COBI — Chancellor's Office Business Intelligence

The CPL Project Tracker, renamed. The masthead (cobi_brand.js) is a single row: the seal and the COBI wordmark with its Alpha chip, the site switcher, the "Where To?" search, and a utility cluster carrying your access and the ℹ About menu — with a quiet line beneath saying COBI is in alpha and what it answers from. (The Kobe homage — a rotating Mamba Mentality subtitle and an 8→24 wink — was retired 2026-06-22; the gold CPL superscript and the per-site org tag followed on 2026-09-04.)

A live dashboard, data pipeline, and curation workbench for the California Community Colleges Credit for Prior Learning Initiative — tracking statewide CPL adoption, articulation, and project work across all 116 California Community Colleges (CCC).

Live site: https://cpl-initiative.github.io/cpl-project-tracker/

Each morning the dashboard opens with First Light — a once-a-day greeting featuring a fresh public-domain California plein air painting each day (revealed grayscale → full color) with the artist's story, an optional read-aloud, and an anonymous "thought for the day" reflection box. Opt-out respected; reopen anytime via the "Today's painting" chip in the header.

The dashboard combines (a) project-management data for the CPL Initiative itself (workplan, budget, personnel, deliverables) with (b) live statewide CPL metrics scraped daily from the CCCCO MAP CPL Dashboard, plus (c) a synthetic knowledge-base layer that unifies course identities and CPL exhibits (credentials) across colleges so adoption can propagate from one college's articulation decision to the rest of the system.

---

What this project covers

The whole ecosystem, end-to-end:

1. The public dashboard (GitHub Pages) — KPI cards, project cards, CPL Analytics, plus a dedicated Exhibit Adoption & Credit Recommendations tab (moved out of CPL Analytics into its own top-level tab 2026-05-30) that groups cards by unified credential identity: each card shows the canonical title in bold + issuing-agency subtitle, a confidence / quality-flag badge when the underlying classification is shaky, and an "Also entered as N variants" disclosure that exposes every raw MAP title folded into the card (28% collapse vs raw IDs as of 2026-05-26) — now with Local + CCC versions merged into one card, per-course consolidated credit recs under a "Typical CPL" units headline, a master-detail Credential view (one card per credential, the CCC standard on top), and a seeker-framed Student view (pick your college → where you can get credit + the likely local course to ask about); plus Workplan Activities & Projects (Activity/sub-activity KPI cards + a work-item Projects grid — sub-activities render once, as Activity cards, and are immune to project Table/Archive; a ＋ Add-project flow creates new work items), Annual Goals (+ a Projects section listing the work items), Budget, Vision 2030, a per-college Common Course Reference curation tab (expand a course → an inverse view of the EACR: the aligned exhibits/credentials that articulate to it), a Common Subjects Reference tab (opening on its Progress dash since S425, a card per step with its procedure, the table as the Reference view; faculty-facing per-discipline curation with validate workflow + TOP/CTE/CIP columns + collapsible category groupings; a "CPL opportunities" column rolls up how many credentials articulate to each discipline; Variants column shows the raw local college subject codes colleges actually use, sourced from kb/coci_minted_memberships.json), a Common Exhibit Reference tab (sister to CCR — expand a credential for scope/CPL chips, a statewide-or-generated credit rec, green/orange adopted-vs-potential college badges, the college-entered exhibit-title variants, and the local courses with units that articulate to it, an Eligible (units) column (credit waiting to be unlocked = eligible − transcribed, rolled up from MAP's Exhibit CRs Catalog) and a Eligible students column, in three views: Needs you (the titles the nightly procedures run held for a person, the unclassified queue with each title's pre-filled proposal, credentials with no issuer, and merges to confirm), Procedures (each rule in plain words, who set it and what it settled) and Reference (every credential and the industry credentials); the tab opens on Needs you when the run held titles), a quick-start natural-language chat widget on the first screen (Claude API classifies your prompt to one of the 11 tabs and can pre-pop the destination tab's filters — "review unclassified credentials" lands you in Common Exhibit Reference with the unclassified-in-MAP queue already filtered; "apprenticeship initiative" pre-fills the Dashboard search; "subjects needing review" lands in Common Subjects Reference filtered to needs-review), a Letters tab embedding the budget-support letter curator (iframe to budget-support/web/curator.html, passcode-gated), Sierra AI (a conversational RAG assistant — ask CPL questions and get streamed answers drawn from the knowledge base, live dashboard metrics, and 2,300+ statewide exhibits; it talks to the shared cpl-chat Supabase Edge Function that also powers the live map.rccd.edu widget; she opens full screen over any tab — from Sierra Training, the Where To? box or a #chatbot link — and docks in My College, and she replaced the separate CPL Assistant tab on 2026-10-16), a Sierra Training tab (team-only: the assistant's improvement loop — the thumbs-up / thumbs-down feedback queue from both chat surfaces with a review status, a gap miner over the chat logs surfacing the questions the knowledge base had no good source for, and the plain-English instructions Sierra follows on every question — five numbers at the top that filter the lists they count, editable instructions, and a test question you can put to Sierra AI over the tab or in My College; round 1 of Sam's redesign, 2026-09-29), a ⚖️ Governance tab (team-only: the decision-rights register — who decides what, how far each input is trusted, and which cadences actually run; it measures itself rather than asserting compliance, so a loop that was decided but never run says so), a MAP Users tab (per-college MAP platform user counts + role mix publicly; for a signed-in reviewer, the staff roster and a ⚠ No student contact worklist — the colleges whose landing page has no Primary Contact, which is the address MAP routes a student's CPL request to, each with a proposed contact drawn from designations that college already made in MAP, and a Proposed because chip saying which designation it came from and why the cascade fell to it; a reviewer can override any row with a curator-set proposal, labeled with who set it and when, and the CSV export carries both layers separately so it doubles as the correction list to work through in MAP — nothing here writes to MAP, which has no write API, and these proposals are deliberately invisible to Sierra, who answers only from what MAP holds), a 🤝 Noncredit & Learning Partners tab (the noncredit / not-for-credit / adult-school / ROP / high-school-Cx / apprenticeship register — the six modes of Learning Partner CPL, a self-refreshing dormant-statewide-exhibit worklist, and an in-place ✎ Add insight affordance), a 🎓 CPL Pathways tab (audience-facing; a College filter, a Search box and an alphabetical Pathway list, and for a chosen college its My College sections; it also lists each harvested catalog record, Beta draft, laid out by requirement or by term apprenticeship-to-baccalaureate course maps with CPL check-offs derived live from the MAP articulation data — Cerritos's Field Ironworker Supervisor BS (31.5 units of journeyworker CPL) plus Foothill's Dental Hygiene & Respiratory Care BS views; violet ⊕ adoption-option chips show where OTHER colleges already articulate a credential, each with a ⚡ Quick Adopt request form; status stages (Discussion Draft / Active / Tabled) + a print-ready ⬇ PDF extract), an Implementation Funding tab (the DRAFT CPL funding model as a scenario tool: a selectable 2-year window, year-specific funding priorities, each card carrying its own Ed. Code §78093.2(d)(1) outcome — its letter, name, citation and the statute's own sentence — which is DERIVED from what the priority's metric measures rather than from its name, and which a curator can override on the card (Year 1 / Year 2 filter; every metric and description editable, draggable into a different order, with every section below the metric collapsing behind a summary that still carries its own figure; outcomes with no campus measure get their own reported cards, which carry designated activities and a share of the statewide project allocation — the order is stored as a permutation beside the config, so a priority's wording, share, funding factor and strategies all travel with it and no stored figure moves — and an optional Year 2 mirrors Year 1 switch), one funding total over all 118 institutions (the one-pool model, adopted 2026-08-31: $150K base to $400K cap on the combined credit + noncredit award; the three noncredit-only campuses — NOCE / SD Continuing Ed / Calbright — are ordinary rows qualifying by origination), one row per institution led by its minimum-conditions pie and Veteran Star and reading Max and Curr funding by lane and in total (the College Dashboard Sam locked on 2026-09-28: Curr is the funding qualifying so far, and no reserve figure shows), its max award decomposing into credit and noncredit shares, per-priority drill-ins opening on the Minimum Conditions line, district rollups, and one statewide row under a frozen header (measured pins, no lazy loading) — the table is the first section after the introduction, and every other section is folded on open (per visit, since 2026-09-02) — with the project card's named projects folded in live from the Budget table — and layered editing — anonymous edits are a private per-browser sandbox, while a signed-in reviewer saves the base model everyone opens to (Supabase-backed) — plus live per-priority actuals from the daily MAP pull (a measure that has yet to arrive reads TBA and counts $0, and the noncredit shares count $0 until their feeds report)), a TMC Builder tab (align a college's local courses to an ASCCC Transfer Model Curriculum / ADT — pick a college + a TMC and the right column auto-fills the local course already carrying each slot's C-ID; all 45 official TMCs, a GE Breadth companion for the full ADT, curator notes, and a per-college approved-ADT overlay sourced from the COCI program export so each TMC shows whether that college already has the ADT — ✓ Active (live in the catalog) / ✓ Approved (pending activation) / ⏳ in progress / ◐ teach-out — plus the statewide established-college count; the templates carry per-slot acceptance metadata toward an in-tool ADT review/compliance engine), a Knowledge Base tab (a sign-in-gated portal over the public CPL Knowledge Base — read the curated docs, and compose a new doc with optional file attachments + Claude polish, then commit it to GitHub as yourself), a CPL News tab (an auto-curated CPL news feed — California-first, then national, plus adjacent systems (Career Passport, CA Master Plan, workforce/upskilling) and CA budget items; harvested daily from free sources + a suggest-a-story queue and Claude-triaged, read live so it never goes stale), a My College tab (off the menu since 2026-10-22: its sections now open inside CPL Pathways' college view, and #college-briefing still opens it; open to colleges and the public since 2026-09-17, reading published copies with small student counts removed before publication, while the protected tables behind it keep their protection: one page, not 123 — pick a college and get where it stands against the goals, what its already-articulated-and-waiting credit actually consists of, its CPL funding (the seed grant, the implementation max award, the three minimum conditions with its own state, the funding its current outcomes demonstrate, and each priority's target), its tier with the missing criteria named, who MAP has on file for it, what occupations in its region it could already give credit for — matched against its own catalog, with the accuracy disclosed — and a Sierra AI box with questions computed from that college's own figures, which fills the screen while she answers), and a Pipeline progress board.
2. The data pipeline — daily GitHub Actions cron pulls fresh statewide metrics from the CCCCO MAP CPL Dashboard via a Cloudflare Worker proxy, then excel_to_dashboard.py regenerates the dashboard HTML, exports the client-side data files, builds Word reports, and pushes back to main so GitHub Pages re-publishes.
3. The knowledge base (kb/) — synthetic identity layers above the raw MAP exhibit + course data: unified credential titles (so "First Aid / AHA / American Heart Assoc. First Aid" all collapse to one card), unified course identities using a CCN-aligned surrogate scheme (CCN-ID > C-ID > M-ID), with a Phase B fold that consolidates M-IDs under official C-ID/CCN anchors when the underlying members agree, a statewide c-id.net articulation router (per-college official approvals place 8,377 member courses under their C-ID descriptor rows — no titles consulted), a description-evidence worklist lane that surfaces same-course candidates among the M-IDs no official source covers (TF-IDF over catalog descriptions, guard-suite gated, always curator-confirmed), and a title-evidence worklist lane that does the same across near-duplicate titles — including the ~54k single-college stand-alones — with no units gate, because externally standardized curricula (BAR smog, POST academy modules) pack the same spec at different unit loads per college. One deliberately strict exception acts on its own: the twin merge folds M-IDs whose titles contain the same words AND match on subject, discipline, credit, and units — every twin carries an alias receipt. The worklist's core same-title grouping is level-collapsing (Session 57) — "Beginning / Intermediate / Advanced X" surface as one family — so curators consolidate aggressively (Title 5 §55050 grants credit for similar learning), always one confirm at a time. A curated-anchor duplicates lane (2026-09-04) leads the queue: the May 2026 curated common courses that carry the same title and discipline as a catalog course, the catalog course proposed as the survivor, recomputed every build — never merged by a script. The rules in plain language: docs/ccr_rules_brief.md.
4. The Common Course Reference curation tab (formerly "Unified Courses" — renamed to avoid UC/University-of-California confusion and to signal that the tab is a reference faculty consult, not an authority replacing them). It opens on SkyView — the whole reference as one map, filling the window on its own (2026-09-05) — and since 2026-09-07 as the Sky: the reference as the night sky seen through a window, with the Globe one click away on the same canvas (the flat Map is still routable at #map, but its button left the row on 2026-09-08), Night by default and Day one control away, a slow turn that stops at the first touch, and drag and drop on the curve. On a phone (2026-09-09) the control row folds behind the word Controls and opens as a sheet at the bottom, the legend starts closed and the map takes 86% of the window; pinch zooms. COBI's header and side menu fold away and the ☰ at the map's top left brings the menu back, while the window controls at the row's right step between the map inside COBI, the map alone and the browser's full screen; the map's More menu reaches the comprehensive view, the Disciplines and subjects workspace, How SkyView works and the table, and #unified-courses/list opens the table directly. Two words next to Show: Courses | CPL (2026-09-10) swaps the map's universe — the courses, or the 1,987 curated credentials with the local MAP exhibits folded into each, the courses articulated to them (each a door back onto the Courses map) and a second ring on the 84 statewide ones; nothing moves on the CPL map — and Articulations lights what carries one; the coverage line says the universe's own counts and #skyview/cpl links to it. The search box also takes questions (2026-09-09): type a sentence rather than a keyword and Enter asks SkyView, which turns it into a selection on the map — the map answers by moving, and anything it cannot find on the map it says so rather than quietly leaving out. Its answer prints beside the search box, where you typed. Isolate draws only what you selected, a discipline included — authenticated reviewers (Supabase magic-link auth) curate disciplines, merge variants, and verify inferred values. Edits write to a live Supabase table and overlay the git-tracked KB. Eligible-units + Students columns + a 🎯 Cleanup-impact preset (Session 37) rank the cleanup queue by real CPL student-credit payoff (rolled up from the Common Exhibit Reference via the articulation crosswalk), not just a structural proxy. Session 38: the Subject column/filter shows the canonical SUBJ4 (raw local codes on hover), a surfaced affordance leads each row's actions cell, the expanded member table is , and Units shows a (e.g. ) with a ⚠ alarm when the spread exceeds 2.0 — a likely over-merge signal. the per-row ⚇ Merge dialog can now the surviving course (its Common SUBJ shows the new canonical immediately with a ⟲ "re-keys at the next fold" marker; a course with discipline now shows a blank Common SUBJ instead of a non-canonical local code, and the worklist has a lane that flags courses a title keyword likely mis-disciplined, each with a suggested fix — Session 113), and the ✨ Suggested-merges worklist gained , a global slider, (only the ★ target pre-checked), and a so word-order/suffix variants (Conversational ↔ Conversation) group together, with an amber flag on cross-discipline groups.
5. The row Trust-Card auditor (kb/_row_audit.py) — produces a per-row Trust Card for every M-ID + Cluster with a faculty_trust_score (the bar for cross-college articulation adoption) and an mc_ready_score (the destination — Model Curriculum readiness for ASCCC C-ID submission). 11 active rules cross-validate discipline assignments against title, TOP code, description, and member aggregation. The UCL surfaces findings as a "⚠ N · 0.XX" chip color-graded by severity, with a Triage: filter to carve the cleanup queue (8 modes including "3+ findings" for the high-confidence misassignment subset). Full decisions + lessons: docs/unified_courses_audit_lessons.md. See CLAUDE.md §11 for the M-ID → CIDx pipeline framing.
6. Custom Word reports — a per-college "[College Name] CPL Update" .docx generated on demand via a Claude API proxy (also a Cloudflare Worker endpoint), suitable for board / trustee distribution.
7. A public CPL Fact Sheet — a self-contained, shareable page at /fact-sheet/ that recreates the journalist Fact Sheet, pulls the headline KPIs live from live_metrics.json, and prints to a clean PDF. On screen it is on First Light (Playfair Display and Source Sans 3, and a dark mode that follows COBI's theme control, S352); print and the Word download keep Cambria and Calibri. It "sits alone" (no dashboard nav) so it can be shared publicly without exposing the internal tabs; a "📄 CPL Fact Sheet" link in the left nav rail opens it. It carries a rotating "My CPL Stories" section and is reviewer-Curate-editable in place (edit/add/reorder boxes + manage images; edits show for every visitor, the baked page is the fallback). The ✎ Curate button is hidden from the public — it appears for a signed-in reviewer (the COBI session is shared across browser tabs), or via the bookmarkable ?curate=1. That is presentation, not security: writes were always gated by row-level security, never by the button. The page is verified for WCAG AA contrast, keyboard access and phone widths — tests/factsheet_a11y.test.js in CI, plus fact-sheet/check_mobile_layout.js (headless Chromium, run on demand). Built Session 74 — see docs/fact_sheet_lessons.md.
8. A public funding-model explainer — the public view — a plain-language walk-through of the $35M Implementation Funding model at /funding-model/, written for a Chancellor's Office / Finance reader. It is live, not a snapshot: it loads cpl_funding.js and paints every figure from the same engine and the same Supabase config the Implementation Funding tab uses, so a dial change appears on the next reload with nothing to rebuild or republish. Its headline row sets the appropriation beside its two destinations, the institutions box split into credit and noncredit, and noncredit's share follows the tab's Credit and noncredit setting (blank keeps each institution's teaching mix); the words in every box and passage edit on the tab under The explainer's text, with Bold and Italic. Its Max award by institution section, directly after the introduction, is the tab's own institution table rendered in embed mode (window.CPL_FUNDING_EMBED = "college") — the same rows, drill-in, search and Excel export, not a copy — given the window's width rather than the prose column, because the prose cap is what pushed its last column off the edge. Funding outcomes and milestones follows: the four priorities and the milestone list together, open, with each priority's recommended strategies folded closed. The sections that explain the mechanics fold closed on open, each headed by its text title alone, and the titles are the model's own words (How an allocation is computed, Eligibility requirements). Every checkable claim — the priority descriptions, the baseline requirements, the participation deadline — is read from the model rather than typed beside it. Download PDF prints the page itself, so the file carries the model as it currently stands; there is no built artifact to go stale. It replaced a published artifact that had to be regenerated by hand and had silently drifted from the model it explains — see docs/kb-notes/methodology-a-snapshot-of-a-live-model-is-a-claim-that-decays.md.

---

Architecture

                CCCCO MAP CPL Dashboard (Azure)
                            │
                            ▼  REST /api/potential-savings
            Cloudflare Worker  (cpl-proxy.slee-548.workers.dev)
                            │
                            ▼  GET /scrape?secret=…
                    live_metrics.json
                            │
   CPL_Initiative_Project_List_v3.xlsx ──┐
   kb/ (knowledge base, curation)        │
                                         ▼
                  excel_to_dashboard.py  (Python pipeline)
                            │
       ┌────────────────────┼─────────────────────┬────────────────┐
       ▼                    ▼                     ▼                ▼
  CPL_Dashboard.html   CPL_Data.js / etc.   kpi_history.json    reports/*.docx
       │
       ▼  cp → index.html, commit, push
        GitHub Pages  (cpl-initiative.github.io/cpl-project-tracker/)


Supabase (hvuwhnbuahrtptokpqfh.supabase.co)
    ├── projects, workplan_goals   ← READ BY the pipeline (source of truth as of the
    │      Excel→Supabase migration; daily snapshot + Excel fallback). `projects` is
    │      the authoritative sub-activity TREE; `workplan_goals` is a by-id year-ladder
    │      OVERLAY (Path A, #909) — so the Annual Workplan Goals tab reflects EVERY
    │      Activities-tab project, X.Y.Z ids nesting under their X.Y parent. Inline-
    │      editable by allowed reviewers — Activity + sub-activity TITLES + brief
    │      DESCRIPTIONS on the Annual Workplan Goals tab (single-source editor, #902;
    │      `workplan_goals.description` col); blank-ladder rows are read-only on the
    │      year cells (title/description still edit `projects`)
    ├── budget_funding, budget_expenditures, personnel  ← also Supabase-read (Budget
    │      cutover, PR #189); inline-editable. Dashboard config (title/desc/KPI
    │      params) moved to committed kb/dashboard_config.json (Excel-retirement P2).
    │      The master .xlsx is no longer WRITTEN on any run; a few readers remain.
    │      2026-07-30: budget_funding is now the whole CPL LEDGER (45 rows) —
    │      section/parent_id/archived/description/window_label give it Sources,
    │      Uses, the combined $18M project pool and the 2017-forward history,
    │      rendered by budget_ledger.js with collapsible detail and inline
    │      editing on every non-total field. TOTALS SUM PARENT ROWS ONLY.
    └── kb_curation, allowed_reviewers   (UCL curation overlay)

Cloudflare Worker (cpl-proxy.slee-548.workers.dev)
    ├── GET  /scrape    → CCCCO MAP API → live_metrics.json
    └── POST /          → Anthropic API → Custom Report Generator

The worker calls the CCCCO Dashboard's REST API directly — no browser automation. That was a deliberate decision after Chrome-based scraping proved unreliable.

---

Daily pipeline

.github/workflows/daily-dashboard.yml runs on cron 17 10 * * * (≈10:17 UTC / 2:17 AM PT) and on manual dispatch. Steps:

1. Checkout main.
2. Fetch the CustomReport JSON (fetch_custom_report.py).
3. Scrape live metrics via the Cloudflare Worker (6 KPIs + tier breakdown).
4. Sync the curation overlay from Supabase (kb/_apply_curation.py) so the morning's reviewer edits land in kb/coci_curation.json.
5. Run excel_to_dashboard.py — regenerates the dashboard HTML in place, exports the client-side JS data files, builds Word reports, snapshots today's KPIs into kpi_history.json.
6. cp CPL_Dashboard.html index.html (Pages serves index.html).
7. Commit + push to main with a rebase-retry loop for concurrent pushes.

The commit always includes the regenerated dashboard, the Unified Courses data files, and exports/unified_courses.xlsx, so curation + the daily data refresh are captured atomically.

---

Repo orientation

.
├── CLAUDE.md                          ← project memory, rules, roadmap (read first)
├── README.md                          ← this file
├── excel_to_dashboard.py              ← main pipeline (reads Excel + live_metrics, generates HTML/JS/docx)
├── CPL_Dashboard.html                 ← generated dashboard (mirror of index.html)
├── index.html                         ← served by GitHub Pages
├── fact-sheet/                        ← public standalone CPL Fact Sheet (live KPIs + print-to-PDF; sits alone)
├── funding-model/                     ← public "How this funding model works" — LIVE off cpl_funding.js + the Supabase config (no snapshot to rebuild)
├── library/                           ← the public CPL Library: the pieces listed from COBI's Library tab (Malone's look, 2026-10-24)
│   └── cpl-opportunity/               ← one college's CPL Opportunity Analysis (?college=), moved here 2026-10-26; cpl-opportunity/ forwards
├── privacy.html                       ← public privacy page for the Library filer (its Google consent screen links it)
├── CPL_Data.js / statewide_data.js    ← client-side data for filters/search
├── statewide_prescriptive.js          ← EACR prescriptive layer (who could adopt → likely local course)
├── unified_courses*.js                ← Unified Courses tab data + lazy files
├── dashboard_filters.js               ← client-side filter/search/sort
├── kpi_reorder.js                     ← KPI card drag-to-reorder (per-browser, localStorage)
├── report_generator.js                ← Custom Report (Claude API via proxy)
├── college_report_generator.js        ← per-college .docx generator
├── docx.min.js                        ← local [email protected] UMD build (do NOT switch to CDN)
├── jszip.min.js                       ← local JSZip 3.10.1 (MIT), loaded on demand by the Library's remediation (do NOT switch to CDN)
├── library_remediate.js               ← the Library tab's in-browser remediation of a .pptx or .docx (S509)
├── cloudflare-worker-proxy.js         ← dual-purpose worker (scrape + Anthropic proxy)
├── live_metrics.json                  ← latest scrape
├── kpi_history.json                   ← daily KPI snapshots (must have no date gaps)
├── CPL_Initiative_Project_List_v3.xlsx ← master project data
├── reports/                           ← generated Word reports
├── exports/                           ← per-table .xlsx exports + unified_courses.xlsx
├── kb/                                ← knowledge base + curation + auditor
│   ├── README.md                      ← KB schemas + identity precedence + status
│   ├── _row_audit.py                  ← row Trust-Card auditor (see CLAUDE.md §11)
│   ├── _apply_curation.py             ← Supabase kb_curation → coci_curation.json sync
│   ├── _infer_disciplines*.py         ← re-runnable discipline inference passes
│   ├── coci_minted_*.json             ← minted M-ID identities + memberships + singletons
│   ├── coci_unified_courses.json      ← variant-unified clusters
│   ├── coci_articulations.json        ← earned articulations resolved to identity + credential
│   ├── coci_curation.json             ← human curation overlay (regen-safe)
│   ├── promotions.json                ← Phase A/B official-id consolidation manifest
│   ├── reference/                     ← read-only authorities (C-ID, CCN, MQ disciplines, COCI list)
│   ├── remint_out/                    ← re-mint dry-run artifacts + alias_map.json
│   └── row_audit/                     ← per-day Trust Card artifacts (latest.json + <date>.md)
├── docs/                              ← decision docs + lessons-learned (synced to Obsidian)
│   ├── INDEX.md                       ← landing page (generated block; kb/_build_docs_index.py)
│   ├── catalog/                       ← GENERATED per-lane doc catalogs — do not hand-edit
│   ├── coursecontrolnumber_remint.md  ← the 2026-05-22 M-ID re-mint playbook
│   ├── exhibit_unification_vision.md  ← credential-layer canonicalization design
│   └── reference/                     ← C-ID / CCN / TMC reference PDFs from ASCCC
└── .github/workflows/                 ← GitHub Actions (daily-dashboard.yml)

---

Local development

The full pipeline is Python + a few static assets — no build step.

# Regenerate the dashboard locally
python3 excel_to_dashboard.py

# Re-run the row Trust-Card auditor (read-only)
python3 kb/_row_audit.py

# Re-run a discipline inference pass (idempotent — only fills blanks)
python3 kb/_infer_disciplines.py
python3 kb/_infer_disciplines_from_desc.py
python3 kb/_infer_disciplines_from_top.py

excel_to_dashboard.py streams the 24 MB kb/reference/coci_course_list.xlsx with openpyxl read-only — never cat it.

---

Where to read more

• CLAUDE.md — project memory, critical rules, full pipeline reference, knowledge-base build status, M-ID lifecycle + MC vs TMC framing, roadmap, troubleshooting.
• kb/README.md — KB schemas, course-identifier precedence (CCN-ID > C-ID > M-ID), curation workflow, generator inventory.
• docs/coursecontrolnumber_remint.md — the playbook for safely re-keying shared-system identifiers (used for the 2026-05-22 re-mint; any future identifier re-key should follow the same pattern).
• docs/subj4_canonicalization_remint_lessons.md — Phase 1e decisions + lessons (the second re-mint, first under the revised Rule 7 staging framing; complete 2026-05-23 — 65k aliases applied, cleanup receipt zero).
• docs/common_subject_code_tab_lessons.md — Common Subject Code tab evolution (the four-PR series A → D, 2026-05-23). UX patterns, validate workflow design, TOP/CTE/CIP column rationale, render-refactor lessons.
• docs/exhibit_unification_vision.md — the credential-layer canonicalization design (synthetic unified-title layer over MAP's freehand exhibit titles).

---

Contributing

• Feature branches; open a PR to main. Claude Code sessions use claude/<short-description> branches automatically.
• The daily GitHub Actions cron pushes to main — coordinate around the 10:17 UTC window for any shared-system change (Supabase, alias maps).
• Never force-push main (GitHub Pages serves from it).
• Every PR runs three security scans (added 2026-05-23): Dependabot (pip + github-actions, weekly bumps), CodeQL (Python + JavaScript SAST, every push + PR + weekly), and TruffleHog (secret detection with live-verification). Findings show up under Security → Code scanning alerts. Dependabot opens PRs you can merge after CI greens.
• M-IDs are in staging-cleanup phase (per CLAUDE.md Rule 7, revised 2026-05-23). Re-mints are permitted in service of cleanup but must follow docs/coursecontrolnumber_remint.md — dry-run + alias map + Supabase fresh-read + atomic land + cron-window. Once the M-ID layer is faculty-published, Rule 7 re-locks to "stable identifiers, no renumbering."

---

License

© 2026 California Community Colleges Chancellor's Office (CCCCO). All rights reserved — no copying, redistribution, or derivative use without prior written permission (see LICENSE). This project tracks public statewide CPL data and is maintained by the CPL Initiative team at RCCD on behalf of the California Community Colleges system. The underlying public data and any separately-licensed component (e.g. the CPL Knowledge Base, CC BY 4.0) keep their own terms. Permission requests: [email protected].