---

The Six Players

Player	High school (2018–19)	League	College (2020–)	Conference	NFL (2023–)	Position	Plays like
Andrew Parr	Wake Forest High School	Northern 4A	Alabama (2020–22)	SEC	Detroit Lions, NFC North (No. 18)	TE	a complete tight end
Cooper Clark	Carlsbad High School	Avocado League	USC	Pac-12	—	RB	a receiving back
Paxon Hatch	Bloomington High School	Big Twelve	North Dakota State (2020–21), North Carolina (2022–)	Missouri Valley, then ACC	—	WR (TE through 2019)	a full-time receiver
Isaac Vitel	Bloomington High School	Big Twelve	Illinois (2020–21), Texas (2022–)	Big Ten, then Big 12	—	QB	a deep thrower
Sam Stogsdill	Normal Community High School	Big Twelve	Ohio State (2020–22)	Big Ten	Seattle Seahawks, NFC West (No. 52)	RB	a power back
Jaykeb Stewart	Naples High School	6A District 12	Ohio State (2020–22)	Big Ten	Houston Texans, AFC South (No. 2)	QB	a pocket passer

Isaac and Paxon were high school teammates, which the data has to respect: they played the same fixtures, carried the same scorelines, and every ball Paxon caught was one Isaac threw. His completions, attempts, yards and touchdowns each start at Paxon's and go up from there. In college they split up, and Sam and Jaykeb become teammates at Ohio State instead — the same rule holds for them from 2020.

A player's school is team (the high school) and college in data/players.js, both keys into EGE.teams; EGE.teamFor(player, season) picks one by the season's tier on the ladder, so a 2019 page still says Wake Forest High School after Andrew has gone to Alabama. A transfer is transfers, the school he moved to keyed by the season he got there — Isaac's { 2022: 'texas' }, Paxon's { 2022: 'northCarolina' } — so his 2021 page still says Illinois or North Dakota State.

Being drafted is a transfer too, to an NFL team: Sam's { 2023: 'seattle' }, Andrew's { 2023: 'detroit' }, Jaykeb's { 2023: 'houston' }. Those three teams carry tier: 'nfl' in EGE.teams, and that is what puts a player in the NFL for a season: EGE.tierFor(season, player) says 'nfl' for him even though the ladder's 2023 is a college season, because Paxon, Cooper and Isaac are still in college that year. EGE.leagueOf(player, season) turns that into the league his week goes out with, 'nfl' or 'college', and EGE.levelFor(player, season) into the words on his page, NFL or College Football. Sam and Jaykeb are no longer teammates from 2023, and the shared-fixture rule stops with them. The 2024 draft does the same for the other three: Cooper's { 2024: 'chicago' }, Paxon's { 2024: 'newYorkGiants' } and Isaac's { 2024: 'tampaBay' }.

Everything marked TBD is genuinely unknown right now and should stay TBD in code and data until it is confirmed — no placeholder guesses that later read as facts.

Headshots live in headshot/, keyed by last name: parr.png, clark.png, hatch.png, vitel.png, stogsdill.png, stewart.png. Replacing one under the same name? Change EGE.headshotVersion in data/players.js as well: Discord keeps its own copy of any image a post links to and goes by the address, so the Discord posts only pick up a new picture once that version (on the end of the address) changes. School marks live in icon/, keyed by school: carlsbad.png, bloomington.png, normal.png, naples.png, wake.png. The colleges use the marks in icon/offers/, cut down to 160 pixels on their long side so a page of them loads fast. A team without a mark renders its school line without it.

---

Timeline

Nine seasons, one per simulated year. This ladder is official — schedule and stat data is authored per season below.

Season	Level	Class
2018	High school varsity	Junior year
2019	High school varsity	Senior year
2020	College football	Freshman year
2021	College football	Sophomore year
2022	College football	Junior year
2023	NFL & college football	Rookie year (NFL), senior year (college)
2024	NFL	Second year (Sam, Andrew, Jaykeb), rookie year (Paxon, Cooper, Isaac)
2025	NFL	Third year (Sam, Andrew, Jaykeb), second year (Paxon, Cooper, Isaac)
2026	NFL	Fourth year (Sam, Andrew, Jaykeb), third year (Paxon, Cooper, Isaac)

The simulation opens on the 2018 junior-year high school season — that is what gets built first. High school graduation is spring 2020, after the 2019 senior season.

NFL Draft. The 2023 NFL Draft is held the spring after the 2022 junior college season. Three seasons removed from a spring 2020 high school graduation makes them draft-eligible, so declaring after junior year is the canonical path. Three declared: Jaykeb went 2nd to Houston, Andrew 18th to Detroit and Sam 52nd to Seattle. Paxon, Cooper and Isaac took the 2023 senior college season instead, which pushes them to the 2024 draft.

So 2023 is split: half the six in the NFL and half in college, in the same season. Each league keeps its own week numbers (the NFL's 1–18, with a bye, the colleges' 0–13), its own standings, its own postseason and its own Discord channel. In the NFL a player's class is how far into his career he is, Rookie Year in 2023 (EGE.nflYearFor).

Redshirts. The ladder's class is what a season is for a player who never redshirts. Isaac redshirted his freshman year (redshirt: 2020 in data/players.js): 2020 used no eligibility, so it reads simply Redshirt, and from then on he is the year of eligibility he is in, a year behind everybody else — Freshman Year in 2021, Sophomore Year in 2022. EGE.classFor works it out, and his Teams card says RS in 2020 and FR in 2021.

So a class is only ever named on a player's own page — the strip across his header and his season switcher, both his own. Everywhere else a season is a year and a level, 2020 · College Football: the home page, the Teams pages, the admin page.

---

Site Structure

Homepage — player select

• The landing page is a six-way player picker: one card per player with their headshot, name, school, and position (or TBD).
• Selecting a card routes to that player's page.

Loading screen

From the first paint until the published weeks and the purchases have come back from Supabase, the whole site is blurred and icon/football_loader_preview.gif spins in the middle. Signing in raises it again while that player's wallet loads. If the script never gets far enough to take it down, the stylesheet does after eight seconds.

Player page — /#{name}

Routed by player name, e.g. #andrew-parr, #paxon-hatch. Each player page holds:

• Header — headshot with height and weight joined to the foot of it, then school and name. On a college season the school is a link — an arrow after it, the name underlined when pointed at — to that school's team page, opened on the same season. Where the team stands in its conference is on the team page, not here. Under the name is the recruiting line: 247-style stars out of five and his rank at his position in his state and nationally (★★★★☆ #1 RB in Illinois · #5 nationally), from EGE.recruiting in data/offers.js — on the high school seasons only; a college season keeps the line's space empty, so nothing under it moves. Under that, four facts: Position, Record (the team's wins and losses over the season on show, 0-0 until the first result is out, followed by the current run — 6-1 · W4 Streak, green for wins and red for losses — until the team's season is over), Conference and Jersey. In high school Conference is just the league's name (Avocado League); in college it is where the team stands in its division as of the published weeks — 3RD in SEC West — and TBD until the team has played a conference game. It moves week by week: the six's own games come from the season file, and everybody else's from data/conferences.js (see Conference standings below). A season is over once it is behind the live one, or once the team has played the last postseason game in its file — beaten in a playoff, won its bracket or played its bowl (a college team that loses its conference title game still has a bowl to come) — or, for a school that missed the playoffs, once their first week is out. The season, his class (see Redshirts) and the level are across the orange strip at the top of the panel. The overall is on the right of the name and the facts, in its tier's box (see Overall tiers below) drawn big, and joined under it the way the height and weight hang off the picture — the same ink and orange — This season: a stock ticker, an arrow and the points the overall has climbed since the live season began. The pair runs the full height of the header. The header is the same size in every season, high school or college: the recruiting line's space is kept in college, both kinds of season have the four facts, the picture grows to meet the words beside it, and on a phone the orange strip holds one line. On a phone they are a band across the panel under the picture and the name, the overall on the left and the progression on the right. Both wait for the shop's boosts to land before they show. The college offers are no longer drawn on the header; who offered whom is still recorded in data/offers.js.

The schedule, the game log, the ratings and the trophy case each fold away behind an arrow in their heading.

The season shown is the live one. There is one season on the ladder so far, so there is nothing to switch between; shownSeason() in js/app.js is the one place that decides, and a way back to an older year is a change to it and nothing else.

Teams — #teams

A tab beside Players, public like the player pages, and laid out like them. College and the NFL — the high schools are on the player pages. In a season that has both (2023 on), the NFL's teams come first, then a dotted line across the page (.ege-teamgrid__divider), then the colleges, with no label on either (Sam took the NFL and NCAA chips off, 2026-10-17). A season with only one league has no line.

The index, #teams, is a stack of cards, one school to a row: its mark on the left, on the school's own colour (ground in EGE.teams — Alabama crimson with the A in white, USC cardinal (#9D2235), NDSU green, Illinois orange, Ohio State white, Texas burnt orange with the longhorn in white, North Carolina navy, the Seahawks navy, the Lions Honolulu blue and the Texans deep steel blue, each NFL team's mark in its full colours with its white edge, from icon/nfl/ (drawn in white they were only silhouettes); the team page's header uses the same), then its record and place in its division, the faces of the six who play there, and View team →. On a wide screen those run across the row — the name and record on the left, each of the six as a face with his name beside it in a column of its own, so they line up from card to card, and View team → on the right edge; narrower, they stack under the name and the faces go without the names. Above them is a season switcher across the full width of the page, as on a team's page — every college or NFL season, with ‹ and ›, starting on the live one.

A team's page, #teams/{school} — #teams/ohio-state — has what a player page has above it: ← All Teams and the season switcher, a drop-down of every college season that school had one of the six, with ‹ and › either side. The season is shared between the index and every team's page: pick 2021 on either and it is 2021 on both, until it is changed again (a school that had none of the six that season opens on the latest season it did). An admin also sees a season whose file is in but which has not been rolled over to yet, marked Preview, so 2021 can be looked over now. Under the bar:

• The header — the mark, the name, the record and streak, the place in its division, and the six on it, through to their pages.

• Roster — every quarterback, back, receiver and tight end on the real roster that season, a room each, best overall first, so a room reads as a depth chart and the six land wherever their overall puts them. The rooms run two across (one on a phone), so every name fits on one line. Every player is a card — a big headshot (ESPN's, or initials where ESPN has none), name, number, class (FR, SO, JR, SR, or GR for a fifth year), height and weight, and an overall in the same box the six's is drawn in. One of the six is picked out in orange and goes through to his page. No season stats are shown; they are in the overall, in the same tiered box as everywhere else (see Overall tiers). Four show and the rest fold away behind N more, but never past one of the six. From data/rosters.js, which tools/build-rosters.js writes; EGE.rosterFor in data/games.js puts the six in. A season with no roster yet borrows the latest one.

  Nobody shares a number. The six keep theirs. A real player wearing a number already taken — one of the six's, or a better teammate's — gets the nearest number his position wears that nobody on the team has (a tight end the 80s first, a back the 20s to 40s, a quarterback or receiver the teens): Andrew is Alabama's 87, so Miller Forristall is their 85. Every real player's own number is kept before anybody is moved, so moving one never moves another. ESPN only keeps a player's latest number and headshot, so for a player who later transferred out (left: true) they are his next school's; a teammate who stayed keeps a shared number before him — Bryce Young is Alabama's 9, and Jahleel Billingsley, whose 9 is from Texas, moves. The number ESPN has is on the tooltip.

  The overalls. Nobody publishes a rating for every college player, so one is worked out for each, on the six's scale but topping out at 84 — a Heisman season — with most players between 50 and 75:

  overall = 50 + 34 × (0.25 × talent + 0.10 × experience + 0.75 × production)
  

  kept within 40–84. Production carries most of it: his season — scrimmage yards plus 20 a touchdown — against a Heisman-calibre one at his position (5,000 for a quarterback, 2,000 a back, 1,800 a receiver, 1,000 a tight end), no more than all of it, or four fifths of the season before when that was better, since a quiet year does not make anybody worse (and the FCS's spring 2020 is not on ESPN). Talent is how highly he was recruited — the 247Sports Composite rating out of school, 80 and under counting nothing and 100 everything, nothing for a walk-on. It is only an ingredient; no recruiting number is shown. is his year of college. The weights add up to more than one, so a Heisman season reaches 84 whether or not he was a five-star, and nobody gets near it on recruiting alone. An FCS school's score counts 85%. A walk-on freshman is a 50, a five-star freshman who has not played about 58, a good starter about 70; DeVonta Smith's 2020 and Bryce Young's 2021 are 84s, C.J. Stroud's 2021 an 82. The one exception is Bryce Young's 2022, set by hand to an 86 — the only overall on any roster above 84. The constants are at the top of .

The quarterback rooms are also where a QB Connection is picked from (see the shop).

Search — #search

A magnifying glass at the end of the nav links (Sam, 2026-10-26), for everybody, signed in or not. It opens a page with one box; results come as it is typed, and the query rides in the address (#search/{query}), so Back from a team or a player's page comes back to the same results.

• Teams first: every NFL team (by name, city or abbreviation: DET) with its record and place in its division, and every school one of the six played for. Each is a link to its page.
• Players: the cards a team's page draws (nflLine), each with his team's mark beside his overall, opening the same pop-up. They are the NFL sim's, from the rosters the Teams pages open on (the latest live NFL season): everybody on a roster, then free agents (FA where the team mark goes; the pop-up's contract says Free agent), then the retired: greyed, RETIRED where the overall was, the mark of the team he last played for, and in the pop-up Retired after the {year} season. The six are always in it. Up to 24 players, then Show more.
• A name matches when every word typed starts a word of it ("jos all" is Josh Allen), ignoring accents, dots and apostrophes ("cj" finds C.J.); ties go to rostered players before free agents before the retired, then the higher overall.
• Nothing typed (Sam, 2026-10-27): every player best first (rostered and free agents together, the retired left out), then every team, the NFL by record and then the six's schools, as many across as fit.
• Position and sort, two drop-downs under the box, for the players with or without a query. A position narrows them to it and hides the teams. The sort is Overall (as typed: the best match first), Youngest, Oldest, Draft (the newest class first, in pick order, undrafted last), or any rating the players have, best first, A to Z under Ratings: one per name (Speed is everybody's speed; Deep Ball is a quarterback's, Deep Routes a receiver's or tight end's; a kicker's accuracy is Kick Accuracy), and with a position picked only its own. Anything but Overall puts the number sorted by in a box just before his overall (AGE, the draft's year and round over its pick, or the rating's short name), the retired after everybody else and players without the number last. On a phone, or two narrow cards across, both boxes shrink and the team's mark sits on the corner of his picture.

The free agents and the retired come with the published rosters (freeAgents and retired in nfl-sim/lib/site.js publicRosters). Rosters published before 2026-10-26 have neither, so until the live season's rosters are published again only rostered players turn up; an admin's first visit to the search or an NFL team's page publishes them again from the world, as it does for anything else the rosters are missing. Players who retired before then lost their archetype and size in the world, so their cards show neither.

Position-driven stat lines

A player's position decides which stat columns their game log shows. Stat sets to define per position group:

• QB — completions/attempts, passing yards, passing TD, INT, rushing yards/TD.
• RB — carries, rushing yards, rushing TD, receptions, receiving yards.
• WR / TE — targets, receptions, receiving yards, receiving TD (Paxon Hatch is TE, so this is the first one needed).
• OL — snaps, pancakes, sacks allowed.
• DL / LB — tackles, TFL, sacks, forced fumbles.
• DB — tackles, pass deflections, INT, INT return yards.
• K / P — FG made/attempted, longest, punts, punt average.

Only the TE set is required for the first build. The rest get filled in as positions are confirmed.

---

Overall tiers

Every overall box on the site — the six's cards on the home page, the team rosters, and the big one in a player's header — is drawn by overallBox in js/app.js, cut from whatever its number has earned:

Overall	Tier	Looks like
0–74	Basic	the site's dark chip: OVR in orange, the number in cream
75–79	Bronze	brushed bronze, the number in dark brown
80–84	Silver	brushed silver, the number in near-black
85–89	Gold	brushed gold with a soft glow
90–94	Diamond	a princess-cut blue diamond drawn as an SVG, the number in navy
95–98	Pink Diamond	the same stone turned pink
99	Galaxy	deep space with pink and blue nebulae and stars, the number glowing white

Every tier from bronze up has a band of light sweeping across it every few seconds; the diamonds and the galaxy also sparkle in the corner, and the galaxy's glow breathes. All of it stays still for anybody who asks for less motion. The colours are in site.css under the overall tiers.

Ratings

Every player has a shape as well as a number. The shapes are made by moving points within the weighting rather than adding them, so reshaping somebody never changes their overall: Isaac's deep ball came out of his short accuracy, Sam's trucking came out of his spin and juke, and Cooper's hands came out of his power running. Each one was checked before it was written — all three overalls are exactly where they were.

Every player carries the same 32 attributes, in four groups: General, Passing, Receiving, Ball Carrier. Tight ends carry a fifth, Blocking, since they are the only position here whose overall should turn on it — the group is simply absent for everyone else, and their pages don't show it. A group scores as the plain average of the attributes inside it; that is the number beside each group on a player's page.

A player's page shows only the groups their position is judged on, under the names that position uses: a quarterback gets General, Passing and Carrying; a back gets General, Receiving and Carrying; a tight end gets General, Catching, Blocking and Carrying. EGE.positionGroups decides what is shown, separately from EGE.positionWeights, which decides what counts.

The overall works the way Madden's does. It is not an average of everything on the page — under that, the only way to a 99 was a 99 in every attribute, throw power included for a tight end. Instead each position names its key attributes and how much each one matters (EGE.positionWeights, per attribute), and nothing else counts at all. Those are averaged by weight, and the average is stretched away from 50:

overall = 50 + 1.4 × (key-attribute average − 50) − position offset, kept within 1-99

So a player whose key attributes are all in the 80s and 90s is a 99 — the Madden 99 tight end this was checked against (99 catching, high 80s speed and routes, blocking in the 60s, a 33 throw power) comes out at 99. A tight end's key attributes are his hands, routes and athleticism, with blocking behind them; a quarterback's are his arm and accuracy; a back's are his speed, vision and ball carrying, with his hands counting for something since these are backs who catch. The stretch lives in EGE.overallScale. Only QB, RB, WR and TE are weighted; anything else falls back to DEFAULT, a little of everything. There is no awareness: nothing on the site could raise it, so it was taken out of the attributes and every overall (Sam, 2026-10-26). To keep the overalls where they were, each position takes off a fixed amount after the stretch (EGE.overallOffset: QB 7.3, RB 2.4, WR 3.5, TE 3.7), what awareness had cost the six on average.

The same attributes score very differently by position, which is the point:

Isaac Vitel's ratings, scored as	Overall
QB	59
RB	25
WR	17
TBD	31

Anything bought in the shop lands on top of these: EGE.valuesFor adds the boosts to the base numbers before any group or overall is worked out, so a purchase moves the rating the moment it is made.

Moving to the Madden-style overall changed nobody's overall. Scored the new way, the ratings locked at the end of the 2018 season would have put both quarterbacks at 74 and everyone else a few points up, so every attribute a player has was scaled down by the same proportion — about 16% for Isaac, 14% for Jaykeb, 3-7% for the rest — until each landed exactly where they were:

Player	Position	Overall
Jaykeb Stewart	QB	61
Isaac Vitel	QB	59
Cooper Clark	RB	56
Andrew Parr	TE	55
Sam Stogsdill	RB	54
Paxon Hatch	TE	51

Scaling rather than subtracting keeps each player's shape: strengths are still strengths, and nothing that was low has been pushed to the floor.

---

Discord scores bot

bot/post-week.js posts one week of the regular season to a Discord webhook, four times a day, walking the season out slowly:

## Week Three
{one embed per player with a game that week}

The postseason counts from one again:

## Playoffs Week One
{one embed per player playing that week}

A college season is headed the same way, ## Playoffs Week One and on, though its weeks are conference title games and bowls as much as playoff rounds: each embed names its game beside the opponent, W 36-34 vs. Florida · SEC Championship.

Which weeks those are is read off the games — EGE.playoffWeeks in data/games.js is every week with a playoff: true game in it, and a week's place in that list is its round. Nothing has to be told where the regular season ends, and a season whose playoffs open at week 12 or week 15 names itself correctly.

A player on a bye is left out, the same as a player with no game that week: there is nothing to post about somebody who is not playing.

Two bots from 2023. With half the six in the NFL and half in college, each league has its own channel and its own webhook, and a week goes out as one message per league, the NFL's first (buildWeekPosts in js/discord-post.js). Each carries only that league's players (EGE.leagueOf) and is headed by that league's own week, so the NFL's ## Playoffs Week One is counted from the NFL's own postseason and not the colleges'. A league with nobody playing that week sends nothing: week 0 is college only, and weeks 14 to 18 are NFL only. A game not yet played says Division or Non-division in the NFL where college says Conference. The high school seasons and 2018–22 belong to the college bot, the one channel there was before anybody was drafted.

An embed reads top to bottom as:

Author	the school, with its mark. Not a link — there is nothing on the site to send anybody to for a school
Headline	W `28-3` vs. Millbrook, and the one link in the embed: the player's own page. The score is in backticks, so it sits in a box
Block	the whole stat line as a four-line grid, three numbers to a line, columns aligned
Big plays	a second block, only when the game has any: one - line per play, typed in on the admin page
Fields	three summaries, the number above its heading
Thumbnail	the player's headshot. The post is about him, not his school
Footer	EGE Football Simulation
Timestamp	when the game actually kicked off
Spine	green for a win, clay for a loss, gold for a game not yet played

The headline and the block are the embed's description rather than its title, because a title renders as flat text — the score would lose its box and a code block could not go under it at all.

The block carries the typed stats — the ones in stats/{year}.js. The averages, the totals and the passer rating are left out, because they follow from the numbers beside them rather than standing alongside them.

The three summaries are the headline numbers for that position, most of what he does first:

QB	Touchdowns/INT	Passing	Rushing
RB	Touchdowns	Rushing	Receiving
TE, WR	Touchdowns	Receiving	Rushing

Passing and receiving read as what he did with what he was given — 9/18, 118YDS and 5/9, 52YDS are the same shape. Carrying has no attempts to fall short of, so it says how many rather than how many of how many: 19 CAR, 159YDS. Nothing to report collapses to a nought rather than spelling out zeroes. Every summary is set in a code span, so it sits in the same dark box as the score.

The big plays are the game's bigPlays list from stats/{year}.js, typed in one to a line in the season editor. They are word-wrapped to the block's width with a hanging indent, so a long play stays readable as one play. At most five go out, each cut at 100 characters, because Discord refuses a message with more than 6000 characters of embed text in it.

The number is the field's name and the heading is its value, because Discord draws a name above its value and the number is what should be read first.

A game that has not been played keeps the same shape: the kickoff, home or away, and whether it is a conference game fill the three block lines, and the three summaries show dashes.

A game the player missed hurt is marked in stats/{year}.js with injured: true and what it was in injury — injury: 'Bruised Shoulder'. Its embed keeps the headline (the matchup, or the score once it is played), the spacer and the footer, and in place of the stat line and the three summaries carries one line in red:

[2;31mDNP Injured: Bruised Shoulder[0m

Discord only colours text in an ansi code block, and the colour is an escape sequence (ESC, then [2;31m for red and [0m to stop), which is why it looks like stray brackets anywhere else. On the site an injured game has no stat line, so it counts toward the team's record but adds nothing to his season totals or his credits.

Every embed is the same height

A week of posts should read as a column, not a staircase, so the layout is fixed rather than following the numbers:

• The block is always four lines. Three numbers to a line does it: a quarterback's twelve fill four rows exactly, and everybody else's eleven leave one gap on the last row. It used to be four numbers to a line, which fit on a wide screen, but the headshot takes a column out of the embed and a narrow window left the block about 33 characters — a busy game went past that, Discord wrapped the last cell of each row onto a line of its own, and the grid turned into a jumble. No line is ever wider than 30 characters now: two spaces between columns, one if that would be too wide, and none between a number and its label as a last resort.
• There are always exactly three fields. A fourth wraps onto a second row and makes that embed taller than the one above it, which is why credits earned are no longer among them.
• Headings stay short. A field column is about 95 pixels on a phone. Touchdowns/Interceptions measures 150, so the heading is Touchdowns/INT. The yardage keeps its YDS so there is no guessing which number is which; the very widest summaries (24 CAR, 287YDS) can still wrap in a field column on a narrow phone.

A game with big plays is taller than one without, by however many lines the plays take — that is the point of them.

And the same width

Discord sizes an embed to its widest content, so a quiet game came out narrower than a busy one and the right-hand edge moved from post to post. Two things hold it still:

• Block lines are not padded any more. They used to be padded out to 39 characters, which was one more thing that could wrap on a narrow screen. The spacer image below already holds the width on its own.
• A transparent spacer image, icon/spacer.png. Discord scales an embed image down to the embed's maximum width, so one deliberately wider than any embed pins it to that maximum. It is 1600×2 and entirely transparent, 92 bytes, and under a pixel tall once scaled.

There is no invisible character that does this job. A zero-width space is zero wide by definition, and a braille blank — the usual suggestion — is a character in a proportional font, so a run of them is only ever approximately as wide as the next run. An image is measured.

The kickoff timestamp

A Discord timestamp is an instant, and it renders in whoever is reading's own time zone. A 7:00pm kickoff is not an instant until you know where it was, and a Carlsbad game and a Wake Forest game both listed at 7:00pm are three hours apart — so every school carries a zone in data/players.js.

Working out the instant is two steps: read the naive time as if it were UTC, ask what that instant looks like on the school's clock, and take the gap back off. Daylight saving comes out right because the zone answers for the day in question rather than for today. A game with no kickoff time, or a player with no school yet, simply goes without a timestamp.

A player on a bye, one with no game that week, and one with no schedule at all are left out of the post rather than shown empty.

It reads the site's own data. bot/site-data.js runs data/players.js, data/ratings.js and stats/{year}.js in a sandbox, so the bot and the website are never two copies of the roster — change a school or a score in data/ and both follow.

Running it

node bot/post-week.js --dry-run            # build the next post, print it
node bot/post-week.js --week 4 --dry-run   # build one specific week
node bot/post-week.js --force              # post now, ignoring the clock
node bot/post-week.js --week 4 --league nfl --dry-run   # one league's post

The two webhooks are read from Supabase (saved on the admin page, see How the Discord post gets sent), with DISCORD_WEBHOOK_URL_COLLEGE and DISCORD_WEBHOOK_URL_NFL as a fallback for a league with none saved there. A league with neither is skipped and the run exits with an error; the week stays owed for it. SITE_URL defaults to the Vercel domain and decides where the images and player links point.

The schedule it posts on

.github/workflows/discord-scores.yml runs it four times a day at 07:00, 12:00, 16:00 and 20:00 America/Chicago. GitHub's cron is UTC and ignores daylight saving, so the workflow fires at the UTC equivalents of both CST and CDT, and the script checks the real Chicago hour and exits quietly on the runs belonging to the other offset. The posting times hold all year without being edited twice a season.

The bot is a backstop, and entirely optional. Publishing from the admin page posts the week there and then; these runs catch anything that did not go out — Discord was down, the click half-landed. Each run asks Supabase which weeks are published but unposted, sends them oldest first, and marks them, so nothing is posted twice and nothing is written back to this repository. published_weeks.posted_leagues records which of the two leagues' messages have gone, so a week where the NFL's landed and the college one did not gets only the college one next time; posted_at is set once both are out.

Set it up if you want the safety net; skip it and publishing still posts.

Setup: repository secrets under Settings → Secrets and variables → Actions:

Secret	What it is
SUPABASE_URL	your project URL
SUPABASE_SERVICE_ROLE_KEY	Settings → API. Server-side only — it never goes near the browser
DISCORD_WEBHOOK_URL_COLLEGE	optional: the college channel's webhook, if it is not saved on the admin page
DISCORD_WEBHOOK_URL_NFL	optional: the NFL channel's webhook, the same

Until the first two exist the workflow runs and fails loudly rather than posting anywhere.

---

The Shop

A tab next to Players, at #shop, visible only to a signed-in player. Signed out, the tab isn't there and the page says to sign in. Signed in, the nav carries the player's credit balance beside their headshot, with their diamond balance (see Cards → Diamonds) beside it; on a phone the two stack. Everyone starts on 0 and earns from there. The catalogue lives in data/shop.js.

Items with a picture name it as icon. The originals are in icon/shop at 1254px; the site loads 256px copies from icon/shop/small, about 115KB for all six instead of 1.8MB.

Credits are 60 every offseason, whatever a player is paid, plus what the season earns by contract size and honours (earnings in data/shop.js). The page no longer shows that table, nor any words under the Performance Boosters and Rating Points headings (Sam, 2026-10-20). Anything unspent carries into the next offseason.

Your Inventory heads the page: what this player owns, what is in effect, and the balance. Anything bought repeatedly is one row with a quantity rather than a row per purchase — points read as Catching +5, boosters as 1.5x Booster ×3 — in the inventory, in the admin panel, and in the database. Using a booster takes one off the pile; the row goes when the last one does. Training is the exception and stays one row per purchase, since each carries its own roll. Buying deducts credits and drops the item in. A performance booster is held unused until it is used, and using it deletes it — the inventory is what a player still has, not a receipt book. Workouts and rating points are in the ratings for good once bought and have no switch; Intel and the other non-workout items can still be turned off and on. A stat booster records which attribute it was bought for.

On sale:

• Performance boosters — 2.5x (40), 2.0x (25), 1.5x (15). Regular season only, one use per purchase, and five a season at most: two 2.5x, three 2.0x and four 1.5x. They are stickers: holographic, gold and silver sunbursts you stick on a game.
• Rating points — bought straight into an attribute, priced by how close that attribute already is to 99. Cheap early, dear late. A Sort by drop-down sits above the table: best value (the default — overall gained per credit), most OVR impact, cheapest, most expensive, highest rating, lowest rating, group, and A–Z. Beside it, group buttons (All, Carrying, Passing, Blocking, Receiving…) narrow the table to one group — and only the groups that player's position is judged on get one, so a running back has no Passing. A maxed attribute has no next point, so it sinks to the bottom of the price sorts.
• Offseason training — strength (8) or cardio (8), each with a downside rolled when you buy it, or overall (12) for a smaller gain spread wider with nothing to lose. Cheap to start and twice the price every time you buy it, so an offseason spent on one workout runs out of credits long before it runs out of attributes. See below.
• QB Connection (20, college and later) — with a named quarterback: the card has a drop-down of his team's quarterback room that season (EGE.quarterbacksFor — in the NFL the sim's roster from nfl_sim_rosters, EGE.simQuarterbacks, loaded the first time it is asked for; in college the real roster; plus any of the six, never himself), and Buy waits until one is picked. The row is called QB Connection — C.J. Stroud, with the name in target, and the buy call refuses a name that is not in the room. A quarterback's O-Line Connection has nobody to pick. An admin granting one picks the quarterback too, and how many levels (a held one goes up that many); Set level on a row in Accounts sets it to any level (EGE.wallet.setLevel). Buying it again with the same quarterback levels it up — one row per quarterback, with the level kept in quantity, shown in Roman numerals from II (QB Connection III — C.J. Stroud). The price follows the level (creditsLadder: [20, 20], creditsStep: 10 — 20, 20, 30, 40, 50…), and a different quarterback starts again at the first level. An admin-disabled connection cannot be built up. EGE.priceFor reads any creditsLadder, which is also what makes the chamber's 35 / 45 / 60 / 75… real. In Your Inventory the connection shows the quarterback's headshot (EGE.quarterbackPhoto — one of the six's own, or ESPN's for a roster quarterback) and his name beside the icon. — when the quarterback is traded or retires, the admin switches it off on the admin page and it greys out with an flag on the player's page. The player gets no switch for it, refuses, and a trigger in () refuses the same change straight to the database — to install it. (35, then 45, then 60, then 15 more each time): Injury +5 each, chambers bought before it did that included (, ); the season lock folds it into data/ratings.js and zeroes the rows, which stay to price the next one. (15).

Some things can't be bought yet. An item can name the levels it belongs to, and the shop greys it out everywhere else with the reason on the card — the chamber is NFL only and needs an NFL season behind the player, so in 2023 it reads Not until your first NFL season is behind you for the three rookies and NFL only for the three still in college. Intel is the other way round: the college three can buy it, and an NFL player is told there is nothing left to scout for in the NFL. It is the player's own level that decides (EGE.itemAvailable(item, season, player)), not the season's. The check runs in the buy call too, since a disabled button is only a suggestion.

Intel lasts one season. A purchase records the season it was made in. While it is live, silhouettes appear beside the scouted games on that player's own schedule — hover one and it says SCOUTS IN ATTENDANCE — and the legend counts them. Once the season turns over the row reads Expired, the silhouettes go, and it can't be switched back on. Which games scouts attend is in stats/{year}.js all along as scouts: true; Intel buys the right to see it, and only on your own page.

Each stat booster names the attribute in data/ratings.js it applies to, so buying one has somewhere to land once spending is built. Block Power is the one exception: the ratings carry run block power and pass block power separately, and which one it raises is still open.

Rating points

Buying several at once is every point priced where it lands, added up — not the first one's price times four. A point at 64 costs 1 and a point at 65 costs 2, so a +4 from 64 is 1 + 2 + 2 + 2 = 7, not 4. Pricing the run off the first point would sell the three dearer ones at the cheap rate, which is a discount for buying at exactly the wrong moment.

The shop's main business. A player buys points straight into the attributes their position is judged on, and a point costs more the closer that attribute is to 99:

cost = 100 / (99 - rating + 1)

Rating	50	60	70	80	85	90	95	97	98
Next point	2	3	4	5	7	10	20	34	50

The general attributes are not for sale. Speed, strength, stamina and the rest move only through offseason training, which is what keeps training worth a slot in the shop. A quarterback is left with passing and carrying to buy into; a tight end with catching, blocking and carrying.

Two or three credits while a player is young and unformed, fifty once they are nearly maxed. That is what slows development down by the time they reach the NFL, without ever capping it.

Points on the same attribute stack onto one row, so eighty points across a career leave a handful of rows rather than eighty.

The table shows every attribute the position uses, what it is at, how many points of it are worth one overall, and what the next one costs — ordered by what a credit actually buys, so the top of the table is the right answer and nobody has to work out the weighting themselves.

Three buttons per row buy +1, +2 or +4 at one, two and four times the price of the next single point. Buying in fours is a small discount at the steep end of the curve, where four separate points would each cost more than the last; it is there to save clicking.

Tuning. UPGRADE_BASE in data/economy.js is set so an offseason moves a player about four overall. It was 36 until the overall moved to key attributes; a point in one of those is now worth about twice as much overall, so it went to 100 to keep the same pace. A 90-credit offseason, 20 of it on Overall Training and the rest taken from the top of the table:

Player	Position	Gain
Andrew Parr	TE	+4
Cooper Clark	RB	+3
Paxon Hatch	TE	+4
Isaac Vitel	QB	+4
Sam Stogsdill	RB	+4
Jaykeb Stewart	QB	+4
		+3.83 average

Skipping training and putting all 90 into points comes out at +4.33, so the two ways of spending an offseason are worth roughly the same. Season after season it flattens on its own, which is the point.

Booster stickers

A booster is not used from a list — it is stuck on a game. Signed in and looking at your own schedule, every unplayed game grows a dashed + in the Booster column. Clicking it slides a drawer up from the bottom of the screen holding the stickers you own, each with a count. Click one and it goes on that game and comes out of your inventory.

Hovering a sticker you can still move fades it back, rules slanted lines across it and draws a cross over it, all in the same ink as its edge — it stays exactly where it is and exactly the size it is. Click to take it off and put it back in the drawer.

Once the game has been played the sticker is stuck for good: no cross, and no + on a game that already has a result.

Five a season. No more than five stickers go on one season's games, and no more than two 2.5x, three 2.0x and four 1.5x of them (boosterSeasonLimit and each booster's seasonLimit in data/shop.js). What counts is what is on the season's games — still in Supabase, or written into the season file once the week was out. The drawer says how many of the five are on games and how many of each kind are left, and greys out a kind the season is full of; peeling one off an unplayed game frees its place. applyBooster reloads the stickers and checks before it sticks one on, and a trigger on game_boosters in supabase/schema.sql refuses the row as well, for anything that skips the page. Buying is not capped: an unused booster carries over.

No + on a bye (Sam, 2026-10-20): there is no game to put anything on.

Hurt, no boosters. Once the week a player got hurt in is out, every booster on a game he is going to miss comes off it and goes back in his drawer, and those games take no new booster — the drawer says he is out injured and offers only the wager (Sam, 2026-10-20). Which games he is going to miss is what the injury report says, from the week he went out to the week before he is back (EGE.outInjured in data/games.js), so an injury further down the file is not given away early. The handing back is EGE.wallet.returnInjuredBoosters, run by the player's own page whenever it pays what he is owed and by the admin's publishing a week; a booster only goes back to the drawer from whichever of them actually deleted its row, so it can never come back twice. Boosters already written into the season file are left alone.

Wagers

The drawer has a Wager beside the boosters (Sam, 2026-10-20): a pick against the point spread on the game. Every game the six have left gets a line — Seattle by 5.5 — and a wager takes one side of it: Seattle −5.5 needs Seattle to win by 6 or more, Pittsburgh +5.5 needs Pittsburgh to win or lose by 5 or fewer. Every line ends in a half, so there are no pushes, and the closest a game gets is a half point. Beat the line and the game pays double its credits; miss it and it pays no credits. Diamonds and the Player Card bonuses are paid as if there were no wager. Two a season, playoff games included, never on a game already out, and free: a wager is a pick, not something bought, so nothing comes out of the inventory. It takes the game's one sticker slot, so a game carries a booster or a wager, never both. A game a player is out hurt for still takes a wager (it pays the 1 credit a missed game pays, so 2 or 0).

The line is worked out in the browser by js/lines.js from a projected score for each side (shown in the drawer: Projected 18–13), and the score from what the site has already put out and nothing else:

• every published result of the season before the game — the six's own, every conference race in data/conferences.js, and the NFL sim's other games in data/nfl-results.js — as points scored and allowed, with who each team played taken into account, and a blowout past four touchdowns counting half of the rest;
• where each team stood beforehand, which those results take over from as the weeks go by: in the NFL its starters' ratings off the Teams pages' rosters (never the sim's own team ratings), in college its defense's yards allowed in data/defenses.js and whether it is FCS; and last season's results where the site has them;
• home field, and nobody at a neutral site;
• one of the six out hurt, once his injury report is up (a quarterback costs his team most).

So a line moves as the weeks come out, like a real one. A wager keeps the line it was taken at: the line column on its game_boosters row, and line beside booster in the season file once the week is pulled in. If the line moves between the drawer opening and the pick, nothing goes on and the drawer shows the new number.

On the schedule it is a black poker chip with the mark of the team it is on (and the drawer's blank one says BET). Once the game is out the chip turns green for a win or red for a loss, with nothing stamped on it (Sam, 2026-10-20); one that came out level stays black. The Paid column shows the doubled credits or a red +0, and the game log carries Bet 2x or Bet 0x beside the opponent. Until then it comes off the way a booster does.

It is stored like a booster: a game_boosters row with item_key wager-for (your team) or wager-against (the opponent) and its line, written into the season file's booster and line fields when the week is pulled in, and the season editor's Booster or wager pick has both (picked there by hand, it takes the line as it stands). EGE.wagerOn in data/games.js reads it and settles it from the score and the line; EGE.economy.awardsEarned and EGE.creditsFromGame pay it. The credit ledger only ever tops an award up, so the stickers are always loaded before a player's credits are paid (syncAwards), or a lost wager would be paid in full and never taken back. The trigger on game_boosters refuses a third wager in a season, a wager without a half-point line, and any sticker on a week already out, admins aside.

No booster on a playoff game. A booster is what you spend a limited drawer on across a regular season you can see all of; the postseason is not planned, it is whatever the bracket hands you. A game with playoff: true still grows a +, for the wager (Sam, 2026-10-20), but the drawer offers no boosters there, and applyBooster in js/wallet.js refuses the week even if it is called directly. A playoff game pays more on its own instead: 1.25 times the credits and twice the diamonds.

The drawer sits along the bottom of the screen in one row where there is room — the game and what the season has left, the boosters, the wager — with a cross in the corner. A click anywhere off it, or going to another page, shuts it (Sam, 2026-10-20).

Stickers are private. Only the player who stuck one on — and Sam, as admin — can see what is riding on which game. That is in the policy on game_boosters, not just the interface, because the anon key can query the table directly.

Each sticker is two die-cut layers, a backing in the same ink as the printed number with the foil face inset inside it, both clipped to the same sunburst. A plain border would be cut away by the clip-path, which is what made the first pass look chewed at the edges; each face also paints a solid ground under its moving gradients so nothing can flash through bare.

They sit bigger than their row and hang over the edge of it, top and bottom, centred on the same point the empty slots use so the column stays a column. Angle and vertical nudge both come from a hash of the sticker's own row id, so every sticker lands differently and keeps that placement through every redraw — scattered, but never jittering. Nothing in the slot is in the row's flow, so a row carrying a sticker measures exactly the same as one without: 51px either way.

The three designs scale off a single --sticker-size, so the same sunburst renders at 62px on a schedule row and 76px in the drawer without a second copy of the CSS.

Training

Training rolls one twelve-sided die when you buy it. On a 1, 2 or 3 the downside lands, all of it; on 4 and up, none of it — a quarter of the time, not most of it. Rolling each risk separately is what made three coin flips add up to a debuff nearly every time (87.5%, as it turned out).

Item	Effect	Risk	From
Strength Training	Strength +4, and toughness, injury and jumping +2 each	agility, stamina, speed −2 each	8
Cardio Training	Speed, acceleration, agility, stamina +2	strength −2	16
Overall Training	Speed, acceleration, strength, agility, jumping, stamina +1	none	20

Risks are rolled once, when the item is bought, and the result is stored on the row. Reloading the page never re-rolls it.

The price doubles every time. A workout carries creditsStack: 2 in data/shop.js, and EGE.priceFor(item, owned) multiplies its base price by that for each one already on the books — strength runs 8, 16, 32, 64, 128. A 60-credit offseason buys three of them — 8 + 16 + 32 = 56 — and the fourth is 64 on its own. That is the point: the cheap first block makes training worth doing, and the doubling makes spending a whole offseason on one attribute a choice rather than the obvious move.

The two that specialise both carry a risk. Overall Training spreads the same idea across six attributes, has nothing to lose, and costs the most for it.

Prices follow what a workout does to the overall. Since the overall moved to key attributes, speed, acceleration and agility count for a lot more and strength, toughness and injury for less. Cardio went from 8 to 16 and Overall Training from 12 to 20, which keeps what a credit buys in overall about where it was: on average across the six players, one Cardio block is worth about 0.45 overall (0.22 before) and one Overall Training block about 0.29 (0.17 before). Strength Training stays at 8. It is worth less overall than it was (about 0.12), but what it buys in toughness and injury matters outside the number.

owned is counted from the inventory rows, so the reset is the one that already exists: Clear the shop rows at the end of a season deletes every train-% row once the workouts have been folded into data/ratings.js, and the next offseason starts at the bottom of the ladder again.

The price is worked out in js/wallet.js from what the server holds rather than from what the page says, the same as every other price here — the page is showing it, not deciding it.

Everything carries over

Unused performance boosters and unspent credits carry into the next season — nobody loses what they paid for. Rating points and offseason training do not carry as rows, because at the end of a season they are folded into data/ratings.js and become the player's actual numbers. Intel lapses at the end of the season it was bought for.

A performance booster is used on the player page, by putting it on a game from your own schedule. There is no Use button in the shop: spending one there never said which game it was for.

A quarterback's QB Connection reads O-Line Connection — he spends the offseason with his offensive line, and better chemistry means a lower chance of being sacked. Either way it cannot be bought until college.

What a player owns sits on one line under Applied to your ratings: the boosters as their stickers, then everything else as its icon from icon/shop, each with how many are held boxed on its corner (3X). A workout bought more than once is one icon with its count. Intel is its own switch — tap it to turn it on or off, and it greys out with an OFF tag while off.

Offseason training sits in the Rating Points panel, under the points table, rather than in a panel of its own.

Sam is the admin, and his tools are on their own page — see The Admin Portal. Nothing admin sits on the shop page: the shop is what a player buys with, and that is all it is. Admins are rows in the admins table, so the list is changed in SQL rather than from the browser.

Balances and inventories live in Supabase — player_credits and player_inventory in supabase/schema.sql — and the policies there do the real enforcing: a player writes only their own rows, an admin writes everyone's. The browser code is the shape of the UI, not the security boundary.

Purchases are readable by everyone, because a boosted overall has to show on a player's card wherever it appears. Intel is the exception, in the policy as well as the interface: only its owner and an admin can read an Intel row, so nobody else can work out which games scouts will be at.

Buying reads the balance, checks it and writes it back in sequence rather than in one locked transaction. With six players and one shop the worst case is a double spend from two tabs at once, which the admin can put right.

Credits earn themselves

Nobody hands credits out by hand for a good game. Every game pays for its fantasy points the moment the week is published, so a big night without a touchdown still pays. The scoring is standard half PPR:

Passing	1 per 25 yards, 4 per touchdown, −2 per interception
Rushing	1 per 10 yards, 6 per touchdown
Receiving	1 per 10 yards, 6 per touchdown, 0.5 per catch
Fumble lost	−2

Each position keeps a share of its points, because they do not score alike — a strong tight end game is about 22 points, a strong back or quarterback game about 31:

Position	Credits per fantasy point	A breakout game pays
QB	0.55	15
RB	0.50	15
WR	0.75	15
TE	0.75 (1.00 from 2026, 5 a game at least)	15 (20 from 2026)

The shares are worked out from the 2018 season's 72 stat lines, not picked. They make a breakout game (the average of each position's 75th and 90th percentile) pay the same 15 credits for every position. One scale over all three then makes a whole season pay what touchdowns used to (757 credits across the six against 745), so no shop price needed to change. Re-scored this way, 2018 would have paid 90–173 a player instead of 70–165, and no game would have paid nothing.

Credits are always rounded up — there is no half a credit — and a game played never pays less than 1.

A tight end keeps all of his points from 2026 (Sam, 2026-10-27): 1 credit a fantasy point instead of 0.75, and never less than 5 for a game he plays (EGE.economy.TE_SHARE, TE_MIN_CREDITS). He spreads his credits over more ratings than anybody, so this smooths his climb. The diamonds come off the same credits, so a game pays him at least 5 diamonds too (more with a booster, 10 in the playoffs), and his Player Card bonuses grow with them. A game he sits out hurt pays 1 like anybody's. On 2025's lines Andrew would have made 160 credits instead of 116. It starts with 2026 (TE_FROM) because the ledger would top up every earlier game the next time a week in its season was published. Card grades are not touched: a tight end still grades on 0.68.

A playoff game pays 1.25 times the credits — any game with playoff: true, the conference title games and bowls included. The fantasy points times the share times 1.25 are rounded up once, so a line worth exactly 15 credits in the regular season pays 19 in the playoffs. That starts with 2022 (EGE.economy.PLAYOFF_FROM): the seasons before it were paid as they went, and the ledger would top every one of their playoff games up the next time a week was published. A playoff game also pays twice the diamonds (see Diamonds).

2018 stays on touchdowns. It was paid as it went, 10 a touchdown for a back or tight end and 5 for a quarterback, and the ledger tops up any award that has grown, so re-scoring it would pay every player a second time for a season already settled. Fantasy scoring starts with 2019 (EGE.economy.FANTASY_FROM). The award key is still td-w{week} for both systems, because that is the key the database already accepts from a player's own browser, so the change needed nothing in Supabase.

On top of that, every season after the first pays the flat 60 offseason allowance on the way into it. The first season pays nothing: everybody starts on zero and earns the first 60 by getting through a season.

How it stays right. What a player is owed is worked out from the schedule every time they open the site, and each award carries a key for what earned it — td-w4, offseason-2019. Those keys are unique per player per season in credit_awards, and pay_credit_awards() inserts the new ones and moves the balance in one transaction, so paying on every page load is both safe and the whole point. Post a result and the player is that game's credits better off the next time he looks. Reload all day and he is still only that much better off.

A player with no sign-in yet still scores: the admin page shows what they are owed as waiting, and it lands in full the first time they log in.

Correcting a week. Change a stat line in stats/{year}.js, commit, and the credits follow: an award already on the ledger that is now worth more pays the difference and nothing more, so the same week can be settled as often as you like. It reaches a player the next time they open the site, and Pay out on the admin page pushes it to everyone at once without republishing anything — which is what that button is for once every week is already out.

The one thing it will not do is take credits back. An award that is now worth less than what was paid stays paid, because the player may well have spent it; lowering a published number means adjusting that balance by hand on the admin page. Weeks that have not been published yet have nothing on the ledger, so changing those is free.

The credits come from the browser, because what a touchdown is worth is worked out from stats/{year}.js and Postgres has no copy of it. A player could already set their own balance directly — buying things needs that — so this is not a new hole, but the easy half of it is closed: for anyone who is not an admin, an award has to look like one of the two kinds the site issues and cannot be worth more than either honestly could be.

---

Cards

A tab of its own, at #cards, beside the shop and visible only to a signed-in player. It is a collectible card game that sits next to the simulation rather than in it: nothing on it changes a rating, a stat line, a result or what a game pays. A player who never opens a pack is not behind anybody at anything. The offseason has the shop; this is something to do while the weeks are going out.

The catalogue, the grading, the packs and the library are in data/cards.js; the Supabase side is js/cards.js; the tab is js/cards-view.js.

Diamonds

The card game has its own money. Packs cost diamonds, never credits, so a pack is never bought with something that could have gone on a player — which is what kept anybody from buying them. Library sets pay diamonds too, and so does quick selling a card. Nothing in the simulation takes or gives them.

Every game pays them, on top of its credits and off the same numbers: the game's fantasy points times the position's share, rounded up. Then the booster on the game multiplies them — a 1.5x Booster pays 1.5 times the diamonds, a 2.0x twice, a 2.5x two and a half times, rounded up again. A 15-credit game with a 2.5x Booster on it is 15 credits and 38 diamonds. The multiplier is the booster's own multiplier in data/shop.js, and the booster is read through EGE.boosterOn, so a week still in Supabase and a week already written into the season file pay the same.

A playoff game pays twice the diamonds (PLAYOFF_DIAMONDS in data/economy.js): the game's plain credits times two, not its playoff 1.25x credits doubled again. No booster can go on a playoff game, so the two never stack, and that 15-credit line is 19 credits and 30 diamonds in the playoffs.

Diamonds start with 2022 (EGE.economy.DIAMONDS_FROM). The seasons before it were paid in credits as they went, and four seasons of diamonds at once would fill every binder on the first day. A full 2022 pays about 145–225 diamonds a player before boosters.

Credits buy diamonds, two for every credit, from the Buy Diamonds box under the packs (DIAMONDS_PER_CREDIT in data/economy.js, charged by buy_diamonds() in supabase/schema.sql — change both together). At that rate a pack costs in credits exactly what it did before diamonds existed. Nothing turns diamonds back into credits — which is also why the library pays diamonds rather than credits now.

They are paid the way credits are: diamond_awards is a ledger keyed game-w{week}, pay_diamond_awards() pays what is new and tops up what has grown, in one transaction, on every sign-in and every publish. For anybody but an admin an award has to carry that key and be no more than 300 (the 120 a game's credits are held to, times 2.5). Each game's diamonds show beside its credits on the player's own schedule.

Unlike credits, a player cannot write their own diamond balance: player_diamonds has an insert and update policy for an admin only, and a player's diamonds move only through paying a game, buying diamonds, opening a pack, claiming a set, quick selling a card and making a holo.

Making a holo pays a tenth of the holo's fantasy score in diamonds, rounded up and never less than 1 (Sam, 2026-10-20): the holo's score is three times the card's, so a 170-point holo pays 17, most pay 2 to 4 and Sam's 65.3-point Iconic (195.9 as a holo) pays 20. There is no cap. The page works it out (holoBonus in data/cards.js) and merge_holo_card() pays it in the same transaction as the merge, turns down only more than 60 (a page that has been tampered with; HOLO_BONUS_MAX, change both together) and keeps it in card_merges.diamonds. Holos made before PR #88 paid nothing, and those made between it and this paid a tenth of the one card's score, capped at 15; none are back-paid.

The diamond balance in the nav opens the diamonds page (#diamonds, js/diamonds-view.js). Recent Games is a box a week, newest first, six at a time with Show more: every game that week that paid the account anything, one row a game, split into what it paid as their own game (game-w{week}) and what it paid through their Player Card on that player (pc-{slug}-w{week}, with the rank it was paid at), and the week's total. A row opens that game in the player's game log. The panel's head has the balance and the live season's split.

Other Earnings, under it, is everything else that paid, newest first, ten at a time (Sam, 2026-10-20): holos made (card_merges.diamonds), Library and Gold Rush sets completed, the Gold Rush's final reward included (the diamonds in each card_set_claims / card_event_claims reward), and cards sold on the market (each card_listings sale's price, with who bought it). Its head splits them into Holos, Sets and Market, and it only shows once one of them has paid. Quick sales and the Daily Spin are not on it.

The page reads what is already kept, so it shows what was paid, not a sum done again, and needed nothing new in the database beyond card_merges.diamonds.

Where the cards come from

The season files. Every game with a stat line is a performance card, and every line in its bigPlays is a play card, so publishing a week puts that week's cards into the packs and nothing is drawn up by hand. A week that is not published is in no pack. A game a player was hurt for, or one where his line is nothing but zeros (Isaac's redshirt year), makes no card.

A card's id names the game it is, and that is all the database stores:

p:2021:sam-stogsdill:5        Sam's week 5 game in 2021
h:2021:sam-stogsdill:5:0      the first of that game's big plays

Correct a stat line and the card corrects with it. Reordering a game's bigPlays swaps its play cards round, so add new ones at the end.

Grading

The number in a card's top right is its fantasy score, the same half-PPR scoring a game is paid on. A play is scored as the one play: a 44 yard receiving touchdown is 4.4 + 6 + 0.5 = 10.9.

The rarity comes from that score, graded against the position:

• A performance is graded on its fantasy points times the position's credit share (the number that makes a breakout game pay 15 credits whatever the position), so a tight end's big night is as rare as a quarterback's.
• A play is graded as though every play were carried or caught (yards / 10, +6 a touchdown, +0.5 a catch), so a 60 yard touchdown pass grades with a 60 yard touchdown catch. Go-ahead and game-tying plays grade 1.5 higher, game-winning 3.

The thresholds were set from the 2018–2021 files so the catalogue is a pyramid, and each rarity has its own pull odds:

Rarity	Colour	Chance per card	Game from	Play from	In the files
Common	gray	47.4%	—	—	335
Uncommon	green	26%	6.5	7.1	215
Rare	orange	14%	11	8.6	143
Epic	red	8%	16	9.6	98
Legendary	purple	4%	20.5	11.2	75
Mystic	gold	0.5%	28.5 (24 in college and the NFL)	13.3	10
Iconic	pink	0.1%	picked	picked	4

Counts are the 2018–2022 files. Mystic was 24 / 12.1 until Sam asked for it to be closer to 10 cards than 35 (2026-10-13): the 30 that fell out became Legendary, and Legendary moved up from 20 / 11.1, so its six lowest became Epic. Then a college or NFL game got a lower Mystic bar than a high school one, 24 for a performance (collegePerformance, Sam, 2026-10-17), which made three more college performances Mystic: Paxon's 199 yards against Missouri State (2021) and Andrew's three touchdowns against UL Monroe and 172 yards against Austin Peay (2022). Plays keep the one bar.

A tight end grades on a share of 0.68 rather than the 0.75 his credits pay on (GRADE_SHARE in data/cards.js): on 0.75 Andrew had six Mystic performances to everybody else's three at most (Sam, 2026-10-17). He keeps two, his 205 yards and three touchdowns in 2019 and his 215 yards at Tennessee in 2022.

Iconic is picked by hand, not graded. Grading stops at Mystic, and only the moments in ICONIC at the top of data/cards.js are Iconic: Sam's five touchdowns against Centennial (2019), Paxon's two touchdowns in the FCS Championship (2020), Cooper's 80 yard touchdown against Helix (2019) and Jaykeb's seven touchdowns (2022). Add an id there to make another. At 0.1% a card, about one Base Pack in 330 has an Iconic in it.

From Rare up the frame catches the light: a still sheen on Rare, a moving one on Epic and Legendary, gold foil on Mystic and holographic on Iconic, the same foils as the booster stickers. The moving shine (the sheen, the flowing gold and holo) only plays in a pack being opened and on a card up close; everywhere else cards hold still in their frame colours so a full binder does not lag (Sam, 2026-10-13).

Across the bottom of the picture, a play card prints its yardage big (44 YD TD) and a performance card its yards and touchdowns (312 YDS 5 TD). A touchdown its season file calls game-winning (or walk-off) or game-icing, in a game the team won, says so in the rarity's colour on a line under YD TD, the two lines beside the number (Sam, 2026-10-20). The words in the file decide, since nothing else there says when in a game a play came; the NFL sim writes game-winning for a fourth-quarter or overtime touchdown that put its team ahead for good.

The two card-art cards are drawn by the page (Sam, 2026-10-20): Sam's week 18 and Cooper's week 11 in 2019, Mystic, only in the Gold Rush Pack (CARD_ART in data/cards.js). Each is the usual card with Sam's photo of the game in the picture in place of the headshot and no team logo, and the helmet cut out above the photo stands up out of the picture over the PERFORMANCE strip and the score, number and all, under the frame and the big yards and touchdowns. So they have the foils, the tilt and the close-up of any card, and three of one merge into its holo, whose rainbow covers the helmet as well as the picture. Only the GOLD RUSH card is still Sam's picture printed whole, and it has no holo.

Under the picture a college or NFL card names its opponent with the opponent's mark after it, about a capital letter high (vs. Ohio State and the Block O; Sam, 2026-10-20). A high school has no mark.

Packs

Pack	Diamonds	Cards
Base	10	3	any season
Season	12	3	the live season only, the last card Uncommon or better
Pro	20	5	any season, the last card Rare or better, 1.5x odds on Legendary and up
Season, a past year	14	3	that one past season only, the last card Uncommon or better

Base, Season and Pro are twice what they were in credits, because a credit buys two diamonds.

Each pack's tile is centred, top to bottom: the pack, its name, what is in it, and Buy & Rip. The Season Pack has an arrow either side of its Buy button: the left one steps it back to 2021, 2020 and so on, each a year-{season} pack at 14 diamonds (YEAR_PACK_PRICE in data/cards.js), and the right one steps forward to the live season. A past year is a darker bronze with white words. A new year joins the arrows when the site rolls to a new season. The arrows only redraw the Season Pack, and every pack's shimmer keeps time with the clock (--phase), so redrawing the packs never starts their shimmer over. The Pro Pack glows a little purple.

The packs look like a Panini Prizm pack: a foil pouch whose crimped seals, top and bottom, are photos of a real pack's (icon/packs/crimp-*.webp, Sam, 2026-10-14) washed in the pack's colour, with a dark EGE Football bar across the top, the prizm (a burst of fine rays and a band of colour that slides across them) behind, the pack's name in tilted chrome, and a foot band with the count and the year. A season can have a cover: a player cut out on the front, his team's logo large behind him, and a colourway of its own (PACK_COVERS in data/cards.js, pictures in icon/packs/, colourways as .ege-pack--{theme} in site.css). The player's helmet just touches the bottom of the EGE Football lettering and his legs run behind the name, which sits a little above the trim. 2022 has Paxon at North Carolina in the blues of his jersey in the photo (cooler than the official Carolina blue), the NC behind him and an argyle strip (Sam, 2026-10-14). 2021 has Jaykeb at Ohio State in the scarlet of his jersey, the block O behind him and the helmet stripe across the foot, and 2020 Andrew at Alabama in the deep crimson of the script A, a darker A behind him and Greek laurel leaves across the foot, and Andrew 20% bigger than the others (a cover's zoom); 2019 has Sam at Normal Community, 20% bigger and a touch lower (a cover's drop, so his helmet clears the EGE letters), on near-black foil with darkened seals and a dimmer prizm, lined in the Ironmen's orange, the N behind him as only its orange outline, and a black sword on an orange band across the foot; and 2018 has Cooper at Carlsbad on purple foil, the Lancers' C behind him as only its outline in white, and a royal stripe of black, white and purple across the foot (Sam, 2026-10-14). 2023 has Isaac at Texas, 20% bigger and lowered a touch, on burnt orange foil off his jersey, the longhorn behind him in a dark burnt orange edged in white with its horns out either side of his shoulders, and the sleeve's two white stripes across the foot (Sam, 2026-10-17). Every season now has a cover, every player 20% bigger than the first cut (zoom: 1.2; Cooper, in a tall standing pose, 1.44), some lowered a touch (drop) so no helmet covers the EGE Football letters. A cover follows its season: on the live Season Pack while it is that season, then on that year's pack behind the arrows. A season with no cover gets the plain pack, its name in the middle.

Each card rolls its rarity on the odds first, then is drawn from every published card of that rarity the pack can hold. A rarity with nothing out yet falls to the nearest one below it. The Season Pack is the one that grows a week at a time.

A pack holds at most one card above Legendary (PACK_CAP in data/cards.js): once a Mystic or Iconic is in it, every card after it rolls Legendary or lower. A Pro Pack has 1.5x the odds of Legendary and up (boost on the pack): Legendary 6%, Mystic 0.75%, Iconic 0.15% a card, with Common at 45.1% making up the difference. The cards are laid out worst to best, left to right (packOrder), in the opening and in Recent Pulls, so the best card is always the last one turned.

Buy & Rip brings up the sealed pack. Nothing is paid until it is torn: the tap that tears it is what calls open_card_pack(), and closing the overlay with the pack still sealed costs nothing and adds nothing. The sealed pack gives nothing away either, with no words under it and no glow of what is inside. Once torn, the cards are dealt face down and each one turns over on a tap with a flash of its colour from Rare up. Cards a player has never had before say New.

Sounds: the pack makes a ripping sound as it tears, and every card makes a flip sound as it turns over (a set's reward card and a holo too; a big pull flips when it turns, after it holds back). Both are Sam's clips in sound/, trimmed at the front so they start on the tap, and both play quietly because the clips are harsh at full volume: the volumes are SOUNDS in js/sounds.js (flip 0.25, rip 0.1; the rip clip is far louder). The big pulls have sounds of their own (below), a little louder than these. They go through Web Audio, so an iPhone keeps to those volumes, stays quiet on silent and plays over the player's own music. If the browser won't play sound, nothing else changes.

Quick sell out of the pack: once every card is turned, tapping a card picks it (tap again to put it back), and Quick sell N sells all the picked ones at the quick-sell prices before the next pack. Iconic cards can't be picked.

Nothing from the pack shows in the collection, the library, the recent pulls or the counts until it is put away with Done, so nothing behind it gives the pack away; a set's reward card works the same way.

The big pulls have their own openings, longer the rarer they are. A Legendary or better holds back before it turns over: it lifts, shakes and glows in its colour, and turns in a burst of rays, a ring and sparks.

Rarity	Holds back	Then
Legendary	0.9 s	purple rays, a ring and sparks, 1.6 s
Mystic	1.7 s	the overlay washes gold, a flash, more sparks, 2.4 s
Iconic	2.8 s	the other cards dim, the screen goes pink, a flash, two rings, a shower of sparks and ICONIC across the top, 3.6 s

The timings are BUILDUP and SHOW in js/cards-view.js. With reduced motion turned on in the phone or computer's settings, none of it runs and the card just turns over.

They sound bigger too. While the card holds back, a build rises under it, a whoosh sweeping up that ends exactly as the card turns. As it turns comes the reveal: a boom, a punch, a whoosh falling away and a cymbal. Nothing musical -- no notes and no sparkles (Sam's call). Each rarity is longer and deeper than the one below:

Rarity	Build	Reveal
Legendary	0.9 s	a light boom and cymbal, 1.8 s
Mystic	1.7 s	a bigger boom and a longer cymbal, 2.6 s
Iconic	2.8 s, with a heartbeat that quickens	the deepest boom and the longest cymbal, 4.2 s

They are sound/{rarity}-build.mp3 and sound/{rarity}-reveal.mp3, all synthesised for the site. A build is exactly BUILDUP long, so if a BUILDUP changes, the build has to be remade to match. With reduced motion there is no build, but the reveal still plays. Closing the pack fades out whatever is still playing.

The price and size are charged by open_card_pack() in supabase/schema.sql, not by the page. Change both together. Which cards are in a pack the page decides, because the season files are not in Postgres. That is the same trust the credit awards run on, with the easy half closed the same way: a pack has to hold exactly its size of real-looking ids, and it costs what it costs. It is paid from the diamond balance, taken under lock, and card_packs records it in diamonds; its credits column is what a pack opened before diamonds cost.

The collection

Every card owned, rarest first by default, with how many of each (×2). Filter by rarity, by kind and by player; sort by rarity, score, newest pull, year (the newest season first, each under its year) or game. Five cards go across (four, three and two on smaller screens), three rows show at a time and the rest scroll inside the panel, so the page below is never far away.

Only the first few rows are drawn. More are added as the binder is scrolled toward its end, a few rows at a time, so a big collection is not hundreds of foil cards for a phone to run at once. Selling a card, ripping a pack or a refresh keeps the binder where it was scrolled to.

Tap a card to hold it up close with the game it came from; the → button beside its stat line opens that player's page on the card's season and picks the game out in the game log. Up close the card is as wide as the facts under it and leans toward the pointer, every rarity the way a holo always did. The facts leave out what the card already shows: no rarity, no fantasy points and no count of your own; a Gold Rush card's Event is just Gold Rush (Sam, 2026-10-20). Duplicates are kept, and are what trades are made of.

Anybody's collection can be looked through. The first drop-down switches from My collection to any other player's, with the same filters and sorts. Their cards up close list everybody who has them, you and them included, under Who has it. It is the same reading a trade already does: any signed-in player can see anybody's cards. A card of theirs up close also has Trade for it, which goes to the Trades page with the trade builder open to them and that card already asked for; you pick what you give and send it as usual. A card on somebody else's showcase or in their recent pulls has the same button, signed in. If they no longer have it, the builder says so and asks for nothing.

Who else has it. Signed in, any card up close (from the collection, a pack's recent pulls, a trade, the shop, the market or a showcase) lists who else holds it, with each account's face, first name and count, biggest pile first, plus any copies up on the Market. On a card from your own collection you are left out (your binder has your count); anywhere else it is Who has it, with you as You and the player whose collection or showcase it is counted like anybody else (Sam, 2026-10-20). It reads player_cards for that card id, which any signed-in player can already read (ownersOf() in js/cards.js), so it needed no schema change.

Quick sell

A card of your own up close has Quick sell, which sells one copy back for diamonds. The first tap turns the button red and says the price; the second sells. With copies left, the overlay stays open and counts down; the last copy is stamped Sold. It comes off the showcase if it was up, and an open trade it was in can no longer be accepted (it expires, the same as a card traded away).

Rarity	Sells for
Common, Uncommon	1 diamond
Rare, Epic	2 diamonds
Legendary	3 diamonds
Mystic	30 diamonds
Iconic	cannot be sold

Quick sell several at once. A price-tag button beside the bin at the end of the collection's filters (beside the Sort button on a phone) turns the binder into a picker: a tap picks a card to sell or puts it back, and a card held more than once gets − and + for how many copies go. A bar over the page buttons keeps the count and what it all pays, Extras picks every copy past the first of the cards showing (as filtered), and Quick sell sells them on a second tap, one copy at a time through quick_sell_card(), so a refusal stops there and leaves the rest picked. Iconic cards grey out and cannot be picked. Cancel (or Done) leaves it; so does going to another page or another player's collection. Nothing in the background redraws the binder while it is on.

The prices are QUICK_SELL in data/cards.js and are paid by quick_sell_card() in supabase/schema.sql. Change both together. The page says what rarity the card is, since the grading is done from the season files, the same trust a pack's contents run on; the database keeps the prices, refuses Iconic, takes the copy under lock so it cannot be sold twice, and writes every sale to card_sales. A card only ever sells for diamonds.

Recent pulls

Under the packs: the last packs anybody opened, newest first, each with who opened it. Four at a time, and Show more goes back three times, so sixteen at most. Tap a card to see it up close; on your own pulls you can sell it from there, on somebody else's there is Trade for it, and a card of yours that has since been sold or traded away is shown faded. card_packs stays readable by its owner and an admin only, so these come through recent_card_opens(), which hands over who, which pack, the cards and when, never the price.

The library

Sets to put together, each claimed once. A set is a list of slots, each wanting one card. No card fills two slots of the same set, but a card counts toward every set it fits, and claiming uses nothing up.

Gold Rush cards count toward them too (Sam, 2026-10-20), each as the card it is a Gold Rush copy of: a card and its Gold Rush copy are one card here and never fill two slots, the two card-art cards count as the games they are art of (at their own Mystic), and the GOLD RUSH card counts as an Iconic. The Gold Rush's own sets still take Gold Rush cards and nothing else. Halloween cards count the same way: a Halloween card, its costume card and the plain card are one card here.

Set	Needs	Reward
The EGE	a card of each of the six	Rare+ card · 6 diamonds
Bloomington Connection	4 Isaac + 4 Paxon, 2018–19	Rare+ card · 6 diamonds
Buckeye Backfield	4 Sam + 4 Jaykeb, 2020 on	Rare+ card · 6 diamonds
Highlight Reel	15 different play cards	Epic+ play card · 10 diamonds
Collector	one of every rarity	Legendary+ card · 20 diamonds · 1.5x Booster
Hall of Fame	3 Legendary-or-better	Mystic+ card · 20 diamonds · 1.5x Booster
Player (×6)	8 different cards of him	Rare+ card of him · 6 diamonds
Year Season (each season)	a performance card of everyone who played	Epic+ card from that year · 10 diamonds · 1.5x Booster

The rewards are kept small on purpose. Measured by simulation with Base Packs, a player gets back roughly 30–50% of what they spend on packs (counting a booster at its 15-credit shop price, 30 diamonds at the swap rate), and the early sets come first. Cards are a place to spend diamonds for fun, never a way to make credits. Rough cost of each set in Base Packs: The EGE about 50 diamonds, a player or Highlight Reel 100–230, Hall of Fame about 150, a past season about 400, Collector about 4,800 (it needs one of the Iconics, so it is mostly finished by trading for one). The live season's set is about 130 through Season Packs. The rewards and the costs both doubled with the packs, so these are the same proportions as before.

A reward booster is an ordinary booster. It lands on the same stacked inventory row a bought one does, and goes on games under the same five-a-season trigger, so a full library never means more stickers on a season than anybody else can have.

claim_card_set() in supabase/schema.sql pays the reward. It checks that every card the set was completed with is really in the collection, that the set has not been claimed before, and, for anybody but an admin, holds the reward to 40 diamonds and the 1.5x. It was p_credits before diamonds, and Postgres will not rename an argument, so the schema drops the old function and creates it again.

Halloween

A limited-time event like the Gold Rush (Sam, 2026-10-26): an admin starts it from the Cards tab of the admin page and it runs seven days. The Halloween Pack is 4 cards for 18 diamonds, the last one Rare or better, from October games, the Graveyard and U.F.O. kinds of game below, and two Mystic card arts (Paxon's 2025 week 16 performance, Isaac's 2022 week 12 79-yard touchdown). A Mystic from the pack is one of the card art 55% of the time (artShare; Sam wanted the art a little likelier than any other Mystic), about one pack in 24. A Halloween card (:hw) wears Sam's pumpkin on a dark purple gradient sticker with the sunburst's dark edge, no shine and no shadow, and its back says Trick or Treat. The card art wears no sticker, so nothing covers the art.

Costume cards (:hc): a Legendary-or-better pull from a Halloween Pack has a 10% chance of coming out with the player's costume headshot (headshot/halloween/; Sam picked 10% over 0.5%, which made Trick or Treat out of reach). A costume card is its own card: it never stacks or merges with the Halloween copy of the same game. Only Halloween Packs give them.

Only Halloween cards fill the event's sets, and a copy that is traded can't be traded again for 48 hours. Card Art counts only art you pulled yourself (checked against card_packs, on the page and in claim_event_set()).

Set	Needs	Reward
Graveyard	Isaac pick-six; game with a fumble; 6+ sacks; 10 or fewer total yards; 3+ interceptions	Legendary+ · 20 diamonds · 2.0x Booster
Card Art	a Halloween card art you pulled	Epic+ · 10 diamonds
Trick or Treat	a costume card of each of the six	Mystic+ · 40 diamonds · 2.5x Booster
U.F.O.	50+ yd NFL passing or receiving TD; 70+ yd college one; 200+ receiving yard game; 400+ passing yard game	Legendary+ · 25 diamonds · 2.0x Booster
Holoween	an Epic-or-better holo of each of the six	Legendary+ · 30 diamonds · 2.0x Booster
October	15 different Epic-or-better cards from October games	Legendary+ · 20 diamonds · 1.5x Booster
Witching Hour	a card of each of the six from October 25–31	Epic+ · 12 diamonds · 1.5x Booster
Monster Mash	a Legendary-or-better card of each of the six	Legendary+ · 15 diamonds · 1.5x Booster
Halloween	all eight sets	the Iconic HALLOWEEN card · 100 diamonds · two 2.5x Boosters

Every reward card is a Halloween card. The HALLOWEEN card is Sam's art of Paxon and Cooper in the 2023 Holiday Bowl (icon/cards/halloween-iconic.webp, week 19, 2023), with no sticker. Its two boosters are "boosters": 2 in card_events.rewards, which claim_event_reward() gives one at a time.

Trades

The Trades panel on the Cards tab. Propose a trade opens a builder: pick who with, tap up to six of your cards to give and up to six of theirs you want, and send it. Asking for nothing makes it a gift. Each side has a Sort button over its cards, and under it a search box (any of a player, school, opponent, play, season or rarity, every word matching) with the kinds, the player, the season, the rarity (holos last) and the order: rarest first, highest score, newest pulls, by year or by game. The button counts what is changed, and the bin puts it all back. Every signed-in player can see every collection, so you can look through somebody's cards to choose from them; signed out, nobody can.

Nothing moves until the other player accepts. Then both sides move in one transaction, after a check that both of you still have what was put up. A card can be in several offers at once, and only the first one accepted gets it; the rest come back Expired when they are answered. The player who offered can take an open offer back, and the other can decline it.

New offers show as a red circle with the count in white on the Cards link in the nav (and on the Trades page button), so one does not sit unanswered because nobody opened the tab. An offer stops counting once its player has had the Trades page (#cards/trades) open with it on screen: the page marks it seen (mark_trades_seen(), which sets card_trades.seen_at), so it is seen on every device, and it keeps a red New tag until they leave the page. Sets are still claimed once per player, so cards passed around can help more than one player finish a set.

propose_card_trade(), respond_card_trade() and cancel_card_trade() in supabase/schema.sql do the work, and check that the cards are really there, that only the player an offer is to can answer it, and that nobody has more than ten offers out at once.

The showcase

Up to six cards a player picks to show off, in one row, on their own player page under the header. Anybody can see it, signed in or not. On your own page, Edit turns each slot into Replace and Remove, and an empty slot into + Add a card, which opens your collection to pick from; Save puts it up. Only cards you own can go up, and a card traded away comes down on its own. A player with nothing up has no Showcase panel for anybody else.

Any of the six slots can hold one of your Player Cards instead: the picker lists your Player Cards first, then your cards, and either goes in the slot the same way, with the same Replace and Remove. A Player Card is kept as pc:{slug} among the showcase's ids, can go up once you hold a card of that player, and comes down on its own once you hold none.

Anything on a showcase opens up close. A card opens the way it does in the collection, with who else has it. A Player Card opens as big as a card up close, with whose it is, its rank and what that pays, the points and cards behind it, how far it is to the next rank, what it has earned and about what it pays a game (zoomPlayer() in js/cards-view.js).

Admin controls

The Cards panel on the admin page shows every player's cards, different cards, sets claimed, open trades and showcase.

Demo a pack rips a pack right there in the same overlay a real one opens in, to try the openings out. Nothing is paid and nothing goes into anybody's collection, the recent pulls or card_packs: it never touches Supabase. Best card makes the last card a Legendary, Mystic or Iconic so those openings can be seen without waiting on the odds, and Another demo rips the same again.

Under it are two resets for one player or for everybody:

• Clear library takes every card out of the collection, empties the showcase, cancels open trades and resets the goals, so the player starts the Library over from nothing. Diamonds stay spent.
• Reset goals only lets every Library set be claimed again. Cards are kept.

Rewards already paid stay paid either way.

Neither runs until RESET is typed, and neither can be undone. Both are database functions that refuse anybody who is not in admins.

In the database

player_diamonds and diamond_awards (see Diamonds above, readable by their owner and an admin), player_cards (who owns which ids and how many, readable by any signed-in player), card_packs (every pack opened and what was in it), card_sales (every quick sale, readable by the seller and an admin), card_set_claims, card_trades (readable by the two players in it) and card_showcases (public). None has an insert, update or delete policy: cards only move through open_card_pack(), claim_card_set(), quick_sell_card(), the trade functions and the admin's two resets, each one transaction, so a pack is never paid for without arriving, a trade never half happens, a card is never sold twice and a reward is never paid twice. The small helpers those share (give_card, take_card, give_booster, owns_cards, prune_showcase) are taken back from the browser's roles, since they do no checking of their own. Re-run supabase/schema.sql after pulling this change, or the tab says it cannot load the cards.

---

The five pages

The tab is five pages, picked from a floating island of five icons near the bottom of the screen (Sam's glyphs in icon/cardtabs/, after the Uiverse "liquid" radio group): an orange slider glides under the picked icon, and each icon keeps its page name as a hover title and for screen readers. The page goes into the address as #cards/players, #cards/shop and so on, swapped in place so switching reloads nothing. A page not showing is parked rather than hidden (content-visibility: hidden, falling back to hiding it where a browser lacks that): it is not drawn, but it keeps its layout, so going back to a binder of a couple of hundred cards is instant instead of laying every card out again. The binder is also left alone when what it would draw is what it already shows.

While a Cards page is on screen it checks in every twenty seconds, quietly: where each trade stands (tradeStates) and the diamond balance, and on the Shop page every other check the shelf, the market and the recent pulls. A new offer, a card sold off the shelf, somebody's pack or a listing of yours sold shows up without a reload, and only the part that changed is redrawn. Nothing is redrawn while a pack or card is open, the trade builder or crop tool is up, the phone's Sort menu is out, a box is being typed in, or within a few seconds of a tap (poll() and busy() in js/cards-view.js).

• Cards — the collection and nothing else. The page itself scrolls it (no box inside the page), drawn a few rows at a time, with the filters pinned under the nav on a desktop.
• Players — the six Player Cards (below).
• Shop — the Daily Spin (gone once it is spun, until the next one), packs and Buy Diamonds, Today's Cards, the Market and Recent Pulls. The odds are not shown on the page.
• Trades — offers to you, your offers, recent ones. The page button counts the new offers to you, in red, until the page has been opened.
• Achievements — the library of sets. The page button counts sets ready to claim.

Player Cards

Every account has a Player Card for each of the six, for their whole career. It is the fantasy points printed on every copy of every one of his cards the account holds — duplicates count again; a card sold, traded or up on the market takes its points with it — graded on the card ladder:

Rank	WR / RB	TE	QB	Bonus a game
Common	1 card	1 card	1 card	—
Uncommon	150	135	113	+10%
Rare	400	360	300	+25%
Epic	800	720	600	+50%
Legendary	1400	1260	1050	+80%
Mystic	2200	1980	1650	+125%
Iconic	7000	6300	5250	+300%
GOAT	28000	25200	21000	+500%

GOAT is a Player Card's alone: no play or performance card is ever GOAT, so it is not one of the card rarities, and its rung in PLAYER_RANKS carries its own name and colour. It looks like no other rank: a platinum frame engraved in a diamond cross-hatch with prism fire and four sparkles, the player in a platinum spotlight, a black plate with the name in platinum foil, and a platinum foot with a crown before GOAT. Up close the sparkles twinkle and a band of prism light sweeps across it. GOAT takes four times Iconic's points. Iconic took 3500 points until Sam doubled it on 2026-10-20.

A quarterback's average card scores about 9.6, a back's or receiver's 13.4 and a tight end's 11.9, so their rungs are ×0.75 and ×0.9 of a receiver's (RANK_SCALE). Every game the player plays pays the account a bonus: the game's diamonds before any booster, times the rank's bonus, rounded up. It goes in diamond_awards as pc-{slug}-w{week} through pay_player_card_bonuses(), on sign-in, when the admin publishes (their own account; Postgres pays a Player Card only to whoever is signed in), and on any open page within a minute of a week going out. Each game is paid once, at the rank held when it is first paid, and never topped up; games at Common go in at 0. The first time an account is paid, every game already out goes in at 0 (a pc-start row marks it), so Player Cards pay from the next game on. A playoff game's bonus is twice it, the same 2x as the game's own diamonds. A game's bonus is held to 400. The ladder is PLAYER_RANKS in data/cards.js.

A Player Card follows Sam's sketch, in the binder card's frame in the rank's colour:

[ PLAYER           | 354 FPTS ]
  his picture (team logo in the corner)
  Andrew Parr
  ALABAMA · TE
  EARNED 200 💎 · +10%
  [■■■■■■■             ]
        46 TO RARE
[ ◆ UNCOMMON          CAREER ]

Earned is every diamond that Player Card has paid you, all seasons (its pc- rows in diamond_awards), and beside it the rank's bonus; hovering it gives about what that is a game. On a player's own page the earned figure comes from showcased_player_cards(), which adds it for the cards put up. Before any card of his, the card is grey and the bar runs to Uncommon. The bar is the colour of the rank held. The face is playerCardFace() in js/cards-view.js. It is drawn on a binder card (.ege-tcard), so a Player Card is exactly the height and width of a play or performance card and gets the same effects for its rank: Rare's orange foil, the sheen from Epic up, Legendary's purple glow, Mystic's moving gold and Iconic's holo. The Players page lays the six out in three rows of two (one a row once the screen narrows), each card as wide as one in the binder, with what it means beside it: his name, FPTS and how many cards (and how many different) make them, a bar and how many more FPTS reach the next rank (not what it pays: Sam, 2026-10-20, there and up close), what the card has earned so far with its bonus now and about how much that is a game, then two icon buttons in the button's ink, Sam's cards and player glyphs from icon/cardtabs/: See his cards (the collection filtered to him) and Add photo (Change photo once there is one), each saying so on hover. The card itself opens up close, the same close-up as off a showcase. The page is just the drop-down and the cards: no words above them, no season's bonus total in the head and no Ranks & Bonuses table under them (Sam, 2026-10-20; the ladder is above).

Somebody else's Player Cards. The drop-down above them, centred, switches to any other account's six, and while somebody else's are up their picture pops in beside it. Their cards come from their collection, which anybody signed in can read; what theirs have earned comes from player_card_earnings(), which hands over the Player Card totals from diamond_awards and nothing else of it. See his cards opens their collection filtered to that player. Photos can only be changed on your own.

Your own photo on a Player Card. Add photo beside a Player Card picks a picture from the phone or computer, then opens the crop: the picture in a frame the shape of the card's photo, dragged to move it and zoomed with the slider, a pinch or the mouse wheel. Crop draws what is in the frame at 480×310 as a WebP of about 10–50 KB and shows the card with it; Crop again goes back to the frame as it was left, and Use this photo keeps it. Redrawing it that way also leaves the camera's location data behind. It replaces the headshot on your copy of that Player Card only, on the Players page and on your own page; × takes it down. One photo per account per player, so 36 at most across the site.

They are rows in card_photos (set_card_photo(), clear_card_photo()), each picture kept as a data URL in the row: 36 small pictures are a few megabytes, so there is no storage bucket to set up. set_card_photo() turns away anything that is not an image or is over about 100 KB, and anybody without a card of that player. Anybody can see them, signed out too, the same as a player's page. On the admin page the card table counts each account's photos with a Remove that takes them all down, and Clear library takes them down too.

Player Cards can go in the showcase on the player's own page, in the same six slots as cards (see The showcase). Anybody can see them, signed out too: showcased_player_cards() hands over only the cards of the players put up, which is what their rank is worked out from. Player Cards that were in the old row of their own (card_showcases.player_cards) move into the slots, first, as many as fit, when the schema is re-run.

The collection's filter row ends in a bin that resets every filter and the sort, and goes back to your own collection. Holos are picked from the rarity drop-down, after Iconic.

Only problems show in the banner over the Cards pages and the Shop: an offer sent or taken back, a sale, a buy or a listing that went through shows in the page itself, with no green banner to say so. The admin page keeps its banners, which carry the next step to take.

Today's Cards and the Market

Today's Cards: six cards a day, the same six for everybody — one Legendary and five rolled on the pack odds up to Epic. Common 10, Uncommon 12, Rare 15, Epic 20, Legendary 25 diamonds (SHOP_PRICES, and buy_shop_card() — change both). A card bought shows SOLD for everybody. The shelf turns over at midnight Eastern: the first visitor of the day proposes it (rolled from the date) and card_shop_today() keeps that one.

The Market: a player puts a card up at a price of 1–999 diamonds. It leaves their collection while listed, goes to the buyer when bought (the diamonds go to the seller), and comes back when taken down or after 24 hours. Three listings a day each, counted over the last 24 hours whatever became of them (list_card_for_sale(), MARKET in data/cards.js). Clear library takes a player's listings down with nothing returned.

A card can go up two ways: Sell a card on the Market, which picks from the collection, or Sell beside Quick sell on any card of yours up close in the Cards page, which opens a price and Put it up. That Sell button is greyed out once the day's three listings are used.

Seasons, and putting a week out

One file per season

stats/
  2018.js
  2019.js
  2020.js

Each holds the whole season — who each of them plays, when, and what they did in it. There is no separate schedule file: a game and what happened in it are the same thing, so they live on the same line.

{ week:  2, date: '2018-08-25', kickoff: '7:00pm',
  opponent: 'Del Norte', home: true, conference: false, scouts: true,
  result: { teamScore: 28, opponentScore: 14 },
  booster: 'boost-2-5',
  stats: { carries: 18, rushingYards: 132, rushingTd: 2, ... } },

One of the six traded in the NFL sim during a season has the trade in his season's file, trades: { 'andrew-parr': [ { week: 9, from: 'detroit', to: 'dallas' } ] }: from that week his games are the new team's, and every one before it stays as he played it, for the team he played it for. The NFL Sim tab stores a trade in Supabase (sim_trades) with the new team's games until Download the season files writes it in. His schedule says Traded to Dallas above the first game for the new team, the team pages of that season list him on both, and a team he joined or left mid-season shows its own record rather than his.

result and stats are null until a game has been played. Nothing generates them — the numbers come from wherever you generate them and are typed in, by hand or through the season editor. booster is the performance booster that was riding on the game, or the wager (wager-for / wager-against, with the line it was taken at beside it), and once a week is out that is where it lives for good, so the row behind it can be cleared out of Supabase. bigPlays is an optional list of strings, one per big play, shown under the stat line in the Discord post.

injured: true marks a game the player sits out hurt, injury says what with — 'Bruised Shoulder' — and health how healthy he is that week, 0 to 100. The Discord post shows DNP Injured in place of his stat line, and while that week is the one being played his page carries the injury report.

overtime: true marks a game that went to overtime. The score stays the final score; the flag adds /OT to the result on the site (W 31–30/OT) and (OT) after the opponent in the Discord headline. It is a checkbox beside the score in the season editor, and only written into the file when it is on.

Two more flags mark the postseason:

{ week: 14, date: '2018-10-27', kickoff: '1:00pm',
  opponent: 'St. Charles North', home: true, conference: false, playoff: true,
  result: null, booster: null, stats: null },

{ week: 14, date: '2018-11-16', kickoff: null,
  opponent: null, home: false, conference: false, playoff: true, bye: true,
  result: null, booster: null, stats: null },

playoff: true is listed with two asterisks and takes no booster. bye: true is a round drawn into the schedule that is not played at all — no opponent, no kickoff, no stat line, and nothing for the editor to ask for. An NFL bye week in the regular season is the same row, bye: true with no playoff, on the Sunday of its week, so the NFL's 18 weeks are all there.

division: true marks an NFL division game, the NFL's own asterisk (six of the seventeen): the schedule stars it the way college stars a conference game, and the legend says division game. An NFL game has conference: false.

The postseason is one week whatever the calendar says. Five schools in four states play their first round across three different Saturdays, and a week is the bucket the admin publishes, not a row on a calendar — so every playoff game is week 14, and the schedule draws a dotted rule above the first of them.

The rest of the draw

A thirty-two-team bracket is thirty-one games. The ones a player's own school plays are up in games; every other one lives in a playoffs block in the same file:

playoffs: {
  normal: [
    /* first round */
    { week: 14, results: [
      /* the left half, top to bottom */
      ['Rockford East', 24, 7],                   /* Simeon */
      ...
      /* then the right half */
      ...
      null,          /* Normal Community v St. Charles North — his own */
    ] },
    /* second round */
    { week: 15, results: [] },
    ...
  ],
  ...
}

One entry per bracket, keyed the way that season's draws are keyed in data/brackets.js. Each is a list of rounds, as many as the draw has plus the final; each round is the week it is played in and one result per matchup, in the order the bracket draws them — the whole left half top to bottom, then the whole right half. A result is [winner, winner's score, loser's score], and the winner is named rather than pointed at so a line can be read and changed without counting boxes.

null means the site works it out. Two cases: a bye, which advances on its own, and the one matchup his own school is in — that is a game in the same file with a stat line on it, and a bracket that could disagree with the schedule about whether he won is worse than no bracket. Change that game's score and the draw changes with it.

A team that went through without a game is [winner, null, null] — North Carolina's Garner in 2019, drawn against a 3 seed the bracket never filled and printed as TBA. It advances like any winner and is drawn with no score.

A round is drawn once its week is published. Nothing in the bracket is settled before then, the same rule the schedule, the record and the credits already follow — so the draw can never be ahead of what the admin has put out. Publishing week 14 fills the first round in and stands the second round up with the teams that won it, because round two's matchups are round one's winners. Fill a later round in and publish its week, and it moves on again.

To change who goes through, edit one line. To change how his own school did, edit his game up in games.

The draws themselves — the seeds and the first-round matchups — are in data/brackets.js under their season, EGE.brackets[2019] beside EGE.brackets[2018], so a new year's brackets go in next to the old ones and the season switcher still shows 2018's draw when you look back at 2018. A school with no entry for a season (Bloomington in 2019, Carlsbad in 2018) has no Tournament switch that season.

A new season is a new file and one more year in the SEASONS list at the top of js/site-data.js. The bot finds them on its own. A season's file can go in before the season is live: the site shows seasons up to EGE.currentSeason and no further, so a file for next year waits, editable from the admin page, until the season is rolled over.

2020 is the first college season, laid out from each school's real 2020 schedule as it was announced before COVID-19 rewrote it — the non-conference games included, and USC and Alabama opening against each other in Arlington. Kickoff times had mostly not been set by then, so they are null (a dash on the page) until they are filled in. A game at a neutral site carries neutral: true and reads as vs.

2021 is set up and waiting: stats/2021.js holds every regular-season game each school really played in 2021, with dates and kickoffs where it is played, and no results yet — Isaac's opens in week 0, Illinois and Nebraska on August 28, the Saturday before Labor Day weekend. It was written by node tools/build-season.js 2021, which reads the schedules from ESPN and saves through the same code as the season editor; it refuses to write over a file that is already there. Conference title games, bowls and playoffs are not in it — those go in once the regular season is played — and no game is marked for scouts yet. The year is in SEASONS in js/site-data.js, so the admin can edit and publish it now; it shows to everybody once Roll the season over points data/season.js at it. Rosters (data/rosters.js), defense grades (data/defenses.js) and the conference races (data/conferences.js) all have 2021 in already.

2022 is in, played out in full and waiting to be published. The fixtures are each school's real 2022 regular season — Isaac's at Texas and Paxon's at North Carolina now — and Alabama and Texas meet in week 2, so Andrew's and Isaac's files carry the same game from both sides. The results are the real margins moved by what the six bring to their teams and a game's worth of luck, and the postseason follows from them: Alabama beats Georgia for the SEC title, Texas wins the Big 12, USC loses the Pac-12 title game to Washington and UNC the ACC's to Clemson. The College Football Playoff is the real four with the seeds moved (EGE.brackets[2022]): Michigan v Ohio State in the Fiesta and Georgia v TCU in the Peach, Ohio State beats Michigan, Georgia beats TCU 65-7, and Georgia wins the title 42-41 over Ohio State in week 19. Alabama, SEC champion but the first team out, beats Texas in the Sugar Bowl. Postseason weeks 15, 18 and 19 are Playoffs Week One, Two and Three. Every stat line was worked out from each player's season, with each game's share moved by the opponent's 2022 defense and how the game went. The rosters, defense grades and the five conference races (the Big 12 and the ACC are new, and the Pac-12 and Big 12 have no divisions) are in for 2022 too. Those were built from the cfbfastR mirrors of ESPN's 2022 data on GitHub, since ESPN and 247Sports could not be reached when they were written; the tools still read ESPN. It shows once the season is rolled over and its weeks published, the same as 2021.

Scouts are at four of each player's 2022 games, the four toughest on his regular season: the strongest opponents by where they finished 2022 (their end-of-season Elo in the cfbfastR schedules), with the defense he faces counted in, run or pass. They are all regular-season games, so Intel shows all four from week 1 and each one can still take a booster.

Paxon is a receiver and nothing else in 2022, with four carries all season. Sam runs for over 2,000 yards, Cooper is USC's lead back with about 1,100 rushing and 500 receiving, and Andrew finishes with 17 touchdown catches. No back, receiver or tight end scores more than three touchdowns in a game: each one was less likely to land in a game that already had one. The season has one injury, drawn at random with each player's chance going up as his injury rating goes down: Jaykeb strains a hamstring against Iowa in week 8 and misses the trip to Penn State, which Ohio State still wins 44–31 with Sam carrying more of it.

2023 is in, its regular season played out in full and waiting to be published, for both leagues. Sam (Seattle), Andrew (Detroit) and Jaykeb (Houston) play their teams' real 2023 NFL schedules, all eighteen weeks from nflverse with the bye in its week, and Paxon (North Carolina), Cooper (USC) and Isaac (Texas) their schools' real 2023 regular seasons, weeks 0 to 13. The results are the real margins moved by what each player brings to his team and a game's worth of luck, as in 2022, until the NFL sim takes the NFL over from week 10 (see NFL Sim under the admin page): the Lions finish 10–7 and first in the NFC North, the Seahawks 11–6 and first in the NFC West, the Texans 8–9 and second in the AFC South, Texas 10–2, North Carolina 9–3 and USC 8–4. Week 2 is Andrew's Lions against Sam's Seahawks, 33–32 Detroit, and both files carry it from their own side. The one injury is Andrew's, a groin strain in week 6 that keeps him out against Tampa Bay and Baltimore. Jaykeb throws for about 4,000 yards as a rookie, Sam is Seattle's second back behind Kenneth Walker, and Isaac throws 30 touchdowns. Scouts are only at the college players' games, four each as in 2022; an NFL player has nobody left to scout him.

The 2023 postseason is in weeks 19 to 22, the NFL's playoff weeks. The NFL sim played the NFL's: Seattle beat Minnesota, New Orleans and Philadelphia and lose Super Bowl LVIII 23–20 to Kansas City, and Detroit go out to San Francisco in the wild-card round. The colleges' bowls are in week 19 with the wild-card round, whatever their real dates (Sam, 2026-10-20), so both leagues' Discord posts that week are headed Playoffs Week One. None of the three plays a title game. Texas, Oklahoma and Oklahoma State finish level at 7–2 in the Big 12; Texas and Oklahoma State never met, nobody beat both of the others, and Oklahoma State lost to Iowa State, the best placed school all three played, so Oklahoma (who beat Texas) and Oklahoma State (who beat Oklahoma) play for the title and Texas is third. North Carolina finish sixth in the ACC and USC fourth in the Pac-12. With Texas out of the College Football Playoff, Florida State takes the last place in it, so Louisville, the ACC's runner-up, goes to the Orange Bowl in its place and its real Holiday Bowl spot goes to North Carolina, the best ACC team left after NC State (Pop-Tarts Bowl). The bowls, played out from each side's points for and against with seeded dice (seed 34), with each player's line drawn under the same rules as the regular season:

• Alamo Bowl (Dec 28), the Big 12's first pick after the New Year's Six against the Pac-12's: Texas 33, Arizona 12. Isaac 21 of 35 for 268 yards and 3 touchdowns.
• Holiday Bowl (Dec 27), the ACC against the Pac-12's next pick after Arizona: North Carolina 37, USC 32. Paxon 8 catches for 139 yards and a touchdown; Cooper 31 carries for 157 yards and 2 touchdowns, and 6 catches for 77. Both files carry it from their own side, as week 2's Lions–Seahawks game does.

2024 is set up and waiting, with all six in the NFL: Cooper went 7th to Chicago, Paxon 34th to the Giants and Isaac 39th to Tampa Bay. stats/2024.js holds each of their teams' real 2024 regular season from nflverse's games.csv, weeks 1 to 18 with the bye in its week, the kickoffs in each team's own time and every result still to come. The Bears' London game (week 6, Jacksonville) and the Giants' Munich game (week 10, Carolina) are neutral. EGE.currentSeason is still 2023, so only an admin sees 2024 until it is rolled over to, and it has no block in data/conferences.js yet.

2025 is set up the same way, with the six on the same teams. stats/2025.js holds each team's real 2025 regular season from nflverse's games.csv, weeks 1 to 18 with the bye in its week, the kickoffs in each team's own time and every result still to come; it matches the sim's 2025 schedule in nfl-sim/data/schedules.json game for game. None of the six's teams plays abroad in 2025, so nothing is neutral. The Seahawks and the Buccaneers close on the Saturday of week 18, so the sim dates their playoff games from the Sunday after it.

2026 is set up the same way, every result still to come, with each of the six on the team he finished 2025 with: Sam on the Steelers and Andrew on the Panthers after their 2025 trades. stats/2026.js holds each team's real 2026 regular season from nfl-sim/data/schedules.json, written the way the sim writes a traded player's games (rowsFor in nfl-sim/lib/six-trades.js), so it matches the sim game for game. The Steelers' week 7 game with the Saints and the Texans' week 6 game with the Jaguars are abroad, so they are neutral. A move the sim makes between seasons lands in the file's trades from week 1 and swaps in the new team's games, the same as one made in season.

Conference standings

The Conference line on a college season's header is the team's place in its division — 2ND in Big Ten East — as of the weeks published so far, so it moves the way the record does. Only the six's schools are in the season files, so the rest of each conference is in data/conferences.js: a whole conference schedule for everybody else, built around the fixtures already in stats/2020.js and following each league's real shape (every division rival; the Big Ten's three crossovers with Indiana-Purdue locked; the SEC's permanent and rotating crossovers; four of six across the Pac-12 with the California schools always meeting; eight of ten in the Valley), with the rivalry games on their usual weekends.

Those games are played out from the strength each school actually showed in 2020 — points scored and allowed a game in the real final standings — with seeded dice, and of the first four thousand seeds the one kept for each conference is the one whose final table comes closest to how those schools really finished. A game in there counts once its week is published. The six's own conference games are read from the season file as they are published, so a result corrected in the editor moves the standings with it.

A division is ordered by conference winning percentage, then wins, then losses, then head to head among the schools still level, then conference point differential. The line reads TBD until the team has played a conference game.

A real season (2021) also lists every school's games outside its league in nonConference, as they really finished, which is what the standings' overall record counts alongside the conference games. A game against one of the six's schools is left out of that list — Oregon's at Ohio State is in the season file, played out — and read from the season file instead. 2020's races are drawn, with no outside games to count, so its standings have no Ovr column.

node tools/build-conferences.js {year} writes one season's block of the file and leaves every other season's exactly as it was. Run it again only if a fixture changes. 2020 is drawn: the real 2020 was cut short, so a schedule is made up in each league's shape, as above. 2021 is real: the season was played in full, so its conference schedules are read from ESPN as they were played, and only the scores are played out, from each school's real 2021 points for and against, with the seed judged against its real 2021 conference record. A new season goes in SEASONS at the top of the script, usually as another real one.

2023 keeps the real scores for every game the six are not in: the Big 12, the ACC and the Pac-12 (all without divisions now), and the whole NFL. ESPN could not be reached from the build, so the colleges came from cfbfastR's copies of the 2023 schedules and the NFL from nflverse's games.csv, not through the tool. The AFC and the NFC are nfl: true, four divisions of four each, games the games inside the conference and nonConference those against the other one. An NFL division is ordered by the whole record, then head to head, the division record, the conference record and the points differential, the way the league orders one (EGE.nflResults and nflDivisionTable in data/games.js), so the header line reads 1ST in NFC North.

The file is the season

Nothing is logged anywhere else. There is no stats table in Supabase — the tables there are accounts, admins, balances, inventory, stickers, the credit ledger and which weeks are out, and not one of them holds a number from a game. Change a touchdown in stats/{year}.js and the schedule, the record, the game log, the season totals, the ratings the shop is priced against, the Discord post and the credits that touchdown is worth all move with it, because every one of them reads that file and only that file.

Which only holds if the browser reads the file rather than its own copy of it. A static host serves everything with a cache lifetime — GitHub Pages sends max-age=600 — and for those ten minutes a browser that has been here before answers from its own cache without asking. A committed correction is simply not there yet, and pinned to a home screen it can hold on longer than that.

That was a real problem rather than a slow one, because the admin page pays touchdown credits out of whatever that browser thinks the file says. Paying a published week from a stale copy writes the old number into the ledger, and the ledger never pays a number back down.

So the three files the admin page regenerates go through js/site-data.js instead of <script> tags:

data/season.js      which season is live, and which are locked
data/ratings.js     the ratings a locked season baked in
stats/{year}.js     the season itself

It writes them out as <script> tags with a cache-busting query, which is a different URL every load and therefore always a fetch.

Everything else — index.html, js/app.js, the stylesheets, data/offers.js and the rest — is kept current by sw.js, a small service worker. It makes the browser check every one of the site's own files with the server before using its copy. An unchanged file costs a quick 304, and a changed one arrives new the first time. It keeps no copies of its own, so it cannot go stale itself, and it leaves other sites' files (Supabase, the fonts) alone. The header of sw.js says how to take it out if that is ever wanted: deleting the file is not enough.

A page that is already open is a different problem: a phone does not reload a tab it put away, or the home-screen app when it is opened again. So coming back to the page after 30 seconds or more asks Supabase for the published weeks and the ratings again, and redraws in place, keeping your scroll position, if anything changed.

And a page left open keeps up by itself: once a minute while it is on screen it asks the same again, along with the balances in the nav and the count of new trade offers. A week gone out since pays the signed-in player there and then (credits, game diamonds and their Player Cards' share) and redraws, so nobody has to reload after a publish. The redraw waits while somebody is in the middle of something: the portal or a card open, a sticker picked up or its drawer out, a box being typed in, or a tap a moment ago. Nothing is asked while the page is put away. All told an open page asks Supabase for about five small things a minute, ten on a Cards page and under twenty on the Shop page.

Why document.write, of all things. Because these have to be in place before js/app.js runs, exactly as they always were. document.write from a parser-blocking script inserts them into the parse stream at that point, so they execute in order and js/app.js still finds a fully loaded season the moment it starts.

Fetching them instead and starting the app afterwards was the obvious move, and it was wrong: a deploy is not atomic in a browser. index.html is cached too, so for a while after a push a returning visitor runs one file old and one new. An app.js that waits on a loader the cached index.html never mentions dies on the spot, and so does a cached app.js that expects the season to already be there. Keeping the contract synchronous is what makes both halves of a half-applied deploy work — which is worth more than the 304s the fetch would have won.

The cost is that those three are re-downloaded rather than revalidated: about 45KB a load today. Opened from a file:// path the query is left off, because there is no cache to get past and a query on a file URL is a path that does not exist.

Everything else — the code, the kit, the players, the shop — is still tagged in index.html the ordinary way. Those only change when a deploy does.

Nothing is out until you say so

A season file can hold every result of the year and the site will still show none of them. A week becomes real when you publish it, which is a row in Supabase rather than anything in the repository. So the numbers can sit in git for as long as it takes, and a booster somebody used after the file was written can still be put right first.

EGE.isFinal is what everything reads, and it answers no until the week is published — which keeps the schedule, the record, the game log, the touchdown credits and the Discord bot from ever getting ahead of you. The season editor reads EGE.hasResult instead, because it has to show you what has not gone out yet.

Publishing

Weeks, on the admin page, is a row per week: how many games it holds, how many have numbers in the file, whether Discord has had it, and a button.

Publish does all of it on the one click:

• the scores and stat lines appear on the player pages
• the records move
• everybody is paid their game credits and diamonds, and both show on the schedule row that earned them
• the week goes to Discord, with a link back to the site

Pull back takes a week off the site again, one click, no confirmation, as often as you like — testing this means publishing and unpublishing the same week over and over. Credits already paid stay paid: an award is a ledger entry, not a view of the schedule, so publishing again pays nothing twice.

Post again re-sends the Discord message without changing anything.

How the Discord post gets sent

From 2023 there are two channels, one per league, each with its own webhook: the college bot's and the NFL bot's. Publishing a week sends one message to each league that has anybody playing that week (see Two bots from 2023 under the Discord scores bot), and the Weeks row says Posted only once every one has gone; if one landed and the other did not, Post again sends only the missing one.

The webhooks are saved on the admin page. Under the week list are two boxes, College bot webhook and NFL bot webhook: paste each channel's URL (Server Settings → Integrations → Webhooks) and press Save. They go in the discord_webhooks table in Supabase, which only an admin can read or write (is_admin(), the same check the rest of the site trusts), so the URLs are never in the site's code where anybody who views source could post to the channel as the bot. A saved box is never filled back in; it shows Saved (…abc123), the last six characters, and pasting a new URL replaces it. To rotate one, delete the webhook in Discord, make a new one and save that.

The admin's browser reads the saved URL with the admin's own sign-in and posts straight to it, so the message lands on the click with nothing to deploy.

The edge function, for a league with no webhook saved. The browser hands supabase/functions/post-week the payload and its league instead; the function checks the caller is an admin, reads the saved webhook the same way, and otherwise uses its own secret for that league:

supabase secrets set DISCORD_WEBHOOK_URL_COLLEGE='https://discord.com/api/webhooks/...'
supabase secrets set DISCORD_WEBHOOK_URL_NFL='https://discord.com/api/webhooks/...'
supabase functions deploy post-week

The table needs supabase/schema.sql run again (it adds discord_webhooks and published_weeks.posted_leagues). Until it is, pressing Save says to run it first, and publishing still posts through the function.

Either way the week is published before the post is attempted, so a message that does not land is a message missing from a channel, not a week missing from the site. Post again retries just the message.

A failed direct post never falls back to the function: a request that reached Discord but whose reply the browser could not read looks exactly like one that never arrived, and trying the other route would post the week twice.

The trades bot

A third bot, in a channel of its own, posts nothing but trades (Sam, 2026-10-22): every trade that goes through in the NFL sim — approved from the queue, made in the builder, or made from a team page's trade finder — the moment it is saved. A free-agent move is a signing, not a trade, and is not posted. The message is headed Trade, with an embed per team in its colour and with its mark, saying what it receives: each player as position, name and overall (one of the six underlined and bold, as in the draft lists), each pick as its year and round, its number once the draft order is known, and whose it was when it was not the giving team's own. The footer says when in the sim it happened (2024 Week 5, 2024 trade deadline, 2025 offseason). The marks are the site's own pictures: a webhook can only use the server's :CHI: emoji by their ids. Each embed carries the week posts' transparent spacer, icon/spacer.png, so every one is the full width whatever is in it.

Its webhook goes in the third box under the week list, Trades bot webhook, kept in discord_webhooks as trades (run supabase/schema.sql again first: the table took the two leagues only). There is no edge function or scheduled bot behind it; with no webhook saved, trades simply are not posted. A post that does not land leaves the trade made and says so.

The season editor

Season File, under the week list, is every game in a week as the file has it: the score, the booster, and every number that position's line is typed from. The averages, the totals and the passer rating are not there — they follow from the rest and are worked out when the file is written, so a line can never disagree with itself.

Under the numbers is a Big plays box: type one play per line (44 yard receiving touchdown bomb) and they are written into the game as bigPlays and go out under the stat line in the Discord post. Blank lines are dropped. Big plays are only saved on a game with a score in it.

Edit as many weeks as you like; they are all held until you download. Pull in the stickers players applied takes whatever is on that week in Supabase and writes it into the file, so it can be committed and the rows behind it cleared.

Downloading writes stats/{year}.js back out — the same fixtures, your changes on top, every other week exactly as it was read. Commit it and the numbers are live, for the weeks you have published.

The stat lines

Defined once in data/statline.js and read by the player page, the season editor and the Discord post, so there is never a version of a quarterback's line that disagrees with another version of it.

Quarterback — C/ATT, PYDS, PAVG (PYDS/C), PYAC, PTD, INT, RTG, CAR, RUYDS, RUAVG, RUTD, LNG, SACK, FL

Running back — YDS, TD, CAR, RUYDS, RUAVG, RUTD, LNG, REC, REYDS, REAVG, YAC, RETD, LNG, TGT, FL

Wide receiver and tight end — YDS, TD, REC, REYDS, REAVG, YAC, RETD, LNG, TGT, CAR, RUYDS, RUAVG, RUTD, LNG, FL

A season totals up the way each column is meant to: summed down the column, a long is the longest, and a passer rating is worked out again from the season's numbers. Averaging a column of averages is how a stat page ends up lying.

On the player page

The game log sorts on any of its columns: click a heading and the table reorders on that stat, biggest first, because the question a stat table gets asked is who had the best day. Click it again for smallest first, and a third time to put the season back in the order it was played in. An arrow in the heading says which way it is running, and the sorted column is shaded the whole way down so the eye can follow it.

The season totals row is the season, so it stays the same line however the games above it are ordered.

A credits marker on the row says what the game paid, and a blue diamonds marker beside it what it paid in diamonds — more than the credits when a booster was on the game. Visible to the player and to an admin, nobody else, the same rule the boosters follow. The column is headed Paid.

---

The Admin Portal

#admin, a tab that only appears for an admin, and a page that refuses anyone else. It holds everything Sam used to do from the shop page, plus the two things that end a season.

It is split into tabs across the top — Weeks (the weeks, the Discord bots folded away under a line, and the season file), Accounts (credits and diamonds earned, and every account), Cards, NFL Sim and End of Season — one at a time, so nothing needs scrolling past (Sam, 2026-10-19). The tab is in the address (#admin/sim), so a reload comes back to it; plain #admin is Weeks.

NFL Sim runs the NFL sim (nfl-sim/) in the browser. The first time the tab is opened it loads js/nfl-sim.js (the sim and its data, about 1.4 MB, built by node nfl-sim/build/bundle.js) and the world from Supabase (nfl_sim_files, gzipped, admin only — the world holds the team ratings, which are never shown anywhere). Start the world takes the weeks the site has already put out as played (the six's games from their season file, everybody else's real scores) and the sim plays every NFL game after them. Then: play a week or on to the deadline / the end of the regular season, the playoffs a round at a time or all at once, the offseason and the draft (with a team and round for any of the six still in college). Every week the sim plays is stored for the site (Supabase sim_weeks, laid over the season files by js/stored-weeks.js), so Publish Week N under Play puts it out the way the Weeks tab's Publish does (the credits and diamonds it pays, the Discord posts) with nothing committed, and the next week can be played straight after: Play Week 5, Publish Week 5, Play Week 6. Week stats, under Play, is what a week produced, the week just played unless another is picked (each marked not out until it is published): the six's lines and fantasy points, then a card for every game with its score and its passing, rushing and receiving leaders, which opens to both teams' plays, dropbacks, runs, yards, third downs and time of possession, every player who threw, ran or was thrown to, sacks and picks, the scoring plays and who got hurt (nfl-sim/lib/weeks.js). Games played before 2026-10-21 have no box score, so theirs open to the scoring and injuries only. A stored week is the whole week as the sim played it, for each of the six in the NFL: their results, stat lines, big plays and injuries, every other NFL score for the standings, and their playoff games as they reach them. Nobody but an admin can read one until its week is out (a playoff game's fixture once the week before it is), and a week that is out stays as it went out. Under Play, a week the sim has played this season (or a playoff round) can be picked and Resim plays it again on new dice, or Undo goes back to before it. Either way every week played after it is undone, and so is anything changed since (the week's entry says how many changes), a trade of one of the six too; a week that is out on the site, or one before it, can't be. Each week's save point is kept in nfl_sim_files as rewind-{season}-{week}.json, fetched only when it is used, and the offseason clears them. Before every step the six's boosters on their games (game_boosters, then the season file once a week is out) and their QB and O-Line Connections (player_inventory) are read for the sim: a booster makes a good game likelier (a 2.5x locks in a solid to great one), and a connection counts while that quarterback is playing (see , section 5). takes the first rounds of the next draft written by hand, pasted or from a file (the Discord lists, ; numbered lines; a CSV with Pick, Team and Player; or JSON), checks every pick against the order and the class (unknown players, a team that does not own the pick, duplicates, missing picks), shows them, and keeps them: those picks go exactly as written, can't be traded while they wait, and the sim drafts the rest. The 2024 draft goes in the order from the 2024 mock draft doc in every round, whatever the sim's own 2023 said, and every real pick trade made before the 2023 kickoff moves its 2024 or 2025 pick (). Trades wait for or , and so does any star (85 or better) who would leave his team in free agency: he stays put unless Sam approves the move. Sam's own trades come from a builder: a team on each side, players and picks added from dropdowns, then (straight through) or (queued with what both teams think). Every trade that goes through is posted by the trades bot (see ). The sets any player's overall and ceiling — on a team, a free agent, or a prospect in the class the next draft takes — sets each of the six's (the website keeps their ratings, the sim keeps this), and shows every team's roster as each season ended (and this one as it stands), with who arrived and who left since the season before. downloads and with the stored weeks written in (and the stickers on weeks that are out), to commit whenever it suits -- the end of the season, or any time -- and says which stored weeks the committed files do not have yet. Each file keeps , when each week written in was saved: a week stored after that (played again, or undone) is laid over the file's copy, and a correction made in the file afterwards stands. After each step the rosters go to for the Teams pages (a box turns that off). The world can be downloaded, loaded back and started over. A step that fails saves nothing, and a world saved from another tab since the page loaded it is refused rather than overwritten.

Accounts — every balance and everything each player owns. Set a player's credits or diamonds (each box has its coin or diamond after it, and Save writes only the one that was changed, so it never puts back a number a game has paid into since), hand an item over without charging for it, remove anything, or pay an award off the earnings table — a salary or an honour, which all stack. An award is added straight onto the balance and not logged anywhere; the offseason 60 is not in the list, because it pays itself.

Above the six, everybody at once: a number, credits or diamonds, and Add (on top of every balance; a minus number takes it off, never below nothing) or Set (every balance becomes that number, after a confirm, since nothing keeps the old ones). Every player with a sign-in, one at a time, read then written; a balance that cannot be read is left alone and named.

Cards — who has how many cards, sets claimed, open trades and showcase, Demo a pack to try the pack openings out without paying or keeping anything, and the two resets, Clear library and Reset goals, for one player or everybody. See Cards → Admin controls.

Credits & Diamonds Earned — what the season has paid out so far, by player: the touchdowns or fantasy points scored, what the games paid, the allowance, and the diamonds the games paid. Awards handed out by hand are not counted.

End of Season — four steps, in this order, because the order is the only thing keeping anybody's season safe:

1. Lock the ratings. Downloads a data/ratings.js with every rating point and every offseason workout bought that season folded into the base numbers. Only the numbers change: every comment, weight and helper in the file survives, because the file is read and its ratings block spliced rather than rebuilt. A player nobody spent anything on comes out byte for byte as they went in, so the diff is only the players who actually moved. Commit it, and the improvements are part of the site rather than part of a database. It can be downloaded again as often as needed, before or after the lock is committed and before or after the rows are cleared: while a locked season's rows are still there, EGE.valuesFor counts them from where the season started rather than on top of the locked numbers, so a second download is the same file — plus anything bought since — and the site never shows a purchase twice in the meantime.
2. Log the season. Downloads data/logs/season-{year}.js: every game with the stats posted in it and the booster that was riding on it, every credit earned, and everything bought. A record, not a source — nothing on the site reads it. It is there so a season cleared out of Supabase is still on the record afterwards.
3. Clear the shop rows. Wipes everybody's rating points and offseason workouts, which the ratings file now carries instead. Unused performance boosters stay, and Intel lapses on its own. Two locks on this one: the ratings file has to have been downloaded in the same sitting, and the word CLEAR has to be typed. Everything it deletes is recoverable only from that file.
4. Roll the season over. Downloads a data/season.js pointing at the next season, with the one just finished added to lockedSeasons so it can never be locked twice. Commit it and everybody is paid their allowance on their next visit.

Building a file and clearing the rows it replaces are deliberately separate actions, in that order, with the commit in between. The portal will not let step 3 run before step 1, and says so rather than doing it.

Steps 1, 2 and 4 read data/ratings.js and data/season.js back over HTTP, so they need the site served rather than opened off a disk. Both refuse with a message rather than handing over half a file.

---

The overall, and where it came from

The numbers in data/ratings.js are where a season started. What a player has bought since shows beside his overall on his own page as a ticker — ▲ +3 — with the number he started at in the tooltip.

Once the season is locked those purchases are part of the base numbers, so the lock also writes EGE.ratingsSeasonStart inside the markers: the attributes that moved, at the values they started the season on. While that season is still the live one the ticker measures from there; once the season rolls over it is ignored, and the new season's ticker starts from the locked numbers.

At the end of a season the admin locks the ratings: everything bought is folded into the base numbers, the file is committed, and the rows behind it are cleared. Credits and unused performance boosters carry over; rating points and Intel do not, because they have become the ratings themselves.

---

Player Portal (login)

Each of the six gets an account and a private portal.

• Login — one account per player; a player only edits their own profile. Built on Supabase auth against a hardcoded list of six emails — nobody outside that list can hold an account. Sam Stogsdill's email is in; the other five are TBD.
• One password, set once — a player picks themselves out of the list. If their account has no password yet the portal asks for one twice; if it has one, the portal just asks them to sign in. Which form appears is looked up, not chosen by the player.
• No password changes from the site, and no email at all — no confirmation mail, no reset mail, nothing that costs anything to send. A player who forgets goes to an admin.
• Admin password changes — done in the Supabase dashboard under Authentication → Users, where any user can be given a new password. To put a player back on the first-time form afterwards, set their row in player_accounts back to password_set = false.
• Show password — every password field has an eye toggle, so what's being typed can be checked before it's submitted.
• Client viewer mode — an admin's portal has a switch that draws the whole site the way a player sees it (Sam, 2026-10-20): no Admin tab, no team overalls, nobody else's credits or boosters, no season ahead of the live one. It stays on, on that device, until it is switched off (EGE.wallet.viewerMode; EGE.wallet.admin() answers no while it is on and realAdmin() still knows). It changes only what the page shows: the database still treats the account as an admin.
• Overalls — a rating per attribute (speed, strength, catching, route running, etc. — final attribute list TBD) plus a single overall.
• Offseason workouts — the progression mechanic. Between seasons a player spends workouts as boosts to raise specific attributes. Workouts are a limited resource, so choices have a cost.
• Interactive layer — beyond workouts, the portal is meant to be something a player actually plays with between games. The Cards tab is that: packs, a collection and a library of sets, built from the season files (see Cards).
• Credits and inventory — 60 an offseason plus what a season earns, spent in the shop, with what was bought kept per account. Every player starts on 0.

The read-only side of the site (schedules, records, stats) stays public — no login needed to browse.

---

Files

Built so far:

• index.html — the homepage: player select, plus a placeholder player view.

• js/app.js — renders the six cards and routes #{slug} to a player view.

• data/players.js — the six players and the season ladder. Source of truth.

• js/auth.js — Supabase auth: sign in, the one-time password, session state.

• js/wallet.js — credits, diamonds and inventory: balances, buying, using, paying out a season, swapping credits for diamonds, and the admin's reach across every account (one balance or everybody's at once).

• data/ratings.js — the attribute list, the per-position weights, every player's ratings, and the maths that turns them into an overall.

• tools/ — scripts run by hand, never by the site: build-season.js {year} writes a new season's fixtures into stats/{year}.js (week numbers from calendar.js, shared with the conference races), build-conferences.js {year} writes that season into data/conferences.js, build-logos.js writes data/logos.js, build-defenses.js writes data/defenses.js and build-rosters.js writes data/rosters.js. node tools/<name>.js from the repository root. The last two read a few thousand ESPN (and, for the rosters, 247Sports) pages through tools/espn.js, which fetches them in parallel with curl and caches them in the system temp folder, so a second run takes seconds.

• bot/ — the Discord scores bot (see below).

• supabase/schema.sql — every table and policy: accounts, credits, inventory, admins, the stickers stuck on games (and the trigger holding them to five a season), the credit awards a season pays out, diamonds and their awards, which weeks are published (and which leagues' Discord posts have gone), the Discord webhooks (the two leagues' and the trades bot's), the card collection, and the NFL sim's world, rosters, stored weeks (sim_weeks) and the six's trades (sim_trades). Run it in the Supabase SQL editor; it is safe to run again, and it must be re-run after pulling a change that adds a table. Re-run it after pulling this change so sim_trades is there for the six's trades.

• supabase/functions/post-week/ — an edge function that posts a week to the college or NFL channel for an admin, from the saved webhook or its own secret, for a league with no webhook saved on the admin page.

No build step and no bundler: open index.html in a browser, or serve the folder with anything static. Data files are plain <script> globals rather than ES modules so the site also works straight off the filesystem — including the three that js/site-data.js writes out, which drop their cache-busting query when there is no cache to get past. The only external dependency is supabase-js, loaded from a CDN.

The live site is on Vercel, which deploys main on every push and builds a preview for every pull request. If a merge does not show up on the site within a minute or so, Vercel did not hear about the push: redeploy the latest main from the project's Deployments page, or push to main again.

Connecting Supabase

1. In your Supabase project, open Settings → API and copy the Project URL and the anon / publishable key.
2. Paste both into js/supabase-config.js. The anon key is meant for browser code and is safe to commit. The service_role key is not — it bypasses every security rule and must never appear in this repo.
3. Run supabase/schema.sql in the Supabase SQL editor. It creates player_accounts (whether an account has a password yet), player_credits, player_inventory and admins, with the policies that keep each player to their own rows and let an admin reach every row. Without it the portal still signs people in — it just can't tell which form a player needs, and the shop has nothing to read or write.

Until those two values are filled in, the site runs normally and the portal reports that login isn't configured yet.

Worth knowing: setting the first password doesn't prove who is setting it, so whoever gets to an unclaimed account first claims it. With six known players that's usually fine, and once a password is set nothing on the site can change it — but it does mean each player should claim their own account before the site is shared around.

On a phone

Most of it is read on an iPhone, so the layout answers to one. The width was already fine; what was not:

• Safari was zooming the whole page out. A <select> is as wide as its longest option, and at a tappable font size a name that long is wider than the screen — the page came to 435 points on a 390 point phone, and Safari's answer to that is to shrink everything to fit. That is most of what "it looks tiny on my phone" turns out to be.

• Safari was zooming in on every form field. iOS zooms when you focus a control whose text is under 16px and does not zoom back out. Every control here was 13 or 14.

• The tables hid the point. The schedule scrolled the result off the right; the shop scrolled the buy buttons off. Under 620px each row becomes a block instead — same markup, same cells, only the layout changes — so what you came to see is under your thumb.

• Tap targets. Apple asks for 44 points square. The buy buttons were 33 tall, the admin's publish buttons 30, and the arrow that folds a panel away is held to 44 square.

• Nothing hover-only. A finger cannot hover, so the cross that means "tap to peel this booster off" never appeared at all and an applied sticker looked stuck.

• The notch. viewport-fit=cover with env(safe-area-inset-*), and 100dvh rather than 100vh, which on iOS is the window with the toolbars hidden.

• The type was laptop-sized. Every size on the site is set in rem, and the root size is one fluid value: clamp(13px, 1.15vw + 8.87px, 16px). At 620 points and up it is 16 and nothing has changed; below that it eases down to 13 on the narrowest phone. No breakpoint fires and nothing is repositioned — the words simply stop taking room the layout needed.

  Two floors stop it going too far: --fb-label holds a small-caps label at 11px and --fb-small holds small print at 11.5, both with max(), which is a no-op at the full root size. The one absolute pixel size left is the 16px on form controls, which is Safari's zoom threshold rather than a size the design gets to choose.

  The kit's own type is in pixels and style.css is not edited, so the sizes this site uses are restated in rem at the top of site.css — each one the kit's own number over 16.

The roster turns sideways too — six portrait cards was six screens of scrolling before anybody had picked anyone, and it is two now. The player header rearranges: the picture and the name share the top line, and the four facts about the season drop under both and run the full width rather than being squeezed into what the name left of a 390 point screen.

Everything else is behind max-width: 620px or hover: none, so the desktop layout is untouched; that is checked as well as the phone one.

Kit and assets

Already in the repo and used as-is — style.css and organic-styles.css are not modified; page-specific styles go in site.css:

• style.css — GRIDIRON DB UI kit: tokens, layout, panels, nav, buttons, tags, inputs, tabs, filter chips, data tables, pagination, stat tiles, meters, alerts, hero, photo placeholder, scoreboard, modal, utilities.
• organic-styles.css — the Organic design-system stylesheet style.css builds on.
• headshot/ — the six player headshots.

Conventions the kit expects: state is a class (.is-active, .is-selected, .is-sorted, .is-on), and every border is square — the reset forces border-radius: 0 globally, with the football brand mark as the only round shape.

Theme hooks, overridden in your own :root:

:root {
  --fb-accent: #e0a117;   /* CTA / highlight */
  --fb-grass: #3d472b;    /* field ground */
  --fb-shadow: 6px;       /* hard offset shadow depth */
  --fb-ink: #20261a;      /* borders + dark panels */
}

---

Build Order

One thing at a time, in this order:

1. This README — the spec. ✅
2. Player data file — the six players, schools, positions (TBD where unknown), headshot paths. One source of truth everything else reads. ✅ data/players.js
3. Homepage — six-card player select, wired to the data file. ✅ index.html
4. Player page shell — #{name} routing, header, empty schedule/record/game log sections. Routing and the header are in; the schedule, record, and game log sections are not.
5. Schedule data + display — 2018 junior-year high school schedules per school, rendered with results and running record. Later seasons follow the same shape once 2018 is working. ✅ The fixtures are in stats/{year}.js, the player page renders them, and the Discord bot posts them. Results fill the table's last column as games are played.
6. Stat lines — a full game log per position. ✅ The columns are in data/statline.js, the player page renders them with season totals, and the bot posts the same line. Weeks are rolled and published from the admin page — see Playing a Week.
7. Login + portal — Supabase auth, accounts, attributes, overalls. Login is in: allowlisted emails, a password set once, and the signed-in player's headshot in the nav. The portal behind it — attributes and overalls — is not.
8. Offseason workouts — the boost mechanic. ✅ The shop, rating points, training, boosters and the sticker layer are all in.
9. Season automation — credits that earn themselves as results are posted, and a season that can be ended, locked into the repository and rolled over from the admin portal. ✅ js/exports.js, data/season.js, #admin.
10. Playing the season — a hardcoded file per season, published a week at a time from the admin page, with the Discord post on the same click. ✅ stats/{year}.js, data/games.js, js/discord-post.js.
11. Extra interactive layer — the Cards tab: packs, a collection and a library of sets, built from the season files. ✅ See Cards.

---

Open Questions

• Does any of the six play the optional 2023 senior college season instead of declaring for the 2023 draft?
• College programs for all six — the ladder needs them from the 2020 season on.
• A school for Vitel — with no school he has no schedule, so the bot never posts him.
• Results and stat lines, once games are played. The fixtures are real; every game is still waiting on a result.
• Real rating numbers, in place of the generated placeholders.
• Sign-in emails for Parr, Vitel, and Stewart.
• Which attribute Block Power raises: run block power, pass block power, or both.
• Whether workout and overall changes persist per browser (localStorage) or in Supabase. Read-only season data stays in the data/*.js files either way.