[ohld]

What

Experiment records are in one place now: experiments/YYYY-MM-DD-slug.md, one file per experiment, flat folder. Status lives in front matter (id, status, hypothesis, primary_metric, guardrails, unit, variants, started, check_on, decided, result, learning, code, readout), so files never move.

• 37 experiment files. 18 moved from active/ + completed/ (git mv, front matter added, bodies kept). New files for H2, H3, H3b, H3c, H4, H5, H7, H9, H10, H11, H13 from docs/growth/HYPOTHESES.md (section text moved verbatim; H1/H6/H8 appended to their existing files). Also new: channel_hits_v1, first_meme_nudge, meme_like_count (numbers from the 2026-10-30 closure), friend_challenge_h12 (killed: 0 participations), monthly_reminder_v1 (split out of the cadence holdout note), onboarding_default_language_v1 (fix(onboarding): every new user gets a meme on /start #385), raw_impr_rank_reaction_rate (feat(stats): rank memes within source by reaction rate, not views #384), phash_shadow_v1 (feat(dedup): perceptual hash (dHash) duplicate detection in shadow mode #381), ocr_dedup_watermark_fix (fix(dedup): strip watermark and require 0.6 similarity for OCR duplicates #380/fix(ocr): narrow watermark regex, unbias boolean prompt example #382 + restore).
• scripts/experiments.py (stdlib): list [--status], due, index (regenerates experiments/README.md: how to add/close + template, then Open with check dates and Decided with result + learning, newest first), readout <id> (prints the psql "$ANALYST_DATABASE_URL" -f ... command; does not run it).
• tests/test_experiments_ledger.py: every file parses with required keys, a valid status and ISO dates; README equals index output.
• docs/growth/HYPOTHESES.md and experiments/HYPOTHESES.md: reduced to a pointer / removed. Growth bar moved into the README header; the healthy-snapshot lines moved to docs/analyst/README.md. Links updated in AGENTS.md, SPEC.md, TODOS.md, docs/, specs/ and src comments. Historical readouts still link HYPOTHESES.md, which now points to the ledger.

Open points

• result: unknown where no readout exists: early_channel_popup (shipped with no readout) and crossposting_forward_intent_ranker (never built). Every running/proposed experiment is also unknown.
• phash_shadow_v1 has no check date or ship rule, so due flags it as NO DATE.
• monthly_reminder_v1 is proposed because MONTHLY_REMINDER_ENABLED defaults to false. Set it to running when the flag goes on.
• The friend challenge readout SQL may be deleted by the PR that removes the code. The record keeps the path for history.

Tests

pytest --noconftest tests/test_experiments_ledger.py: 2 passed. With a hand-edited README the test fails. ruff check passes on src/ tests/ scripts/.