Each bar = one day. Forge-orange highlights the peak day; cyan marks active days; faint bars are zero-commit days. The dashed cyan line is the 14-day mean. Hover any bar for that day's count.
The velocity curve is the public face of the engineering culture this codebase commits to (per principles).
Governance layer
7
Active RFCs
4
Published ADRs
5
Cadence rituals
28
Trust claims
Seven public RFCs at /rfc (versioning policy, anti-platitude, hum-score calibration, voltage coach, GPT-4o critic, cohort engagement, reproducibility audit). Four ADRs in docs/adr/ (factory + DI, strangler-fig, ring + dashboard, cmd-K palette). Five cadence rituals running on schedule per principles. Twenty-eight trust claims CI-enforced with historical flap-rate tracking at /admin/trust-claims-history.
Punch-list items shipped
No punch-list item closures in this window.
CI ratchet state
11 of 12 at zero · Tier-A 11/12
Each row is a CI gate that fails the build on regression. Ceiling: 0 means the floor — a single new occurrence anywhere in the codebase fails CI. Source-of-truth is the gate script referenced in each row; ceilings live in .github/.
Ratchet
What it counts
Ceiling
Closed at
Tier
Lighthouse mobile floors
Per-route mobile performance/a11y/SEO floors + FCP/LCP/CLS/TBT caps
0
B2201
A
as-cast count
All `as T` / `<T>x` type assertions in src/ (any type shape; const/never/literal/unknown excluded)
1334
—
A
Component LOC ratchet
Per-component line ceilings (SongDetail.tsx, dashboard/page.tsx)
0
B2189
A
Design tokens (hex literals)
Raw hex/rgb in JSX inside src/app + src/components
0
B2098
A
strict-tsc violations
Files exempt from --strict TypeScript
0
B1600
A
unchecked-index
Array/object indexing without bound checks
0
B1597
A
console.log calls
Direct console.log in production code (vs. structured logger)
0
B1586
A
Capability-surface gaps
Capabilities (genres, tradition presets) without a UI surface
b1909dc07-21Build 4408 (War Room II, flip-window cluster 2/2): every pre-registered prediction HIT — the corrected gate numbers, the vindicated arbiter, and a dead thread killed for $0.
1db090007-21Build 4402 (Master Punch List, Poncho): the gate arming enters the record — and this deploy activates SF_GAUNTLET_GATE_AB=1.
a51ba8707-21Build 4401 (the cost program, ruling 2 executed): the draft-score gauntlet gate ships dark — SF_GAUNTLET_GATE_AB arms the first live A/B in the experiment registry.
5ae98b807-21Build 4400 (Master Punch List, Poncho): the cost program's probe items close against B4399 and the production gauntlet-gate rollout files as MIXED.
e92897e07-20Build 4388 (Model-Lab, Rung 2 RESULT): KIMI misses strict promotion by half a point — CI lower bound -1.54 vs the pre-registered -1 gate — while GPT confirms a third time and GLM's parity dies under replication.
b27ab3007-20Build 4383 (Model-Lab): battle panel top-4 -> top-5 — the flagship run's top-4 by mean composite excludes KIMI (78.7, 0.3 behind opus), leaving criterion (d) undecidable for exactly the arm the shortlist gate exists to test.
038388d07-19Build 4365 (FORGE LAB): the operator-approved 8-arm field — Opus ceiling + GLM/GPT/Gemini via the Vercel AI Gateway.
bf513c207-17Build 4333 (Linear council, crown finding): the DB failure contract made real — the refund gate now keys on the failure signal that actually exists.
02351f207-17Build 4330 (SEO initiative, P1 complete): check:meta-lengths becomes gate #90 — metadata quality is now a ratchet, not a hope.
2b3ff5107-17Build 4328 (SEO initiative, linking cluster): hub unification, /lyrics de-orphaned, BOTH link-graph gates flipped to blocking — 89 gates.
029ffc107-16Build 4323 (shadow rollout, tranche 5): fifteen content/SEO/parity gates on the run() contract — 46/88 in-proc-ready, parity build #6.
8cafa3a07-16Build 4320 (quick-wins bundle): two ratchet tightenings, five agent model pins, the NUL-byte forensic fix + 31st gate migration — parity build #5.
3dc380c07-16Build 4319 (shadow rollout, tranche 4): ten more gates on the run() contract + guard retrofit on all earlier tranches — 30/88 in-proc-ready, parity build #4.
7d40e2307-16Build 4318 (shadow rollout, tranche 3): nine more gates on the run() contract + a never-wired ratchet discovered, operator-re-based, and ENFORCED — 88 gates, 20 in-proc-ready.
746377f07-16Build 4317 (shadow rollout, tranche 2): ten hygiene-ratchet gates migrated to the run() contract — 11/87 in-proc-ready, parity-green shadow build #2.
eb7696b07-23Build 4421 (Master Punch List): new skill songforge-forge-frontend — the /forge client/React conventions the 25-skill knowledge base never covered.
fae83a107-23Build 4420 (Master Punch List): recalibrate the prosody FATAL threshold — the 79%-FATAL /admin/prosody-report reading was a miscalibrated rule, not a catalog-wide prosody emergency.
d4ac88607-23Build 4418 (root-cause follow-up, B4417 forge incident): forge parse-rejects now leave a durable, queryable footprint — the failure rate is measurable in real time instead of blocked on a 30-day Vercel-log export.
ed1920f07-22Build 4417 (PROD INCIDENT — operator "getting this a lot"): the two forge-failure paths fixed — a false-positive regression removed, and the truncation class now auto-retries instead of making the user re-click.
a8e67b507-22Build 4414 (Distribution, Substack war room): the market-and-engage plan enters the record — "Measured Craft", one ICP, ~3 hrs/week, claims-cleared.
a6e347007-22Build 4413 (Master Punch List, Poncho): the autonomous stretch enters the record — three deep-validity items now wait only on the operator.
124a01007-22Build 4411 (War Room II follow-through): the judge-tier pin lands before anything can breach it, and the counterfactual bracket runner refuses to run unsigned.
4fe348f07-21Build 4410 (Master Punch List, Poncho): the nightly run enters the record — two closes, four partials, the flip-window cluster executed to its unblocked edge.
394570007-21Build 4409 (cadence, Trust Decay Audit #10 + War Room II standards rider): the lab claims enter the register, the stale flagship claim is struck, and the /vs claim goes on WATCH.
b1909dc07-21Build 4408 (War Room II, flip-window cluster 2/2): every pre-registered prediction HIT — the corrected gate numbers, the vindicated arbiter, and a dead thread killed for $0.
28fb03507-21Build 4407 (War Room II, flip-window cluster 1/2): the unit correction and the pre-registered flip rule — committed BEFORE the measurements run.
85ff4e007-21Build 4406 (Model-Lab, War Room II): the second war room enters the record — 64 agents, 20 rounds, the flip-window cluster and the deep validity pair.
86c718507-21Build 4405 (Model-Lab, D2b close-out): the critic-panel verdict enters the record — parity at worst, a modest edge at best, and it never lost.
a51ba8707-21Build 4401 (the cost program, ruling 2 executed): the draft-score gauntlet gate ships dark — SF_GAUNTLET_GATE_AB arms the first live A/B in the experiment registry.
5ae98b807-21Build 4400 (Master Punch List, Poncho): the cost program's probe items close against B4399 and the production gauntlet-gate rollout files as MIXED.
c2eded307-21Build 4399 (Model-Lab, the cost program): the sizing lands and both recommended probes execute — the writer was 12% of the bill, and the program found its halving anyway.
0f584f907-21Build 4398 (Model-Lab, D2 sizing instrument): the per-task cost decomposition lands on /admin/cost-per-song — the pipeline-level split the draft-swap prize is sized from.
dd426c207-21Build 4397 (Model-Lab, D2 RESULT): repair-closure PROVEN — chain output is draft-source-invariant (kimi paired diff literally +0.00 vs the all-Sonnet chain), and the mechanism is selection, not erasure.
3715b2507-21Build 4395 (Model-Lab, Rung 3 build 3 + Probe B close-out): the gauntlet-chain cell lands — cheap draft + wound-fed house repair, the D2 repair-closure engine — and the Probe B verdict enters the record.
340c4fa07-21Build 4394 (Model-Lab, Probe B hardening): working notes can never be scored as a song again — the scaffolding vocabulary learns the new generation, the lab gains a sectionless-output backstop, and the fallback flag detects the parse path instead of the preamble.
dbe700207-21Build 4393 (Model-Lab, SA#28 guard + the silent-neutral bonus run): prompt mode is now surfaced at the point of action — run header + start log announce PRODUCTION PROMPT vs neutral.
ebbea3907-21Build 4392 (Master Punch List, Poncho check-off): the SF_FORGE_PROVIDER seam row flips [~] — B4390+B4391 recorded as the shipped build half, routing promotion held open for Rung 5.
c935d2507-20Build 4390 (Model-Lab, Rung 3 build 1 + newcomer criterion d): the writer-override seam lands in the forge core — dark, attribution-safe, stamp-true — and DeepSeek V4 Pro clears the full screening rule.
06c4fd807-20Build 4389 (Model-Lab, newcomer screen RESULT): DeepSeek V4 Pro outscores the control at 1/24th the cost — the new savings frontrunner — Terra is a cost-parity quality candidate, Luna is eliminated.
e92897e07-20Build 4388 (Model-Lab, Rung 2 RESULT): KIMI misses strict promotion by half a point — CI lower bound -1.54 vs the pre-registered -1 gate — while GPT confirms a third time and GLM's parity dies under replication.
a144c0207-20Build 4387 (Master Punch List, Poncho check-off): the candidate-arm row flips [~] — B4386's three arms recorded as the built half, the screening run as the honest remainder.
816346407-20Build 4386 (Model-Lab, operator charge): three candidate arms land — DeepSeek V4 Pro, GPT Terra, GPT Luna — registry 8 -> 11, every price verified live, zero new keys.
a9d278e07-20Build 4385 (Model-Lab, Rung 2 prep): the harness gains the replicate-variance decomposition and the Rung-2 promotion rule — the 288-cell verdict becomes one command.
cd3430807-20Build 4384 (Model-Lab, Rung 1 criterion d): the 240-battle Elo record lands — KIMI K2.6 clears the FULL pre-registered shortlist rule (a-d), and pairwise puts both Claude arms last.
b27ab3007-20Build 4383 (Model-Lab): battle panel top-4 -> top-5 — the flagship run's top-4 by mean composite excludes KIMI (78.7, 0.3 behind opus), leaving criterion (d) undecidable for exactly the arm the shortlist gate exists to test.
23f575b07-20Build 4382 (Model-Lab, Rung 1 close-out): the accidental E1 replication enters the record — sigma_j brackets 1.15-1.50 — and Poncho books the replicate-item close.
ed63d9407-20Build 4379 (Model-Lab, Rung 1): the E1 judge noise-floor battery lands — sigma_j, paired-n MDD, and the perturbation probe, ready to fire the moment a valid local key exists.
3bf810907-20Build 4378 (Master Punch List, Poncho check-off): the deepseek re-pin P0 and haiku price P1 close against B4377; the unbuilt v4-pro candidate arm spins off as its own P3 row.
29cc5dc07-20Build 4377 (Model-Lab, post-run): deepseek arm re-pinned to explicit deepseek-v4-flash four days before the alias dies — plus the haiku/kimi price truth and the 24-seed run's analysis of record.
0bda3fe07-19Build 4375 (Model-Lab war room, Rung 0): the analysis harness + the SA#31 full-song Elo battle machinery land — the 192-cell export becomes analyzable in minutes, and shortlist criterion (d) gains its instrument.
905668d07-19Build 4374 (Model-Lab war room): the 20-agent strategy council lands — six-rung experiment ladder, both Song-Novel designs, the Rung-0 analysis plan, and the live price/identity snapshot that caught the DeepSeek alias deadline.
ca4494d07-19Build 4366 (MODEL TRUTH, operator-ruled): the signed seal reports the model that actually ran — six mis-stamps corrected, prose derives from the registry.
e9a9f1507-18Build 4359 (Master Punch List / DE-TELL defect): the cleanup prompts stop manufacturing the next cluster — the war room's last P1-auto row closes.
5fdc50207-18Build 4358 (Master Punch List / DE-TELL defect): the B3391 anti-rotation lesson finally reaches the model — class notes ride the banned-terms prompt.
7545ad307-18Build 4357 (Quality Council, cadence): the 2026-07-19 entry — first to carry the de-tell standing line; vault number honestly UNKNOWN.
d3745af07-18Build 4356 (DE-TELL war room, Phase 1b): banned words re-enter the cross-song loop + the lexicon-free recurrence signal — Phase 1 autonomous scope COMPLETE.
25e03e107-17Build 4341 (Linear council, performance trio): the ~1MB-per-forge vault re-fetch memoized; PostHog + CommandPalette code-split out of every content page.
168e5ff07-17Build 4339 (push-cost bundle, the fix): excluded paths are now EVICTED from the fingerprint's temp index — the stamp survives the bot rebase.
9eb2c6c07-17Build 4338 (Linear council, push-cost bundle): the ritual-stamp handshake — the pre-push hook stops re-running what the ritual just proved.
13b5fdc07-17Build 4337 (Linear council): the Database generic lands on both untyped Supabase clients — four caught lies fixed, the stale-types blocker surfaced honestly.
6329b7207-17Build 4336 (Linear council): the RepairFlow finalize deadlock retired at its MECHANISM + the Enter-key data loss closed — net -13 LOC on the god file.
44e635307-17Build 4335 (Linear council, quick wins 1-2): three strictness flags internalized into the compiler; the test-count floor restored from 24% stale.
bf513c207-17Build 4333 (Linear council, crown finding): the DB failure contract made real — the refund gate now keys on the failure signal that actually exists.
cd8d88c07-17Build 4329 (SEO initiative, authority/content cluster): the crown page becomes the extractable source; /crucible aims at the winnable SERP.
489b10c07-17Build 4324 (SEO initiative, phase 0): the eleven-lens SEO war room — evidence record persisted, 31 items filed, both operator actions executed and verified.
60deefa07-16Build 4321 (Master Punch List, P1): first-ever coverage for the finalize-song money path — 44 characterization tests + a live refund-window bug FOUND and filed.
3dc380c07-16Build 4319 (shadow rollout, tranche 4): ten more gates on the run() contract + guard retrofit on all earlier tranches — 30/88 in-proc-ready, parity build #4.
7d40e2307-16Build 4318 (shadow rollout, tranche 3): nine more gates on the run() contract + a never-wired ratchet discovered, operator-re-based, and ENFORCED — 88 gates, 20 in-proc-ready.
The 3-Year-Plan operating system. Four moats, four cadence rituals, ten trust-claim predicates. All public; all in the repo at docs/THREE-YEAR-PLAN.md.
Moat citations · per-build tagging
Data
61
Standards
506
Distribution
346
Velocity
296
Quality
1483
Cadence rituals · days since last entry
Quality Council6d / 7d thresholdcurrent
Trust Decay Audit4d / 22d thresholdcurrent
Bet Review9d / 60d thresholdcurrent
External Audit31d / 60d thresholdcurrent
36 public-claim predicates run as a CI ratchet via npm run audit:trust. Trust drift fails the build.