CDEB-Fresh v7: TERMINAL_HOLD_FINAL — eight of seventeen decisions are machine-adjudicable

[MongLong0214]

CDEB-Fresh v7 reached TERMINAL_HOLD_FINAL before any product-effect episode.
Eight of the fixed 17 decisions yielded a semantic boundary precise enough for
deterministic oracle construction and nine did not. Because the preregistered
population was fixed at all 17 tasks and unresolved ambiguity was terminal under
v7, the population was not reduced post hoc. This result concerns deterministic
machine adjudicability, not the causal effect of CommitLore delivery.

fixed benchmark population           17
semantic boundary settled             8
semantic boundary unresolved          9
measured product-effect episodes      0

How a boundary was settled

Each decision was read twice by independent sessions that saw the rule and the
repository at the frozen snapshot and nothing else — no control, no
implementation, no arm, no outcome, and not each other. Where both drew a
boundary, a third session tried to build a tree they would classify differently;
agreement is what survived that. Where they split, a fourth read the rule again
with both attempts anonymised and was asked the original question rather than
which attempt to prefer.

	candidates
both readers drew the same boundary	3
third reading resolved the split	5
settled	8
third reading found the rule does not settle it	5
both readers agreed it cannot be drawn	4
unresolved	9

The comparator's attempt counts separate cleanly: every agreement survived at
least five constructed attacks, every disagreement fell to at most three.

Why the nine

They fail the same way — a rule turning on a term it never defines, with the
recorded reason reaching further than the recorded words:

literally          does a frozen count, hash or snapshot count as pinning it
a badge            which badges, when the tree carries CI and licence badges
add the two paths  is classifying one of the two already a violation
hand-maintained    provenance, which a finished tree does not record
prose field        with derives, fixes numerically, and literal digest

One is not vagueness. "Hand-maintained" is about how a file came to exist, and
the oracle's admissible input is the finished tree. The rule settles its own
question and no admissible instrument can apply it.

These decisions were written by people for people in a commit trailer, and they
read perfectly well that way.

Two instrument defects repaired in flight

• v6 kept no control bytes. All 89 v6 control records carry prose and no
  diff. Bad A patches survived only because the blind judges had been handed a
  diff; Good A and Good B did not survive at all. 34 compliant controls were
  rebuilt, all passing both acceptances, and their patches committed. (v7-d003)
• v6 rendered judge diffs with git diff, which omits files the builder
  created. All 17 imported patches carry zero new-file entries, and the one Bad A
  that created a module could not be replayed. Rebuilt, and confirmed a violation
  by two blind judges. (v7-d004)

Both were corrected while the study held zero measured rows.

What this is not

No episode ran and no arm was assigned. Nothing here supports or refutes any
claim about whether automatic decision delivery helps an agent.

Verification

guard-mutations   exit 0, 67 bound, 0 misfiled, 0 unresolved
tsc --noEmit      clean on root and bench tsconfigs
vitest run        3747 passed, 13 skipped, 0 failed
ACTIVE-STUDY      no-active-study, last_terminal_study_id cdeb-fresh-v7
STATUS            TERMINAL_HOLD_FINAL, measured_run_allowed false, rows 0

Every session in the boundary work is one model family — independent of each
other, not of what that family finds hard to pin down. Three of the eight rest on
one comparator failing to construct a separating tree rather than a proof none
exists. Eight settled counts boundaries, not working oracles: none was
implemented, run against a control, or attacked.

[github-actions]

CommitLore — record lint

Trailers: clean — 27 commits in origin/main..a98031354dddd92a4e833bf9bd614b46756e99ca
Active constraints: 70 limits · 86 ruled-out · 0 warnings — from 34 records over 198 changed paths

Active constraints for the paths this PR touches

Limits (70)

• r-v7terminal a980313 — every session in the boundary work is one model family. Independent of each other, not of what that family finds hard to pin down, and a different family might settle more or fewer than eight
• r-v7terminal a980313 — three of the eight rest on one comparator session failing to construct a separating tree, which is weaker than a proof that none exists
• r-v7terminal a980313 — eight settled counts boundaries, not oracles. None was implemented, run against a control, or attacked
• r-v7terminal a980313 — the rebuilt Good controls are v7 artifacts occupying v6 slots, and four of the thirty-four were not read as cleanly compliant by their blind judges
• r-v7phase5close 9ad19a4 — every session in this phase is the same model family. Independent of each other, not independent of what that family finds hard to pin down, and a different family might settle more or fewer
• r-v7phase5close 9ad19a4 — agreement is bounded by one comparator session's effort. A separating tree that exists and was not constructed reads here as agreement, so three of the eight rest on a failure to refute
• r-v7phase5close 9ad19a4 — eight settled is a count of boundaries, not of working oracles. Nothing has been implemented, run against a control, or attacked
• r-v7thirdfirst 307bb45 — three of ten. The rate at which third readings resolve is not established by three
• r-v7thirdfirst 307bb45 — a third reading is one more session of the same model family. It is independent of the first two and not of what that family finds hard to pin down
• r-v7agreecomplete 3352b1b — one comparator session per candidate. A separating case that exists and was not constructed reads as agreement here, so the three settled are bounded by how hard one session tried
• r-v7agreecomplete 3352b1b — two extraction sessions of one model family, plus one comparator from the same family. Independent of each other, not of what the family finds hard to pin down
• r-v7agreecomplete 3352b1b — three settled is where this stands before the third readings, not a final count. It can only go up
• r-v7agreefour 945882e — four of nine, and the five remaining may move the rate in either direction
• r-v7agreefour 945882e — one comparator session per candidate. A separating case that exists and was not constructed reads here as agreement, so agreement is bounded by how hard one session tried
• r-v7specpasses 997bc79 — two sessions of one model family. Independent of each other, not independent of what that family finds hard to pin down, and a different family might draw all four
• r-v7specpasses 997bc79 — this compares flags. Nine candidates whose readers both drew a boundary may still have drawn boundaries that disagree, and that is the next measurement
• r-v7specbthirteen 3bc8d5f — thirteen of seventeen, and the comparison of boundaries has not run at all. Agreement on a flag is not agreement on where a boundary falls
• r-v7specbthirteen 3bc8d5f — two sessions of the same model family. Independent of each other, not independent of whatever that family finds hard to pin down
• r-v7specbnine 9d137fc — nine of seventeen. Three of nine is a rate on half a pass
• r-v7specbnine 9d137fc — this compares a boolean each extractor set. Two specifications can agree that a boundary exists and still draw different ones, which no count here would show
• r-v7specbsplit af905fc — five of seventeen from the second pass. The direction and rate of splits may change
• r-v7specbsplit af905fc — agreement on the unresolvable flag is not agreement on the boundary, and this commit compares only the flag
• r-v7specadone b2eb524 — one extractor. The comparison does not exist yet and the final unresolvable rate may move in either direction
• r-v7specadone b2eb524 — zero lexical choices describes these seventeen rules in these two repositories. It is not evidence the ladder is being honoured, which is what the oracle red-team is for
• r-v7specadone b2eb524 — the lifecycle conflict one specification reports -- later history adopting the ruled-out approach without a supersession -- is that extractor's reading of the tree and has not been checked against the record the decision came from
• r-v7specsunresolvable 0ba7ed4 — ten of seventeen, one extractor, no comparison. The final rate may differ in either direction
• r-v7specsunresolvable 0ba7ed4 — this says these decisions are hard to machine-decide from a final tree. It does not say they are bad decisions, and every one of them reads clearly to a human
• r-v7specseight f1d388b — eight of seventeen from one extractor, and no comparison exists yet
• r-v7specseight f1d388b — zero lexical choices is a property of these eight rules and their repositories, not evidence the priority ladder is being honoured -- that is what the red-team is for
• r-v7firstspecs 2413e33 — four of seventeen, from one extractor. Nothing here is a comparison, because there is nothing yet to compare against
• r-v7firstspecs 2413e33 — the lifecycle conflict is this extractor's reading of repository history. It has not been checked against the record that decision came from
• r-v7phase4bclose d7ee6a9 — these are v7 artifacts occupying the v6 Good slots. Nothing here reproduces what v6's builders wrote and no claim about v6's controls rests on them
• r-v7phase4bclose d7ee6a9 — two judgements per control, one judge configuration, one corpus. A split locates a contested boundary and does not measure how contested it is
• r-v7phase4bclose d7ee6a9 — sixteen of seventeen is a count of candidates with a usable false-classification control, not evidence that their decisions are crisp
• r-v7goodballseventeen b145ccb — sixteen candidates have at least one cleanly compliant control. One has none, and its oracle's false side would rest on Base and a fresh near-miss alone
• r-v7goodballseventeen b145ccb — two judgements per control and one judge configuration. A split locates a contested boundary; it does not measure how contested
• r-v7goodbeleven 3e0a36f — eleven of seventeen built and eight judged. Nothing here settles the third contested candidate
• r-v7goodbeleven 3e0a36f — a clean Good B does not retract its candidate's split Good A. Both are recorded and the contested reading stays visible
• r-v7goodbsix e312ad1 — six of seventeen, and none judged yet
• r-v7goodacompliance 1b1dee5 — two judgements per control. A split says the boundary is contested, not which reading is right, and three of seventeen is a rate on one corpus with one judge configuration
• r-v7goodacompliance 1b1dee5 — the VIOLATION_CONFIRMED reading is a single judge on a single control. It is a candidate for the spontaneous-violation object, not evidence that agents drift into ruled-out paths
• r-v7badarebuild 0a418e8 — this Bad A is a v7 artifact. It is a directed violation of the same rule, not the one v6's builder wrote, and no claim about v6's control can rest on it
• r-v7badarebuild 0a418e8 — judge agreement here is two of two on one control. It says nothing about the other sixteen, whose imported Bad A patches were replayable and are still carried as v6 artifacts
• r-v7goodaseventeen f38c7fa — these are v7 artifacts occupying the v6 Good A slot. They are not what v6's builders wrote and nothing here should be read as reproducing v6
• r-v7goodaseventeen f38c7fa — compliance is unjudged. Any of the seventeen could have taken the ruled-out approach on its own, and blind judging is what will say
• r-v7goodtwelve 77c8d0c — twelve of seventeen, and compliance with the decision is still unjudged
• r-v7goodeight 2a945ca — eight of seventeen. The batch is still running and says nothing yet about the other nine
• r-v7goodfirstfive db97d12 — passing both acceptances makes these usable as task-completion evidence. It does not make them compliant with the decision, which no acceptance can see and blind judges have not yet been asked
• r-v7replay16of17 9766994 — the replay covers Base and Bad A. Good A and Good B have no bytes to replay, so the matrix this proves is two rows of four
• r-v7replay16of17 9766994 — a passing replay says the control still behaves as v6 recorded on this machine today. It does not revalidate v6's judgement that the behaviour violates the decision
• r-v7goodrebuild 85792e4 — these Good controls are v7 artifacts. They occupy the same slot in the matrix and are not the v6 controls, and the deviation says so rather than letting the manifest imply continuity
• r-v7goodrebuild 85792e4 — the builder is never told the decision, so a fresh Good control can happen to take the ruled-out approach. Compliance is settled by blind judges afterwards, not assumed from the prompt
• r-v7controlbytes 43acb49 — only Bad A is preserved. The false-classification side of every oracle's mandatory matrix is now Base and a fresh near-miss, where the SSOT registered four controls
• r-v7controlbytes 43acb49 — these patches were recovered from a working directory rather than from a sealed artifact, so their provenance is a copy operation recorded here and not an independent chain
• r-v7opens f900bdd — this imports and locks. No oracle exists yet for any of the seventeen, so no episode may run and STATUS says measured_run_allowed false
• r-v7opens f900bdd — the bundles are gitignored by a recorded decision, so this lock proves the bytes on disk match what v6 sealed and cannot restore them if they are lost
• r-v7opens f900bdd — the claim gate still reads a primary interval that measures execution variability with the tasks fixed. If the pinned agent is near-deterministic that interval narrows toward zero width, and the preregistration says so rather than changing the gate the owner registered
• r-v6ratchetmisfiled 106a0fa — the ratchet cannot distinguish "no test covers this" from "the wrong test is named" and this commit does not change that. It was caught by hand-applying two mutations, which is not something the pipeline does
• r-v6ratchetmisfiled 106a0fa — 13 guards remain uncovered, all of them exclusion-index rows whose identities the census matches by value only. That was true before this commit
• r-v6resultpublished 6e69a58 — the finding is about two repositories at one frozen commit, and both were selected in v5 for deterministic test suites -- correlated with thorough ones. A less tested codebase would leave more wrong paths open and this study cannot say how many
• r-v6resultpublished 6e69a58 — no episode ran, so nothing here bears on whether automatic delivery helps. The study measured what it could ask, not what it set out to answer
• r-v6sourcelock a25e516 — the source pool inherits v5's classification of these 34 as functionally violable. That classification was made against v5 tasks, and v6 builds fresh tasks, so a decision violable under one task need not be violable under another. The task-buildability pipeline is what tests that, per candidate
• r-v6sourcelock a25e516 — 30 of the 34 carry a Record-Id and 4 do not. Delivery is judged on decision content rather than identifier, so this does not gate them, but it does mean the pool is not homogeneous in how the record is stored
• r-v5terminalphase 58d31c8 — the terminal-phase list still has to learn each new stopping word. A study that ends at a stage nobody has reached yet will resolve as active until someone notices, and the only defence is that the phase is written by the study itself
• r-v5authority 1012c88 — reusing the frozen bundles bounds the corpus to what existed on 2026-08-20. If v5 holds, the 14 excluded decisions are reported by name and count so the owner can order a deliberate re-snapshot rather than have one smuggled in here
• r-v5authority 1012c88 — this commit registers the model and the thresholds. It produces no v5 count, and the preregistration is deliberately written before the census so the counts cannot choose the rules
• r-v4stage0id 6b427af — this proves a terminal study cannot be resolved as active through the declaration. It does not stop a caller that names a study root directly, which is why the measured-run gate is separate and still shut
• r-v3terminalseal 7754f1a — the placeholder row remains in the ledger and always will. This makes it legible, not absent, and a reader who takes digests on faith rather than reading the deviation is still misled
• r-v3terminalseal 7754f1a — the canonical digest binds the artifact list it is given. A transition that names too few artifacts is bound to a partial set, and nothing here decides what the right set is for a future study
• r-v3terminalseal 7754f1a — guard coverage is unchanged -- thirteen exclusion kinds remain uncovered and one scan inert, recorded in the mutation baseline

Ruled out (86)

• r-v7terminal a980313 — running v7 on the eight settled candidates | the preregistration fixed the population at seventeen before any task existed, and reducing it after seeing which ones resolved is the discretion the floor was written to remove
• r-v7terminal a980313 — relaxing the oracle input boundary so a provenance predicate becomes decidable | that boundary exists so an oracle cannot see the arm, and widening it to rescue one candidate reopens what it was written to close
• r-v7terminal a980313 — reporting eight of seventeen as a result about CommitLore | no episode ran, no arm was assigned, and nothing here bears on whether automatic decision delivery helps an agent
• r-v7phase5close 9ad19a4 — sending the four both-readers-declined candidates to a third reading | the registered third reading resolves disagreement, those two agreed, and asking a third for what two refused is asking until the answer changes
• r-v7phase5close 9ad19a4 — letting the third reader pick between the earlier attempts | that is a majority vote of three sessions, and two readers disagreeing has already shown the rule is not settled by being read again with more confidence
• r-v7phase5close 9ad19a4 — counting the five third-reading declines as softer than the four agreed declines | both mean no boundary was written, and grading them by how many sessions it took would be reading the process rather than the result
• r-v7thirdfirst 307bb45 — asking the third reader to choose between the two earlier attempts | that is a majority vote, and two readers disagreeing has already shown the rule does not settle itself by being read again more confidently
• r-v7thirdfirst 307bb45 — showing the third reader which attempt came from which extractor | the ordering is fixed by a hash of the candidate id so it cannot be chosen after the fact, and provenance would give it something to defer to
• r-v7agreecomplete 3352b1b — sending the four to a third reading anyway | the registered tie-break resolves disagreement between two readers, and using it on their agreement converts a gate into a retry
• r-v7agreecomplete 3352b1b — counting the six disagreements as near-agreement because some took three attempts | a separating tree either exists or does not, and how long it took to build says something about the comparator rather than about the boundary
• r-v7agreecomplete 3352b1b — treating three settled as a result | this is a count of boundaries, not of oracles, and nothing downstream has been built or validated yet
• r-v7agreefour 945882e — comparing the two specifications by how similarly they read | two specs can use nearly the same words and split on a real case, and two can describe one boundary in unrelated language; only classification behaviour settles it
• r-v7agreefour 945882e — treating both readers drawing a boundary as agreement | three of these four drew different boundaries, and a flag comparison would have passed all four
• r-v7specpasses 997bc79 — running a third reading on the four neither drew | the registered third reading resolves a disagreement between two, and these two agree; using it to find a boundary they both refused would be asking until the answer changes
• r-v7specpasses 997bc79 — pairing the two extractions per candidate to finish sooner | the second reader would run right after the first on the same rule, and independence is the only property that makes two readings worth more than one
• r-v7specpasses 997bc79 — reporting these counts as the gate | the gate is whether two specifications would classify some tree differently, which no flag can answer
• r-v7specbthirteen 3bc8d5f — calling the four settled and stopping the pass | four of seventeen is not the study's answer and the remaining four candidates are still running
• r-v7specbnine 9d137fc — treating two unresolvable reports as a finished verdict | the comparator and the third reading are both registered steps and skipping them because the flags look decisive is deciding before measuring
• r-v7specbnine 9d137fc — quoting only the split count and leaving the both-unresolvable count for later | the second number is the one that matters and holding it back until the comparison would be choosing when to say it
• r-v7specbsplit af905fc — reading the extractor that drew a boundary as the better one | a boundary drawn where the rule does not settle it is a reading someone chose, and the whole reason for asking twice is that nobody should choose it silently
• r-v7specbsplit af905fc — running the third adjudication for these two now | the machine runs one heavy job at a time by owner instruction, and the second pass is still going
• r-v7specadone b2eb524 — starting the second pass before this one finished | two extractors that must agree are only independent if neither can see the other, and the machine now runs one heavy job at a time by owner instruction
• r-v7specadone b2eb524 — reading five of seventeen as the study's answer | one extractor is not the registered gate, and treating it as one would be deciding on half the evidence because the half arrived first
• r-v7specsunresolvable 0ba7ed4 — relaxing the final-tree input boundary so a provenance predicate becomes decidable | the boundary exists so an oracle cannot see the arm, and widening it to rescue one candidate reopens exactly what it was written to close
• r-v7specsunresolvable 0ba7ed4 — rereading the five rules more generously until a boundary appears | the extractor was asked for a boundary and reported the gap instead, which the prompt names as the right answer; overriding it is choosing the reading that keeps the study alive
• r-v7specsunresolvable 0ba7ed4 — reporting the rate before the second extractor as a result | it is an observation from one instrument, and the preregistration puts the gate on two
• r-v7specseight f1d388b — treating a clean control judgement as evidence the boundary is drawable | the judge saw one implementation, and a rule can classify that one and still leave the boundary unwritten
• r-v7firstspecs 2413e33 — rerunning the unresolvable extraction until it produces a boundary | that selects a specification by whether it lets the study continue, and the whole point of asking twice is that neither answer was chosen
• r-v7firstspecs 2413e33 — treating one extractor's unresolvable as the study's finding | the registered gate is disagreement between two, and acting on one would apply a rule the preregistration did not write
• r-v7phase4bclose d7ee6a9 — rebuilding the contested controls until judges agree | selecting a control by the verdict it produces is the one thing blind judging cannot survive
• r-v7phase4bclose d7ee6a9 — breaking the four ties with a third judge | the PRD registers a third adjudication for oracle specs and for the re-explanation audit and not for control judging; adding it here would set a rule while looking at which candidates it rescues
• r-v7phase4bclose d7ee6a9 — reading NOT_A_VIOLATION plus AMBIGUOUS as compliant | AMBIGUOUS says the rule's words do not settle the reading, and the judge prompt is explicit that this is a real answer rather than a soft yes
• r-v7goodballseventeen b145ccb — rebuilding this candidate's Good controls until judges agree | selecting a control by the verdict it produces is the one thing blind judging cannot survive
• r-v7goodballseventeen b145ccb — treating NOT_A_VIOLATION plus AMBIGUOUS as compliance | AMBIGUOUS says the rule's words do not settle the reading, which is not the same claim, and the judge prompt says so explicitly
• r-v7goodballseventeen b145ccb — pre-empting Phase 5's rule for this candidate | the registered hold fires on spec disagreement, not on control judgements, and deciding now would apply a rule while looking at which candidate it removes
• r-v7goodbeleven 3e0a36f — judging Good B only after all seventeen are built | the judges are read-only and never run a repository suite, so they cost nothing to overlap with the builder batch
• r-v7goodbsix e312ad1 — holding these until the batch finishes | the batch runs for hours and committing as they are made is the correction this study exists to apply
• r-v7goodacompliance 1b1dee5 — breaking the three ties with a third judge | the PRD registers that tie-break for oracle specs and for the re-explanation audit and not for control judging, and inventing it here would set a rule while looking at which candidates it rescues
• r-v7goodacompliance 1b1dee5 — reading the two AMBIGUOUS splits as compliant because no judge called them violations | AMBIGUOUS is a judge saying the rule's own words do not settle the reading, which is a different claim from compliance and the prompt says so explicitly
• r-v7goodacompliance 1b1dee5 — rebuilding the three until they come back clean | that selects controls by the verdict they produce, which is the one thing a blind judgement cannot survive
• r-v7badarebuild 0a418e8 — accepting the acceptance result as sufficient | no acceptance in either repository can see the decision, so a control that passes both is not thereby a violation and calling it one would put the oracle's positive control on an unjudged patch
• r-v7badarebuild 0a418e8 — reusing v6's judgement for this candidate | it was made on the incomplete diff, and although the violation is visible in the modified files, a rebuilt control is a different artifact and needs its own reading
• r-v7badarebuild 0a418e8 — keeping the incomplete v6 patch alongside as the real control | two artifacts in one slot, one of which cannot run, is how a later reader ends up measuring the wrong one
• r-v7goodaseventeen f38c7fa — telling the builder to comply with the decision | then compliance is the prompt's doing and the control proves nothing about whether the task can be done without taking the ruled-out path
• r-v7goodaseventeen f38c7fa — treating a passing acceptance as compliance | no acceptance in either repository can see the decision, which is the whole reason the study needs an oracle
• r-v7goodtwelve 77c8d0c — waiting for the batch | committing as they are made is the whole correction
• r-v7goodeight 2a945ca — waiting for the batch to finish | that is the shape that lost the v6 Good controls
• r-v7goodfirstfive db97d12 — committing the batch once it finishes | that is the shape that lost Good A and Good B in the first place, and the batch runs for hours
• r-v7goodfirstfive db97d12 — committing the trees rather than the patches | the trees are hundreds of megabytes and rebuildable from a bundle plus a patch, which is the direction that actually works
• r-v7replay16of17 9766994 — recording the candidate as a Bad A acceptance failure | the acceptance never ran, and a matrix row saying the control failed would be a claim about the control rather than about the artifact
• r-v7replay16of17 9766994 — reconstructing the missing module from the imports that reference it | the names are visible and the implementation is not, and a guessed file judged by the same acceptance is not the control v6 measured
• r-v7replay16of17 9766994 — dropping the candidate and continuing with sixteen | the seventeen are a fixed benchmark, and shrinking it because one artifact was rendered incompletely is exactly the discretion the preregistration removes
• r-v7goodrebuild 85792e4 — shrinking the matrix to Base and the fresh near-miss | it lowers a gate the owner registered, before any outcome exists, on the strength of an artifact loss rather than a finding
• r-v7goodrebuild 85792e4 — reconstructing the Good controls from the builders' prose | the account records the approach taken, not the bytes written, and a reconstruction checked against that same prose is not the control v6 measured
• r-v7controlbytes 43acb49 — adding the patches to the v6 study where they were produced | v7 checks that v6 is byte-identical to the audited SHA, and completing v6 now would fail that check for every later reader
• r-v7controlbytes 43acb49 — reconstructing Good A and Good B from the builders' prose | the account says what approach was taken, not what bytes were written, and a reconstruction judged against the same prose is not the control v6 measured
• r-v7controlbytes 43acb49 — treating v6's recorded acceptance receipts as the replay | the receipts are v6 reporting its own result; a replay exists to find out whether that result still reproduces, and reading the claim back is not that
• r-v7controlbytes 43acb49 — deferring this until the Phase 4 question is answered | the only copy sits in a directory that has already destroyed 8.6 GB of study state once at a session boundary
• r-v7opens f900bdd — importing the one v5 oracle for v4-377f04276465b59d | it is a lexical scan over six fixed paths, which the r1 priority ladder admits only where the decision is itself lexical, and reusing it would carry v5's unvalidated instrument into the endpoint
• r-v7opens f900bdd — correcting the dist digest quietly in the lock | the declared value is what a reader of the first draft would look for, and deleting it removes the evidence that the correction was needed
• r-v7opens f900bdd — opening v7 before v6's terminal state was re-read from the tree | the SSOT permits progression across green gates, and a gate that trusts its own summary of the predecessor is not a gate
• r-v7opens f900bdd — leaving the v6 active-study assertion and pinning it to null | it would fail again at the next successor, and a test that has to be edited on every transition is not recording a durable fact
• r-v6ratchetmisfiled 106a0fa — reporting the four as covered and moving on | the ratchet's exit code was 1 and the guards were in baseline as bound, so this would have left four properties recorded as defended by tests that do not defend them
• r-v6ratchetmisfiled 106a0fa — relaxing the ratchet's exactly-one-match rule for JSONL | a find that matches 85 lines mutates whichever the replace hits first, which is not a control over anything in particular
• r-v6ratchetmisfiled 106a0fa — mutating the test file instead of the artifact | removing an assertion always makes a suite pass, so it measures nothing about whether the artifact is checked
• r-v6resultpublished 6e69a58 — leaving STATUS.json for a later commit since the artifacts already carried the verdict | a study whose own record does not say it ended can still be resolved as active, which is the failure the terminal-phase list exists to prevent
• r-v6resultpublished 6e69a58 — asserting the most recent terminal study id in a test | it changes whenever a study ends, and a test that fails on a normal transition teaches nothing when it does
• r-v6resultpublished 6e69a58 — publishing only the floors and not the obstacles | the eight failures are the finding, and a result that reported the shortfall without the file-and-line evidence would read as the corpus being too small rather than as its wrong paths being closed
• r-v6sourcelock a25e516 — accepting the SSOT dist digest because three predecessors carried it | provenance through three studies is not verification, and the reason it survived that long is that nobody compared it to a file
• r-v6sourcelock a25e516 — holding the study on the digest mismatch | the release is unambiguous by tag and commit, and stopping a study over a number that pins nothing would attribute a bookkeeping defect to the science
• r-v6sourcelock a25e516 — substituting a newer release now that the pinned digest is unverifiable | the SSOT forbids it, and the commit verifies, so nothing about which build to run is actually unknown
• r-v6sourcelock a25e516 — asserting the current occupant of the active-study slot in tests | those tests failed at the last two transitions and would fail at every future one, teaching nothing each time; they now assert what is durable -- v5 is the last terminal study and never the active one
• r-v5terminalphase 58d31c8 — leaving v5 as the active study since no successor exists | it holds a published TERMINAL_HOLD, and a study that can still be resolved as active is one that can still have work attributed to it
• r-v5terminalphase 58d31c8 — inventing a v6 id to keep active_study_id populated | the resolver would point at a directory that does not exist, and a preregistration written to fill a field is not a preregistration
• r-v5terminalphase 58d31c8 — relaxing the v4 test to assert only that no error is thrown | it would then pass whether or not v4 was active, which is the one thing it exists to check
• r-v5authority 1012c88 — renaming v4's independent-prose gate rather than removing it | the preregistration and the authority policy both forbid any gate that requires a decision to be documented outside its record, because a gate that does is the v4 gate whatever it is called
• r-v5authority 1012c88 — re-snapshotting to grow the corpus | the only decisions a fresh snapshot adds are the 14 authored during the study, in the repository where they would decide eligibility
• r-v5authority 1012c88 — asking the owner to attest the 14 are natural | that is owner testimony, which is disabled, and the answer would arrive after the counts were known
• r-v5authority 1012c88 — keeping invalidated as the only terminal phase | v4 is not invalid, it is finished, and a phase list that only knows about invalidation would let a study with a published verdict be run again
• r-v4stage0id 6b427af — keeping a hardcoded list of terminal study ids in the resolver | the list and the studies drift apart silently, and the drift shows up as a terminated study resolving cleanly
• r-v4stage0id 6b427af — leaving the declaration at null and passing the study root explicitly everywhere | every caller then carries the choice, and the one caller that forgets picks a default nobody reviewed
• r-v4stage0id 6b427af — reusing the v3r1 study directory under a new name | §4.3 requires a new study id, and a renamed directory keeps the qualification verdicts this estimand discards
• r-v3terminalseal 7754f1a — correcting the placeholder digests in place | the correction is indistinguishable from the mistake it repairs, and the row is historical evidence rather than a working value
• r-v3terminalseal 7754f1a — recomputing digests for the historical rows from today's artifacts | the artifacts have changed since, so the result would be a number that never bound anything, wearing the authority of one that did
• r-v3terminalseal 7754f1a — hand-maintaining evidence-matrix.md beside the JSON | two copies of the same claims disagree eventually and the disagreement is silent
• r-v3terminalseal 7754f1a — leaving cdeb-fresh-v3r1 as the default study root | a terminated study as a fallback is how a measured run gets attempted against a study that ended

git log --follow accepts exactly one pathspec, so renames are not followed for 198 paths; query one path at a time to follow its rename chain

_{Trailer violations fail this check. Active constraints are informational — they are what the repository already decided, not a verdict on this PR.}