Trader Profitability Database

A research site that answers one question for speculative crypto venues: what percentage of wallets actually finished profitable?

Bloomberg terminal meets crypto culture — credible enough to cite, blunt enough to screenshot. Every percentage carries its numerator, denominator, source, query and methodology, and any figure that cannot be established reads INSUFFICIENT VERIFIABLE DATA rather than an estimate.

Core principle: you don't have to trust this website. You can check the calculation yourself.

---

⚠️ Read this before looking at any number

This repository ships in DEMO MODE. The dataset is synthetic — generated to exercise the interface. It is not a measurement of Pump.fun, Hyperliquid, Kalshi or any other named platform.

That is enforced, not just documented:

• A site-wide DEMO DATA banner and a DEMO badge on every card and figure.
• Demo evidence sources are provider: "DEMO", status: UNVERIFIED, url: null. No real Dune query, dataset or paper is ever cited for a synthetic number — a fabricated figure behind a plausible source link is the most damaging thing this codebase could ship, so a test asserts it cannot happen.
• Demo records score LOW on the evidence rubric, which is the honest result for a figure with no reproducible source.

Production mode (DATA_MODE=production) ignores the demo dataset entirely and serves only statistics that have been reproduced and approved in the database. With no approved records the site is empty — by design.

---

Interface update

The homepage starts with made-money / lost-money percentages and a short path into each platform's results and evidence. Advanced charts and metrics remain available in an expandable section on the detail page.

Public pages omit entries with neither a statistic nor a linked profitability report. This removes Rollbit, Stake, Shuffle, BC.Game, Drift, and dYdX from the demo's public catalogue. Their research records remain in the underlying data.

Fomo now has a report-only entry with public Dune links. Its result rows have not been reproduced here: there is no invented percentage, no synthetic Fomo statistic, and no Fomo entry in numerical rankings or comparisons. Existing example statistics remain explicitly labelled DEMO.

What it does

The database

Platform cards showing profitable %, unprofitable %, wallets analyzed, median PnL, evidence score and last-updated, filterable by category (ALL / PREDICTION MARKETS / MEMECOINS / PERPS / CASINOS / OTHER) and sortable without a page reload.

Platform detail — /platform/[slug]

Headline split, raw counts (analyzed / profitable / unprofitable / break-even / unknown), reality check, PnL distribution histogram, profitability by activity level, the survivors, the graveyard, profitability over time (7D–ALL), pain index with its full breakdown, algorithmic badges, methodological warnings, the evidence drawer, and a share card.

Verification — the point of the whole thing

• SHOW ME THE RECEIPTS — every source with provider, type, dataset date, sample, last-verified, reproducibility and status, plus the evidence score broken down line by line.
• CHECK THE MATH — numerator, denominator, the formula with the actual numbers substituted in, PnL definition, fee and unrealized treatment, exclusions, filters, date range, calculation version and the SQL.

Other routes

/rankings (six boards) · /compare (two-platform face-off) · /methodology · /sources · /admin (research queue) · /api/platforms

---

The rules the code enforces

These are the parts worth reviewing, because they are where a site like this usually goes wrong.

Percentages are never stored. They are computed from counts at render time, so the figure on the page and the figure in CHECK THE MATH cannot drift apart.

Break-even and unknown wallets stay in the denominator. Dropping them would raise the profitable share without any wallet changing outcome. A test pins this.

Wallets are not people. Every figure is phrased as a share of analyzed wallets. The sentence "XX% of traders lost money" does not appear, because one person can run many wallets and no clustering method is good enough to claim otherwise.

Centralized venues get UNKNOWN, not a guess. Casinos settle internally, so PROFITABILITY: UNKNOWN + DATA BLACK HOLE plus a note on exactly what is and is not public. House edge is never presented as observed trader profitability — it describes a game's expected margin, not what anyone finished with.

Nothing publishes itself. Adapters produce pending records. Only a signed POST to /api/admin/review with a named reviewer makes one public. Rejected records are kept, because a rejected calculation is part of the audit trail.

Statistics are never overwritten. Approval appends a snapshot in the same transaction, so a figure quoted six months ago stays inspectable.

Adapters fail loudly. A missing column, counts that don't sum, or absent credentials all return a failure with a reason. No adapter ever infers which column meant "profitable" — a mis-mapped column produces a plausible number with no relationship to reality.

Polymarket is excluded everywhere, per the specification. A test asserts it.

---

Running it

npm install
cp .env.example .env    # works as-is: demo mode, no services needed
npm run dev             # http://localhost:3000

Production mode

psql "$DATABASE_URL" -f db/schema.sql
export DATABASE_URL=postgres://...
export DATA_MODE=production
export ADMIN_TOKEN=$(openssl rand -hex 32)
npm run build && npm start

The site now serves only approved records. Approve one with:

curl -X POST "$SITE_URL/api/admin/review" \
  -H "Authorization: Bearer $ADMIN_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"statId":"<uuid>","decision":"approve","reviewer":"your-name",
       "note":"Re-ran the query, counts match within 0.1%."}'

Deploy target is Vercel (vercel.json included); any Node host works.

Deploy the website to Vercel

1. In Vercel, import gowthamaran/Profitable-traders and select its default branch, claude/build-deploy-github-cfmglm.
2. Keep the repository root and the detected Next.js preset. Use project name profitable-traders and set DATA_MODE=demo for the initial preview.
3. Deploy. No database or API keys are needed to explore the demo.

Demo pages are marked noindex to keep synthetic figures out of search results. Page metadata uses Vercel's deployment domain automatically; set SITE_URL only when you have a custom domain. To publish real figures, configure Postgres, apply the schema, populate and review evidence, then set DATA_MODE=production. Deploying the code does not collect or verify profitability data automatically.

Environment

Variable	Purpose
DATA_MODE	demo or production. Defaults by whether DATABASE_URL is set.
DATABASE_URL	Postgres / Supabase connection string.
ADMIN_TOKEN	32+ chars. Absent ⇒ the review endpoint is disabled.
DUNE_API_KEY	Optional. Absent ⇒ the adapter reports it cannot run.
FLIPSIDE_API_KEY	Optional, used as the independent second source.
SITE_URL	Canonical URL for metadata and share links.

---

Layout

src/
  app/            routes: home, platform/[slug], compare, rankings,
                  methodology, sources, admin, api/
  components/     cards, charts, evidence drawer, share card, primitives
  lib/
    types.ts      a percentage cannot exist without its evidence
    metrics/      profitability · evidence · painIndex · badges · microcopy
    data/         repository · mapping · demo dataset · mode
    adapters/     dune · flipside · defillama · hyperliquid
db/schema.sql     counts CHECK, approval gate, append-only snapshots
tests/            88 tests, no network

Development

npm run typecheck && npm run lint && npm test && npm run build

The suite covers the arithmetic (denominators, bucket classification, percentage identities), the evidence rubric and its grade bands, the pain index including weight redistribution, every badge rule, the microcopy triggers, the demo dataset's internal consistency and its labelling, adapter failure modes, and row mapping's rejection of inconsistent records.

Humour policy

Roughly one playful element per screen, always as secondary copy under the statistic, and never on /methodology or a data warning. Nothing jokes about individuals, addiction, or anyone's real losses — the target is speculative markets in aggregate. Where humour and credibility conflict, credibility wins.

Not advice

Historical observation of wallet-level outcomes. Nothing here forecasts any future result.

License

MIT — see LICENSE.