operations · the gate

Every verdict

260 changes judged by the constitutional panel before they could merge, newest first. Open one for every seat's vote, the rules it cited, and its reasoning in full, verbatim from the record.

• 2026-10-03passed#458The rule for a claim with no evidence, tested twice: it works, and no draft is adopted yet5 comply
• 2026-10-03passed#452Cast, Not Carved: the record the panels tripped on is refused, the case re-told, and a fresh panel judges it5 comply
• 2026-09-30passed#456The founder's two answers, the count behind leaving two verdict words where they are, and the next experiment's design5 comply
• 2026-09-30passed#455Corrections to #453: the direction of a split, the strip's date, the rule on stamps (review note #454)5 comply
• 2026-09-30passed#453Where we are, as of 2026-09-30: the status documents, two mends to the loop, a finding and a correction4 comply · 1 violates (GPT-5.6 Sol (OpenAI))
• 2026-09-30passed#451Before Sputnik, re-told within the reader's budget; a fresh panel contests one claim the case rests on5 comply
• 2026-09-30passed#450The Religion with No Name, re-told within the reader's budget; a fresh panel contests the case verdict5 comply
• 2026-09-30passed#449A slow test is given room: the forced-choice test loads every case several times5 comply
• 2026-09-30passed#446The reply comparison is tested against every place a reply can hold text (review note #444)5 comply
• 2026-09-30passed#445A fresh blind check ratifies State, Not Scar; Cast, Not Carved's panel is held back5 comply
• 2026-09-30passed#443Collapse, Not Computation, re-told within the reader's budget; a panel of four ratifies it5 comply
• 2026-09-30passed#442The Emptied Amazon, re-told within the reader's budget and re-graded; a fresh panel ratifies it4 comply · 1 violates (GPT-5.6 Sol (OpenAI))
• 2026-09-30passed#441The Aeon Before Ours, reconsidered and re-told within the reader's budget; a fresh panel ratifies it4 comply · 1 violates (GPT-5.6 Sol (OpenAI))
• 2026-09-30passed#440A reply's research-status entries are traced to the research file before it is called carried (review note #438)4 comply · 1 violates (GPT-5.6 Sol (OpenAI))
• 2026-09-30passed#439Deep Memory, reconsidered and re-told within the reader's budget; a fresh panel still contests it5 comply
• 2026-09-30passed#437The panel is told what a drafter's replies hold that the edition does not4 comply · 1 violates (GPT-5.6 Sol (OpenAI))
• 2026-09-30passed#436A stamp can add to what a draft was shown and never take from it (review note #434)5 comply
• 2026-09-30passed#435The verb closes a claim span the drafter closed with the wrong bracket, and says so5 comply
• 2026-09-30passed#433A judgment written with the panel's verdicts in hand cannot be ratified by them, contested or not4 comply · 1 violates (GPT-5.6 Sol (OpenAI))
• 2026-09-30passed#432Experiment, second arm and report: written definitions of the verdict words are not adopted5 comply
• 2026-09-30passed#431Two cases re-told within the reader's budget: State, Not Scar and Cast, Not Carved4 comply · 1 unsure (GPT-5.6 Sol (OpenAI))
• 2026-09-30passed#430A run record is read wherever it is filed, and a check's record says what it was sent (review note #428)5 comply
• 2026-09-30passed#429A component is its label: the verb's account of what an edition changed keeps each component's identity (review note #427)5 comply
• 2026-09-30passed#426Experiment: do written definitions of the verdict words make a seat agree with itself? Design and first arm4 comply · 1 violates (GPT-5.6 Sol (OpenAI))
• 2026-09-30passed#425The test of better: an edition is held to the reader's budget and read beside the one it would replace4 comply · 1 violates (GPT-5.6 Sol (OpenAI))
• 2026-09-30passed#424"Carried whole" says what it did not compare (the Anthropic seat's note on #422)5 comply
• 2026-09-30passed#423The build reads the content once; two notes on the reading layer5 comply
• 2026-09-30passed#422The panel reads first what it can read nowhere else; and one check is one measurement (review note #420)5 comply
• 2026-09-30passed#421The reading layer: the case page is for reading the case, and the record is a page of its own5 comply
• 2026-09-30passed#419The Anthropic seat is given room to answer, and a failed seat's cost is recorded4 comply · 1 violates (GPT-5.6 Sol (OpenAI))
• 2026-09-30passed#418Chain: 2026-09-284 comply · 1 unsure (GPT-5.6 Sol (OpenAI))
• 2026-09-30passed#417One failing seat no longer stalls the loop: the Anthropic seat streams, a panel is stale only when it cannot speak, and a failed sitting says so5 comply
• 2026-09-23passed#414A file the diff cannot carry is described to the panel, not merely named5 comply
• 2026-09-23passed#413The house model and the Anthropic seat move to Opus 5.55 comply
• 2026-09-23passed#412Vasocomputation: C001 split into C190 and C191, and the edition grades each on its own records4 comply · 1 unsure (GPT-5.6 Sol (OpenAI))
• 2026-09-22passed#411A dated exemption for the rest of September; an exemption may span days5 comply
• 2026-09-21passed#409The planner's ordering tests pick cases without pending records5 comply
• 2026-09-21passed#408Chain: 2026-09-214 comply · 1 unsure (GPT-5.6 Sol (OpenAI))
• 2026-09-21passed#407§3.2 in doubt: a claim a seat calls compound and the reader calls one proposition is split when two parts are affirmed5 comply
• 2026-09-21passed#406A record that stands against an objection carries the reader's reason in the account5 comply
• 2026-09-21passed#404The anchorless split gets its second round5 comply
• 2026-09-21passed#403A claim anchored by evidence alone is still judged one proposition or not5 comply
• 2026-09-21passed#372Chain: 2026-09-204 comply · 1 violates (GPT-5.6 Sol (OpenAI))
• 2026-09-21passed#366Deep Memory: edition under protocol v11 (the seed page as a source, not a narrator)5 comply
• 2026-09-20passed#402The evidence-anchored shortcut holds only for a ledger claim unchanged, and a frozen agenda item is left as it stands5 comply
• 2026-09-20passed#400A claim anchored by the ledger's evidence is anchored; an agenda item retires rather than stand with no claim4 comply · 1 violates (GPT-5.6 Sol (OpenAI))
• 2026-09-20passed#397A split stays inside the compound, and a part is placed where its quote is5 comply
• 2026-09-20passed#396The packet carries declined candidates compact5 comply
• 2026-09-20passed#395The split recurses once for a part refused as compound5 comply
• 2026-09-20passed#394The second client is named by itself5 comply
• 2026-09-20passed#392A second client for a page that challenged the first4 comply · 1 violates (GPT-5.6 Sol (OpenAI))
• 2026-09-20passed#390An answer that relinks reads what it relinked, against the parts, before the edition5 comply
• 2026-09-20passed#389The ledger writers never alias, and an in-place rewrite expands the aliases it would leave dangling5 comply
• 2026-09-20passed#388Frozen records may reference a claim rejected since; live ones may not5 comply
• 2026-09-20passed#387The splitter anchors each part; a compound claim is replaced by anchored parts or refused, never held5 comply
• 2026-09-20passed#385A held claim and a relinked record carry no weight until read4 comply · 1 violates (GPT-5.6 Sol (OpenAI))
• 2026-09-20passed#383A re-reading must not leave the case worse: held compound claims, relinking to parts, rollback4 comply · 1 violates (GPT-5.6 Sol (OpenAI))
• 2026-09-20passed#382The answer step reads a seat's bare ids and ranges as the case's own5 comply
• 2026-09-20passed#381Retrieval reads PubMed Central through Europe PMC5 comply
• 2026-09-20passed#380The run account names every protected section class5 comply
• 2026-09-20passed#376The run account puts the head first and digests the records a change adds4 comply · 1 violates (GPT-5.6 Sol (OpenAI))
• 2026-09-20passed#374The run account carries the assessment's whole judgment5 comply
• 2026-09-20passed#370The answer step: a parked or noted sitting is put back to the verbs5 comply
• 2026-09-20passed#367The leads pass: what a pass named and could not open is asked of the indexes5 comply
• 2026-09-19passed#365Chain: 2026-09-194 comply · 1 unsure (GPT-5.6 Sol (OpenAI))
• 2026-09-19passed#364The agenda's lifecycle fields do not move the ledger5 comply
• 2026-09-19passed#363Chain: 2026-09-195 comply
• 2026-09-19passed#361The founding inputs are a seed, not a protagonist (edition protocol v11)4 comply · 1 unsure (Grok 4.5 (xAI))
• 2026-09-19passed#359Evidence is cited for the claim it is attached to (edition protocol v10)5 comply
• 2026-09-19passed#357Chain: 2026-09-194 comply · 1 unsure (GPT-5.6 Sol (OpenAI))
• 2026-09-19passed#356The budget window re-judges with the workflow's own token5 comply
• 2026-09-19passed#353Chain: 2026-09-17 (interrupted — no account of the sitting)5 comply
• 2026-09-19passed#347Pre-Columbian Amazon: the dating and the builders are two claims (review note #345)5 comply
• 2026-09-17passed#355Refused evidence is a tombstone on the ledger page5 comply
• 2026-09-17passed#354A split part's sibling never takes the part's own id5 comply
• 2026-09-17passed#350A seat whose reply will not read as JSON is asked once more5 comply
• 2026-09-17passed#348What verification blocked is proposed again, by the reverify verb5 comply
• 2026-09-17passed#346A dropped place goes with its marker (narrowLocator, review note #341)5 comply
• 2026-09-17passed#343A resolved lead says what agreed, and no more (review note #339)5 comply
• 2026-09-17passed#342An Internet Archive item page is a viewer, not the text5 comply
• 2026-09-17passed#340Chain: 2026-09-174 comply · 1 violates (GPT-5.6 Sol (OpenAI))
• 2026-09-17passed#338Leads become documents: the report names what it did not open, the draft pins it through the open indexes4 comply · 1 violates (GPT-5.6 Sol (OpenAI))
• 2026-09-17passed#337Chain: 2026-09-175 comply
• 2026-09-17passed#335Retrieval follows Internet Archive items to their OCR text5 comply
• 2026-09-17passed#334Decisions log: the chronology of the §4 amendment, set straight (review note #333)5 comply
• 2026-09-17parked#332Founder amendment, AGENTS.md §4: the provisional tier5 violate
• 2026-09-17passed#331Provisional admission and the research lifecycle, tightened on two more notes (#329, #330)4 comply · 1 violates (GPT-5.6 Sol (OpenAI))
• 2026-09-17passed#328The research agenda gets a lifecycle, a plan gate, and revision by default4 comply · 1 violates (GPT-5.6 Sol (OpenAI))
• 2026-09-17passed#324The goal, reworded: more AI means a better site, never a worse one5 comply
• 2026-09-17passed#323Re-verification: provisional records are settled when their texts can be read4 comply · 1 violates (GPT-5.6 Sol (OpenAI))
• 2026-09-17passed#322Provisional admission: records whose source exists but whose text could not be read enter unread, labelled, and weightless5 comply
• 2026-09-17passed#321Plan: three search improvements entered as todos5 comply
• 2026-09-16passed#320A spend file is named by the run id as it is; an id that is not a file name is refused (review note #316)5 comply
• 2026-09-16passed#317A claim held on its anchor alone says so on the page5 comply
• 2026-09-16passed#314Edition protocol v8: anchored is not evidenced5 comply
• 2026-09-16passed#313Budget-parked PRs are re-judged when the window has room5 comply
• 2026-09-16passed#312Verify: a quote missed in a substitute copy is blocked, not failed5 comply
• 2026-09-16passed#311One spend file per run: overlapping sittings no longer conflict on the ledger4 comply · 1 violates (GPT-5.6 Sol (OpenAI))
• 2026-09-16passed#310The loop's goal written down; two founder hand merges declared supervised5 comply
• 2026-09-16passed#309A dispatcher not given is recorded as not given (review note #308)4 comply · 1 unsure (Gemini 3.8 Flash (Google))
• 2026-09-16passed#307The founder's door is the founder's: checked inputs and a recorded dispatcher (review note #290)4 comply · 1 violates (GPT-5.6 Sol (OpenAI))
• 2026-09-16passed#306The verification report says what the writer did (review note #304, code half)4 comply · 1 unsure (Gemini 3.8 Flash (Google))
• 2026-09-16parked#305Provenance stamps stay only where they hold (review note #302)5 comply
• 2026-09-16parked#303Chain: 2026-09-164 comply · 1 violates (GPT-5.6 Sol (OpenAI))
• 2026-09-16passed#301One bad reply costs one record: the code #299 described, with its test5 comply
• 2026-09-16passed#300By-hand amendments carry the full stamp (edition amendments, changelog runId/promptVersion)5 comply
• 2026-09-16passed#299One bad reply costs one record, not the sitting5 comply
• 2026-09-16passed#298The last two edition fixtures post-date any real incumbent too5 comply
• 2026-09-16passed#296Chain: 2026-09-16 (interrupted during a later step — partial account)4 comply · 1 unsure (GPT-5.6 Sol (OpenAI))
• 2026-09-16passed#295Counts are measured, not estimated: the edition verb appends its own figures to every rationale5 comply
• 2026-09-16passed#293Edition fixtures post-date any real incumbent5 comply
• 2026-09-16passed#291Chain: 2026-09-164 comply · 1 violates (GPT-5.6 Sol (OpenAI))
• 2026-09-16passed#289The founder's door on a sitting: a forced first choice (case and verb), then the ledger's own4 comply · 1 violates (GPT-5.6 Sol (OpenAI))
• 2026-09-14passed#288Orch OR: the intake entry's drafter line annotated — the model asked, the served model unrecorded4 comply · 1 unsure (Gemini 3.8 Flash (Google))
• 2026-09-14passed#287Chain: 2026-09-145 comply
• 2026-09-11passed#286No dangling ids, everywhere: 32 legacy prose references repaired, the cutoff removed, verify scrubs its own notes5 comply
• 2026-09-11passed#284Orch OR: two claim statements made atomic under the rule that landed after them5 comply
• 2026-09-11passed#283What the first Orch OR sitting taught: five fixes, shifted left4 comply · 1 violates (GPT-5.6 Sol (OpenAI))
• 2026-09-11passed#281Chain 2026-09-11, Orch OR: the first research pass, nine papers read in, a new edition — dispatched by hand4 comply · 1 violates (GPT-5.6 Sol (OpenAI))
• 2026-09-11passed#280The scheduler test stops depending on a case without dispositions5 comply
• 2026-09-11passed#279The stamp reads the fallback: a run records the model that answered, and pays each model its own tariff5 comply
• 2026-09-10passed#278Three review notes answered: a question mark is part of a quote, a bearing fails closed, a correction line carries the whole value, an intake entry labels the drafter's rationale4 comply · 1 unsure (GPT-5.6 Sol (OpenAI))
• 2026-09-10passed#274Chain 2026-09-10, the cosmology case: a research pass, a fifth reanalysis read in, a new edition, a blind check, with the two corrections the panel asked for4 comply · 1 violates (GPT-5.6 Sol (OpenAI))
• 2026-09-10passed#273Chain 2026-09-10, The Emptied Amazon: the first research pass, Peripato 2023 read in, a new edition, a blind check4 comply · 1 violates (GPT-5.6 Sol (OpenAI))
• 2026-09-10passed#272Chain 2026-09-10, Before Sputnik: Villarroel's post taken in, its two preprints read into the ledger, a new edition, a blind check4 comply · 1 unsure (GPT-5.6 Sol (OpenAI))
• 2026-09-10passed#271Verify v6: a bearing per claim and a split by direction; edge punctuation is not part of a quote; a read source is ai_verified; a sitting works on at most two cases4 comply · 1 violates (GPT-5.6 Sol (OpenAI))
• 2026-09-10passed#270The CLI reads --deadline-minutes and --out: a completed sitting had opened its PR as "interrupted" because its account was never written5 comply
• 2026-09-10parked#269Chain: 2026-09-10 — transients (inbox: draft, verify, edition; check), pre-columbian-amazon (report chain; check), ccc (report chain; check)1 complies · 1 violates (GPT-5.6 Sol (OpenAI)) · 3 unsure
• 2026-09-10passed#268A sitting knows its deadline and writes its progress, and its PR opens whatever became of it5 comply
• 2026-09-10passed#267What the first third-party drop exposed: the verifier reads the arXiv paper the drafter read, reasons show the whole span, the reader has room to think, a research item states no unsourced finding5 comply
• 2026-09-10parked#266Chain: 2026-09-103 comply · 2 violate (Opus 5 (Anthropic), GPT-5.6 Sol (OpenAI))
• 2026-09-10passed#265Run links completed on the sitting's records: the edition and the five check records name the run that wrote them, and the run records list what they wrote5 comply
• 2026-09-10passed#264The gate declares the nine supervised landings by hash and keeps the loop's merges in the count: epoch 5 withdrawn5 comply
• 2026-09-10passed#260Gate epoch 5: the construction of 2026-09-09/10 is declared supervised, and the autonomous lane's window starts clear4 comply · 1 violates (GPT-5.6 Sol (OpenAI))
• 2026-09-10passed#259The caching decision names the channel of the founder's direction: the working session, by its id5 comply
• 2026-09-10passed#258Records name the run that wrote them, and runs list what they wrote5 comply
• 2026-09-10passed#257Chain: 2026-09-104 comply · 1 violates (GPT-5.6 Sol (OpenAI))
• 2026-09-10passed#255Prompt caching by rule: nothing automatic on one-shot calls, a marked prefix wherever calls share one4 comply · 1 violates (GPT-5.6 Sol (OpenAI))
• 2026-09-09passed#254The reader's acts are structured provenance: each change it makes at intake is stamped on the record with model, run id, prompt version and date5 comply
• 2026-09-09passed#252Verify v5: the second reader names the direction it finds and the claims the passage bears on, and verify writes them4 comply · 1 violates (GPT-5.6 Sol (OpenAI))
• 2026-09-09passed#250Five evidence records say what the second reader said, and the record names its references: review notes #248 and #247 answered4 comply · 1 violates (GPT-5.6 Sol (OpenAI))
• 2026-09-09passed#249The Deep Memory cover is the founder's lab-site picture, brought over unchanged with its record5 comply
• 2026-09-09passed#246The caption record says exactly what changed, no second drop, and the opening tests let a case grow out of its opening4 comply · 1 violates (GPT-5.6 Sol (OpenAI))
• 2026-09-09passed#245Chain: 2026-09-094 comply · 1 violates (GPT-5.6 Sol (OpenAI))
• 2026-09-09passed#243Plate captions say what the photograph shows and what the file calls it, nothing the photograph does not show: review note #242 answered4 comply · 1 violates (GPT-5.6 Sol (OpenAI))
• 2026-09-09passed#241The image tooling returns and Deep Memory gets its cover and twelve plates4 comply · 1 violates (GPT-5.6 Sol (OpenAI))
• 2026-09-09passed#240The Deep Memory drop, on the founder's grant in his words: the home page taken in as the founding narrative5 comply
• 2026-09-09passed#239Tests get twenty seconds, not five: the CI runner took 5.9 s to assemble an edition over every case and failed a green change5 comply
• 2026-09-09passed#238The opening says the founder's pages are to enter by his drop, not that they have: a second question-only edition answers review note #2375 comply
• 2026-09-09passed#236Deep Memory opens question-only, and the founder's pages come in through the inbox: the genesis test begins4 comply · 1 violates (GPT-5.6 Sol (OpenAI))
• 2026-09-09passed#235The inbox reads HTML pages: tags stripped, headings kept as section markers, the statement in a sidecar as a PDF's is5 comply
• 2026-09-09passed#233The conjecture cards retire into the changelog: a bet is a claim, and the founder's stakes are banked where §7 puts them4 comply · 1 violates (GPT-5.6 Sol (OpenAI))
• 2026-09-09passed#232A reply is the answer only when it says so; the spend copy is date-bounded5 comply
• 2026-09-09passed#228The UI pass, third PR: one operations page — the state, the schedule, what is next, spend against the caps, recent sittings, the review notes4 comply · 1 violates (GPT-5.6 Sol (OpenAI))
• 2026-09-09passed#227The UI pass, second PR: the record layer — every sitting on the case, what it admitted, what it refused and why4 comply · 1 unsure (Gemini 3.8 Flash (Google))
• 2026-09-09passed#225The UI pass, first PR: what happened lately on top, the judgment and the panel in one block under one rule, the accounts side by side5 comply
• 2026-09-09passed#223Subtraction, fourth set: the Arbiter's seats through the metered transport; the panel's cost in the ledger5 comply
• 2026-09-09passed#222Subtraction, third set: the loader split at its seams into loading, editions, standing and history5 comply
• 2026-09-09passed#221Subtraction, second set: shared libraries into src, superseded protocols archived, the review-note label applied5 comply
• 2026-09-09passed#219Subtraction, first set: the Maintain era retired, each script naming the verb that replaced it5 comply
• 2026-09-09passed#218Reassessment of 2026-09-09, evening: subtract, then the UI pass5 comply
• 2026-09-09parked#216founder/amend 3 15 narrow veto5 violate
• 2026-09-09passed#215The automation is on: weekly cron, operation live, the throttle's window cleared4 comply · 1 violates (GPT-5.6 Sol (OpenAI))
• 2026-09-09parked#214Chain: 2026-09-095 comply
• 2026-09-09passed#213The edition owns the case's question; anchors may span pages; compound evidence is split and judged part by part5 comply
• 2026-09-09parked#212founder/amend 3 15 concurrence5 violate
• 2026-09-09passed#211Close the loop: the Arbiter reads a run's account, budgets that crunch then settle, the cadence backs off, the panel priced5 comply
• 2026-09-09passed#210The chain hands the panel its five keys, and the intake its GitHub token5 comply
• 2026-09-09passed#209A founder's drop falls under a recorded standing direction, verified by GitHub4 comply · 1 unsure (GPT-5.6 Sol (OpenAI))
• 2026-09-09parked#208Founder amendment §3.15: the founder's directions are the operating record5 violate
• 2026-09-09parked#207Founding inputs as sources, linked competition, split-and-judge, editions by stakes (code half of #206)4 comply · 1 violates (GPT-5.6 Sol (OpenAI))
• 2026-09-09parked#206The essay's propositions enter: a supplied document that is a ledger source anchors claims; PDF founding inputs are read; scheduler order and per-run half-done1 violates (GPT-5.6 Sol (OpenAI)) · 4 unsure
• 2026-09-07passed#192Investigate cases with web research and existing memory5 comply
• 2026-09-07passed#191Read bounded PDF sources with page and relevance checks5 comply
• 2026-09-07passed#190Add bounded source discovery to the research loop5 comply
• 2026-09-07passed#189Compose and independently compare research editions5 comply
• 2026-09-06passed#188Plan the later review of assessments and the full reading experience5 comply
• 2026-09-06passed#186Manage shared AI spending, use Astra for authoring, and seed Birdmen intake5 comply
• 2026-09-06passed#184Let editions feature catalog claims through versioned assessments5 comply
• 2026-09-06passed#183Drop trailing question marks from case titles4 comply · 1 unsure (Gemini 3.8 Flash (Google))
• 2026-09-06passed#182Unify inbox and watch imports through bounded research proposals5 comply
• 2026-09-06passed#181Save coherent research editions and preserve Deep Memory's opening5 comply
• 2026-09-06passed#180Open Deep Memory with an illustrated question and its first sourced observation5 comply
• 2026-09-06passed#179Common research proposals, bounded source reading, and Deep Memory imagery5 comply
• 2026-09-06passed#177Preserve historical source decisions before the intake cutover5 comply
• 2026-09-06passed#176Preserve historical agenda scores before the intake cutover5 comply
• 2026-09-06passed#175Preserve historical agenda proposals before the intake cutover5 comply
• 2026-09-06passed#174Retain intake decisions and reconsider research proposals in context5 comply
• 2026-09-06passed#173Unify source identity and retain intake decisions in context5 comply
• 2026-09-01passed#145The refusal fallback generalizes to every house-drafting call — and is banned from panel seats5 comply
• 2026-09-01passed#144Maintenance: 2026-09-01 run5 comply
• 2026-09-01passed#142Editorial audit joins the refusal-fallback path; stamps the model that answered5 comply
• 2026-09-01passed#141Bench v2: panel-scored proposals, advancing freezes, and the backfill5 comply
• 2026-09-01passed#138Revised build order + build step 1: the promotion pipe5 comply
• 2026-09-01parked#137Build order revised after the outside review: the promotion pipe leads3 comply · 2 violate (Opus 5 (Anthropic), Gemini 3.1 Pro (Google))
• 2026-09-01parked#135Remove pins entirely; narrative guidance moves into the prompt (founder simplification)1 complies · 4 violate
• 2026-09-01passed#134Pins: history.yaml joins the checkable files5 comply
• 2026-09-01passed#132Founding inputs: freeze the current tellings as source texts; first-edition and plates-survive rules4 comply · 1 unsure (Opus 5 (Anthropic))
• 2026-09-01passed#130Atelier design: narrative inputs anchor rewrites to the founding texts5 comply
• 2026-09-01passed#129VASO-S001 collection: the arteriole compartment is not unexamined, and what was found there is not latch5 comply
• 2026-09-01parked#128Metabolism 1+2: yield-allocated attention, pins, and the regression exam5 violate
• 2026-09-01parked#127The metabolism: full-automation design (docs/AUTOMATION.md)3 comply · 2 unsure (GPT-5.1 (OpenAI), Gemini 3.1 Pro (Google))
• 2026-09-01passed#126Fix build: exclude briefs/ archives from Tailwind source scanning5 comply
• 2026-09-01passed#125VASO-S001 freeze: has the latch ever been demonstrated at the caliber the knot hypothesis needs?5 comply
• 2026-09-01passed#124Transients: adopt TRN-C024 (marks on the original negative) and record the preprint publication4 comply · 1 unsure (Grok 4.6 (xAI))
• 2026-09-01parked#123Archive: Hancock twin-A later checkpoint (verified strict superset)2 comply · 3 unsure
• 2026-08-30parked#121Archive: raw chat-brief captures for Hancock and cast-vs-carved (science cases only)1 violates (Gemini 3.1 Pro (Google)) · 4 unsure
• 2026-08-28parked#118Amazon expansion: coverage-gap wave — full brief coverage, one failed item dropped4 comply · 1 violates (Gemini 3.1 Pro (Google))
• 2026-08-28passed#117docs: brief coverage dispositions and the living-persons tier rule5 comply
• 2026-08-28passed#116Link-preview cards: Open Graph + Twitter metadata with a house og-card image5 comply
• 2026-08-28passed#115Pre-flight citation check: run the arbiter's mechanical verification before the PR exists5 comply
• 2026-08-28passed#114docs: the chat-seeded case-brief workflow (capture → consolidate → verify → construct)5 comply
• 2026-08-28passed#113New case: The Emptied Amazon? — pre-contact population, Carvajal, and the heterogeneous mosaic (AMZ-001)5 comply
• 2026-08-28passed#112Claim genealogy, derivative sources, and two honest assessment states5 comply
• 2026-08-28passed#111LLM client: Fable-first by code default, loud Opus fallback with truthful stamps5 comply
• 2026-08-28parked#109LLM client: Fable-first by code default, one-shot Opus fallback on refusal4 comply · 1 violates (Opus 5 (Anthropic))
• 2026-08-27passed#110Agenda generation keys on the published slug, not the directory name5 comply
• 2026-08-27passed#108Proposal memory: editorial silence retires an idea5 comply
• 2026-08-27passed#107The proposals shelf: /proposals renders the weekly agenda proposals5 comply
• 2026-08-27passed#106Budget epoch 2: supervised sprint landings stop spending the autonomous budget5 comply
• 2026-08-27parked#103Content response: 2026-08-275 comply
• 2026-08-26passed#102Maintenance: weekly agenda generation — the ledger proposes its own next questions5 comply
• 2026-08-26passed#101Operator duty 4: draft needs-approval article updates when the ledger outruns the narrative5 comply
• 2026-08-26passed#100ORCH-S001 collection: the quantitative commitments of the Orch OR literature5 comply
• 2026-08-26passed#99Land the stranded #95 fix: parked-PR exclusion and the restored Villarroel entry5 comply
• 2026-08-26passed#98ORCH-S001 freeze: pre-register the quantitative-commitments study5 comply
• 2026-08-26passed#97CCC-S001 collection: thirteen rows, the signal-model-provenance column unanimous5 comply
• 2026-08-26parked#95Content response batches hourly instead of per canon push3 comply · 2 violate (Opus 5 (Anthropic), GPT-5.1 (OpenAI))
• 2026-08-26passed#94CCC-S001 freeze: the quantitative record of CCC signature claims and searches5 comply
• 2026-08-26passed#91Maintenance: 2026-08-26 run5 comply
• 2026-08-26passed#88Record Villarroel permission on the removed correspondence (§3.15 standard)5 comply
• 2026-08-26passed#86Urgent: remove private Villarroel correspondence from the public tree (§3.15)5 comply
• 2026-08-26passed#85The Study: pre-registered desk workpapers as a first-class domain object5 comply
• 2026-08-26passed#84fix(arbiter): keep balanced parentheses in extracted DOIs; reserve a budget share per diff tier5 comply
• 2026-08-26passed#83Surviving objections, the single-thread test, and two honesty lines5 comply
• 2026-08-26passed#82Reconciliation eligibility covers load-bearing splits (the vasocomputation gap)5 comply
• 2026-08-26passed#81Zero-content bootstrap: placeholder params for empty dynamic routes, empty states for homepage and case index5 comply
• 2026-08-26passed#80Arbiter: mechanical citation verification feeds the seats (prompt v3)5 comply
• 2026-08-26passed#79Fix dead evidence anchors: link evidence ids to the full ledger page5 comply
• 2026-08-26passed#77inbox-response: dispatch job needs contents:read to resolve the default branch5 comply
• 2026-08-26passed#76Vendor seat calls retry transient network failures5 comply
• 2026-08-26passed#75Reconciliation cannot self-ratify: engaged checks never vouch for the draft that answered them5 comply
• 2026-08-26passed#74content-response: fail-loud staging and manual dispatch trigger5 comply
• 2026-08-26passed#73generate-case-art: fall back to OPENAI_API_KEY when IMAGE_API_KEY is unset5 comply
• 2026-08-26passed#72Remove the global research page5 comply
• 2026-08-26passed#71Arbiter: temporal humility — unfamiliarity is not evidence of fabrication5 comply
• 2026-08-26parked#70Reverted: Cloud Agent environment config2 comply · 2 violate (Gemini 3.1 Pro (Google), Kimi K3 (Moonshot, via Venice)) · 1 unsure (Opus 5 (Anthropic))
• 2026-08-25passed#69Editorial-audit history label states the real approval mechanism5 comply
• 2026-08-25passed#67Evidence ledger page: browse every record of a case5 comply
• 2026-08-25passed#66Content response: 2026-08-254 comply · 1 unsure (Kimi K3 (Moonshot, via Venice))
• 2026-08-25passed#65The operator: reconciliation loop, parked-PR duty, issue triage — and the throttle learns eras5 comply
• 2026-08-25passed#64Panel page: full arbiter verdicts on-site, with verbatim-record links5 comply
• 2026-08-25passed#62Immortality Key: caption discipline — captions describe, claims argue5 comply
• 2026-08-25parked#60Transients: record citation permission on the microscopy manuscript5 violate
• 2026-08-25passed#59The gate is live; operations go event-driven5 comply
• 2026-08-25parked#58Immortality Key artwork: house-style cover and six provenance-labeled plates4 comply · 1 violates (Gemini 3.1 Pro (Google))
• 2026-08-25passed#57/panel: the governance made visible5 comply
• 2026-08-25passed#56Stage-4 tail: governance harvest, weekly digest, rate limit, prioritized diffs5 comply
• 2026-08-25parked#55Case #9: The Immortality Key (TIK-001) — ergot kykeon and psychedelic Eucharist2 violate (Opus 5 (Anthropic), Gemini 3.1 Pro (Google)) · 3 unsure
• 2026-08-25passed#53AI-operated pivot, stage 4 (dry period): the constitutional arbiter5 comply