Changelog

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.

the newsletter is assembled and claims-cleared — one publication-create away from launch.

the market-and-engage plan enters the record — "Measured Craft", one ICP, ~3 hrs/week, claims-cleared.

the autonomous stretch enters the record — three deep-validity items now wait only on the operator.

the judge-tier pin lands before anything can breach it, and the counterfactual bracket runner refuses to run unsigned.

the nightly run enters the record — two closes, four partials, the flip-window cluster executed to its unblocked edge.

the lab claims enter the register, the stale flagship claim is struck, and the /vs claim goes on WATCH.

every pre-registered prediction HIT — the corrected gate numbers, the vindicated arbiter, and a dead thread killed for $0.

the unit correction and the pre-registered flip rule — committed BEFORE the measurements run.

the second war room enters the record — 64 agents, 20 rounds, the flip-window cluster and the deep validity pair.

the critic-panel verdict enters the record — parity at worst, a modest edge at best, and it never lost.

the D2b build-half enters the record — the experiment run is the open remainder.

the cross-model critic panel enters the lab — one draft, two wound sources, a blind verdict per cell.

the gate arming enters the record — and this deploy activates SF_GAUNTLET_GATE_AB=1.

the cost program's probe items close against B4399 and the production gauntlet-gate rollout files as MIXED.

the sizing lands and both recommended probes execute — the writer was 12% of the bill, and the program found its halving anyway.

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.

D2b files — the operator's cross-model critic panel — and the gauntlet-chain build row closes against B4395.

the gauntlet-chain cell lands — cheap draft + wound-fed house repair, the D2 repair-closure engine — and the Probe B verdict enters the record.

prompt mode is now surfaced at the point of action — run header + start log announce PRODUCTION PROMPT vs neutral.

the SF_FORGE_PROVIDER seam row flips [~] — B4390+B4391 recorded as the shipped build half, routing promotion held open for Rung 5.

production-prompt cell mode lands — Probe B is now a clickable experiment.

the writer-override seam lands in the forge core — dark, attribution-safe, stamp-true — and DeepSeek V4 Pro clears the full screening rule.

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.

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.

the candidate-arm row flips [~] — B4386's three arms recorded as the built half, the screening run as the honest remainder.

the harness gains the replicate-variance decomposition and the Rung-2 promotion rule — the 288-cell verdict becomes one command.

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.

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.

the accidental E1 replication enters the record — sigma_j brackets 1.15-1.50 — and Poncho books the replicate-item close.

cross-family concordance rho=0.75, self-preference 0.07 — the ranking survives rival judges — and the lab gains replicates for the variance run.

sigma_j = 1.15 — the judge noise floor lands, and every Rung-0 significant gap survives it.

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.

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.

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.

the SA#31 pairwise-Elo row flips [~] — B4375's battle machinery recorded as the shipped Elo half, full-pipeline half explicitly open.

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.

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.

thinking off for kimi + one judge retry — acceptance criteria MET; the 24-seed field is GO.

the warm-instance cache clobber — found by the shakedown, fixed with the live incident pinned.

the full accounting — every seed, every lyric, every timing, one exportable artifact.

the best-of SoundCloud playlist joins /examples — "Hear Them Finished."

the multi-writer bake-off core — arms write, Claude judges, nothing touches user data.

the stamp skip-path gains the worktree/commit divergence guard.

the cleanup prompts stop manufacturing the next cluster — the war room's last P1-auto row closes.

the B3391 anti-rotation lesson finally reaches the model — class notes ride the banned-terms prompt.