SCRATCHPAD.md

SCRATCHPAD — cross-agent coordination

Living status file. ANY system (Claude session, Codex CLI, human) working in this repo: READ THIS FIRST, then update the relevant section when you change state. Keep it short. Timestamps in UTC.

Last updated: 2026-06-15 (Claude/Opus session) · Deadline: 2026-06-15 23:59 UTC (~12h)

---

ONE-LINE STATE

Importer overhaul + reconcile DONE and pushed to Burnmydays/hf- (main e0e2511). Codex parser fixed, ./sigrank --all added, instructions sharpened, wild corpus = 10 tokscale operators, Supabase migrated + board synced (11 rows live). Remaining: deploy to Space (secrets) + Codex handoff → SunrisesIllNeverSee + submission.

DO NOT TOUCH (frozen — these are the thesis)

• metrics.py MO§ES (ccusage) SEED row — canonical (Υ 18436.98). Wild corpus = 10 tokscale.ai operators (updated this session, OK to extend with citation).
• The Υ formula (C·O)/I² and the telescoping identity (o/i)(cw/o)(cr/cw)==cr/i.
• Codex estimation must stay flagged/estimated (*) — never a silent strict assumption. Two pathways live in _codex_input_estimate (ingest.py): Beta = output × real Claude io_ratio; Alpha = output × 2.0 (AA baseline).

RESERVED FOR CODEX (earns the $10k Codex-attribution prize — genuine Codex work only)

See CODEX.md. Still available as fresh Codex commits:

1. test_metrics.py — lock canonical numbers (Υ 18,437, lev 2042, X 3.31, telescoping for every SEED row).
2. Real Codex $/1M via OpenAI per-1M prices in parse_codex_submission meta.
3. The repo upload itself to github.com/SunrisesIllNeverSee.

   ALREADY DONE this session (Claude) — NO longer available as Codex commits:

   • Anchor refinement (2:1 → real Claude operating ratio, Alpha/Beta unified).
   • Board ~/* estimated-row marker. Integrity: only attribute commits Codex actually authored.

CANONICAL VERIFY (run after every change)

cd /Users/dericmchenry/Desktop/moses-sigrank
.venv/bin/python -c "import py_compile,glob; [py_compile.compile(f,doraise=True) for f in glob.glob('*.py')]"
.venv/bin/python metrics.py        # MO§ES must print: Y 18436.98, lev 2042.2, 10xDEV 3.31, $/1M 0.527

---

WORK LOG (newest first)

Devin session — 2026-06-15 (rarity system)

• [done] rarity_class(m) function added after classify() — maps velocity/leverage axes to four tiers: MYTHIC (v≥1, l≥100), EPIC (l≥10, v<1), RARE (v≥0.5), COMMON (fallback). Each tier returns (key, label, passive ability name, effect flavor text).
• [done] card_html() updated — outer .sig-card div gets rarity-{key} class, new .sig-card-rarity badge (COMMON/RARE/EPIC/MYTHIC), .sig-card-passive + .sig-card-effect blocks inserted between archetype and yield. All existing elements preserved.
• [done] board_html() updated — each row gets rarity-{key} class for colored left-border.
• [done] theme.py CSS — rarity color variables (--rarity-common/rare/epic/mythic), board row .mb-row.rarity-* left-borders, .sig-card hardcoded gold border removed (now per-rarity border + box-shadow), .sig-card-rarity badge styles, .sig-card-passive/.sig-card-effect text styles, .sig-card-name color changed to --moses-ink (rarity border carries signal), MYTHIC override for yield/name color to gold.
• [done] Verify green: py_compile all *.py OK · metrics.py canonical (Y 18436.98, lev 2042.2, 10xDEV 3.31, $/1M 0.527) · rarity assignments: MO§ES→MYTHIC, MapleEve/vincentkoc→EPIC, cexll/iamtheavoc1→RARE, Nepomuk5665/kzquandary→COMMON.

Devin session — 2026-06-15 (visual upgrade)

• [done] theme.py CSS upgrade — brighter default bar gradient (0.28→0.72), .mb-row.rank1 gold left-border + dominant #1 row, colored rank numbers (.mb-rank-1/2/3), hero upgrade (32px title, #moses-stat-strip), .mb-raw color #5f573f→#7a7060, .comp-bar composition bar, full .sig-card trading-card styles, and a @media (max-width:700px) mobile rule that hides velocity/leverage columns.
• [done] app.py UI upgrade — hero stat strip HTML; board_html() now adds rank1 class to row 1 and mb-rank-{1,2,3} spans for the top 3; new comp_bar_html(c) and card_html(name,m,rank, total_ops,narration_text) helpers (+ _first_sentence quote truncation); profile_md drops the text composition line and accepts an optional read so narrate() runs once; "Share card" section (gr.HTML) added to Measure-yourself tab; run_ingest now returns 4 outputs (profile, comp bar, card, board); footer simplified (formula soup removed — belongs in README "How the numbers work").
• [done] Verify green: py_compile all *.py OK · comp_bar/card/board helpers render · board contains rank1 + mb-rank-1/2/3 · run_ingest returns 4 values · narrate template fallback confirmed (no torch).

Devin session — 2026-06-15

• [done] GPU FIX — applied @GPU decorator to narrate() in narrate.py (line 60). ZeroGPU needs the inference function decorated or GPU alloc never happens and the model silently template-falls-back every call. _try_load() intentionally NOT decorated. No-op fallback path verified (no torch → template, import OK).
• [done] DEAD-CODE FIX — removed unused ingest() from ingest.py (old lines 103-112). Only ingest_meta() is called (app.py, sigrank.py). ingest_meta() unchanged.
• [done] README rewrite — kept YAML front matter + all math/formulas/SEED identical. Added Origin (word-based ranking → conservation law of commitment → token domain, with academic-links placeholder), "The full metric system" (cascade diagnostic, 4 archetypes from classify(), Scale V, species/quadrant framing — MO§ES = empty quadrant not a ladder), "Benchmark convergence" (AA 7:2:1 → 4.0 baseline = (7+1)/2, two-instrument validation). Restructured so Υ is the ranking metric within the broader system. Demo/Social placeholders untouched. Footer updated with origin.
• [done] Verify green: compile OK · python3 metrics.py → MO§ES Y 18436.98, lev 2042.2, 10xDEV 3.31, $/1M 0.527 (unchanged).

Claude/Opus session — 2026-06-15

• [done] Verified baseline green (compile + canonical metrics print correct).
• [done] FIX 6 — confirmed board CSS grid = 8 tracks, matches 8 header cells (theme.py:58). No code change needed; visual eyeball still owner's call.
• [done] FIX 1 — persistence-boundary honesty note added to README (Notes) + app.py Leaderboard tab + "your placement".
• [done] FIX 2 — cost-provenance note added to README (Notes) + metrics.py SEED comment.
• [done] FIX 4 — wrote species_cards.md (2×2 Scale×Amplification quadrant, MO§ES = empty quadrant, flavor-text fix, honest reuse≠proof line). Deck-only, not the app.
• [done] Supabase persistence build complete (table + seed + db.py + app wiring). See SUPABASE section.
• [done] Final verify green: compile OK · canonical metrics unchanged (Υ 18436.98) · db.py falls back to SEED with no env (7 ops, save no-op) · app.py imports + run_ingest end-to-end + dedup OK · narrate template fallback confirmed (no torch locally).
• NOTE: pre-existing Gradio 6.0 warning (css/theme moved to launch()) is handled by app.py's try/except on both Blocks() and launch(). Warning, not error.

---

SUPABASE — persistence build (decision + state)

• Project: AppFeeder betcyfbzsgusaghriptz · org MO§ES™ iwiixslkceolawfcbbhg · region us-west-1 · pg17 · ACTIVE_HEALTHY.
• Standing rule honored: NEW TABLE ONLY (public.sigrank_operators). 56 existing tables untouched.
• RLS decision: (A) curated — public READ, writes blocked for anon (service key only). Keeps FIX 1 honesty note valid (visitor pastes stay transient). Flip to (B) public-write later only with anti-spam.
• Table status: see WORK LOG / PERSISTENCE_SPEC.md STEP 1–2.
• Code: db.py (load_operators REST-read + save_operator service-key upsert) with SEED fallback safety net — app never breaks if Supabase is down/unconfigured.
• OWNER ACTION REQUIRED (cannot be done from here): set HF Space secrets (Space settings → Variables and secrets):

  SUPABASE_URL          = https://betcyfbzsgusaghriptz.supabase.co
  SUPABASE_ANON_KEY     = <anon / publishable key>     # READ
  SUPABASE_SERVICE_KEY  = <service_role key>           # WRITE (only if app writes)
  

  Until those are set, the app runs on the hardcoded SEED fallback (still works).

AUTO-INGEST — sigrank.py (local-first importer, this session)

• sigrank.py = local-first CLI. Auto-runs ccusage --json (or --codex/--file/stdin), computes via metrics.py, prints a sleek read + live board rank (db.load_operators → SEED fallback).
• No paste, no token, 100% local. The hosted Space paste box stays as the backup.
• --push is NOT yet implemented (would need service key); local read+compute only for now.
• Verified: compile OK; file/stdin/4-number paths produce correct profile (MO§ES Υ 18,437).

BLOCKED ON OWNER (not code)

• FIX 3 — tokscale "MCH (tokscale read)" comparison row: needs a clean re-pull; do NOT freeze the screenshot figure. Add as a LABELED instrument-comparison row.
• FIX 5 — install torch/transformers into .venv if you want REAL MiniCPM narration on camera (optional; numbers identical either way).
• Set the HF Space secrets above.
• Submission: move Space into build-small-hackathon org · record 60s video · social post · paste both links into README.