-
Notifications
You must be signed in to change notification settings - Fork 0
docs: this repository is an instrument utility, and publishes nothing else - #101
docs: this repository is an instrument utility, and publishes nothing else #101test1card wants to merge 14 commits into
Conversation
...cribed here Owner direction, 2026年08月20日: information about the private working layer belongs only in that layer, and this repository is the laboratory instrument and nothing else. Two places in this repository described that layer, and both predate the direction: - `docs/OBLIGATIONS.md` carried a row whose whole subject was the layer's private remote, including its clone URL. The register is append-only by its own rule, and the direction overrides that rule for this row: it is removed rather than restated with a changed status, because a status change would keep the text. - `docs/campaigns/MONTANA_CAMPAIGN_ARCHIVE.md` carried a section describing the layer by name and enumerating its internal directories and what each one holds. Both are removed. What is NOT removed, and the distinction is deliberate: the places that say directions are registered in this repository and that a working corpus existed outside it. Those are statements about how THIS repository is governed, they name nothing and describe no structure, and removing them would delete the reason the obligations register exists. Documentation only, no source change. `tests/docs` and the agent-guidance cross-reference guard are green at 70 passed.
...repository Owner, 2026年08月20日: everything private -- the working relationship, the plans and the development directives -- is stored privately, and this repository is an instrument utility and nothing more. TWO DOCUMENTS ARE REMOVED, and they were chosen by measurement rather than by feel: both had ZERO references from tests, governance and continuous integration, so they are the part of this direction that costs nothing to carry out today. - `docs/OBLIGATIONS.md`, 17782 bytes -- the register of owner directions. It is removed rather than emptied, because an emptied register is still a statement about how the owner works. - `docs/campaigns/MONTANA_CAMPAIGN_ARCHIVE.md`, 137300 bytes -- campaign coordination history. Both were copied out byte-exactly and compared by hash before either was removed here. Nothing is lost; it is simply not published. EVERY POINTER TO THEM IS REMOVED TOO, and none of the replacements says where they went -- naming the place is the same mistake in a smaller font. Fourteen pointers across `PROJECT_STATUS.md`, `README.md`, `ROADMAP.md`, `docs/DECISIONS.md`, `docs/MONTANA_IMPLEMENTATION_AGENT_SPEC.md` and the two ASD-STE100 skill files now say "the obligations register", "kept outside this repository", or nothing at all. The obligation identifiers stay, because a reader who meets OB-012 in a comment needs to know it is a real thing with a name; what disappears is the path. The one token that a trigger depends on -- `phase-1-status` in `ROADMAP.md` -- is untouched, and its comment still forbids rewording it. WHAT IS NOT DONE HERE, and it is the larger half. The same direction reaches `ROADMAP.md`, `AGENTS.md`, `docs/DECISIONS.md`, `docs/OPEN_CELLS.md`, `docs/CLAIM_CORRECTIONS.md` and `governance/agent_preventions.yaml`, which together carry 27 references from tests, governance and continuous integration and are what the required checks read. Moving those is a separate change with a sequencing decision behind it, and it is not smuggled into this one. Documentation only, no source change. `tests/docs` and `tests/governance` are green at 364 passed with the derived pair regenerated.
Fourteen pointers across PROJECT_STATUS.md, README.md, ROADMAP.md, docs/DECISIONS.md, the implementation spec and the two ASD-STE100 skill files now say the obligations register, or kept outside this repository, or nothing -- never a path. Naming the place is the same mistake in a smaller font. The obligation identifiers stay: a reader who meets OB-012 in a comment needs to know it is a real thing with a name. What disappears is the path. The phase-1-status token a trigger depends on is untouched, and its comment still forbids rewording it.
test1card
commented
Aug 20, 2026
@codex review
Head is 8051984edf4d752297070bdd5ada657efdc84ca5. Documentation only, no source change.
Two documents leave the repository and every pointer to them is rewritten so that none names a path. The two were selected by measurement: both had zero references from tests/, governance/ and .github/, which is why this change cannot move a required check.
Please look hardest at three things:
- A pointer I missed. A search for
OBLIGATIONS.mdandMONTANA_CAMPAIGN_ARCHIVEover the tree returns nothing, but a reference spelled differently — a baredocs/campaigns/directory mention, a relative link, a path split across a line break — would not be caught by that search. - A claim that no longer has a source. Several documents cited those two files as the authority for a statement. Where the citation is now "the obligations register", the statement still has to be true and still has to be checkable by someone who cannot read that register.
- The
phase-1-statustoken inROADMAP.md. A registered obligation keys its trigger to that exact string. It is untouched, and the comment forbidding a reword is untouched, but that is the one thing in this change whose breakage would be silent.
tests/docs and tests/governance are green at 364 passed, and the derived documentation pair is regenerated to a fixed point.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 8051984edf
i️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
P1 Badge Supersede the owner decision instead of rewriting it
This rewrites an existing [Owner] decision from keeping obligations in-tree to keeping them outside the repository, while its unchanged heading still says they are "registered in-tree." Because docs/DECISIONS.md is explicitly append-only, this both erases the historical decision and leaves the current source of truth internally contradictory; restore the original entry and record the move in a new superseding decision.
AGENTS.md reference: AGENTS.md:L52-L58
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
P1 Badge Keep OB-002 attached to the phase trigger
When this marker eventually changes to DONE, a repository-only reader can no longer determine that OB-002—the adaptation-skill obligation—is what fired, because the rewrite replaced the identifier with an anonymous "registered obligation" and the register is no longer readable here. Preserve the OB-002 identifier and enough of its required action beside the marker so completing Phase 1 cannot silently strand the deferred direction.
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
P2 Badge Remove the surviving campaign-directory pointers
This changed exemption still points to docs/campaigns/, even though deleting the archive removes that directory. A repository-wide search also finds the same stale path in ROADMAP.md, README.md, README.ru.md, PROJECT_STATUS.md, and docs/architecture.md; readers therefore still see a path that resolves nowhere, and the intended path-free rewrite is incomplete.
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
P2 Badge Retain the P9 rule that derives the roadmap state
Moving the only full P9 definition outside the repository leaves the subsequent MERGED_P9_OPEN conclusion unverifiable for a repository-only reader: the listed facts remain visible, but the rule that maps those facts to that outcome and defines MERGED_CHECKPOINT was present only in the deleted archive. Because the post-Montana programme below uses this outcome to decide which work may start, retain a self-contained summary of the P9 decision and closure criteria here rather than requiring access to the external campaign record.
Useful? React with 👍 / 👎.
Four review findings, and the first one is the sharpest: I rewrote an append-only owner decision in place. AN APPEND-ONLY LOG IS NOT EDITED. `docs/DECISIONS.md` says so about itself, and I changed a 2026年08月09日 entry from "obligations are registered in-tree" to the opposite while leaving its heading claiming the original. That both erased the decision and left the file contradicting itself. The body is now WITHDRAWN with the heading kept, so the sequence has no silent gap and no edited entry claims to be the original, and a new dated entry records the supersession and why the earlier decision was made -- it solved a real failure and traded it for a worse one. A TRIGGER LOST THE NAME OF WHAT IT FIRES. The `phase-1-status` token is keyed by OB-002, and the rewrite replaced that identifier with an anonymous phrase. A reader of this repository could then flip the token to DONE and strand the deferred direction with no way to know what it was. The identifier is back, and enough of the required action stands beside it to act on without reading anything else. A DERIVATION LOST ITS RULE. The roadmap derives the outcome MERGED_P9_OPEN and then uses it to decide which work may start, but the rule mapping facts to that outcome lived only in the removed archive. The bounded-cycle rule and all three terminal outcomes are restated here in full. AND THE PATH-FREE REWRITE WAS INCOMPLETE. `docs/campaigns/` is now an empty directory and six files still pointed at it: PROJECT_STATUS.md, README.md, README.ru.md, ROADMAP.md, docs/architecture.md and the ASD-STE100 skill. None does now. The Russian sentences were edited through UTF-8 and read back afterwards, which is the only way to know a Windows tool did not eat them. `tests/docs` and `tests/governance` are green at 365 passed.
test1card
commented
Aug 20, 2026
@codex review
Head is b851b660c4481aa1e476480dc644b0735b41ec5b. All four findings are closed, and the first one was the sharpest.
- I rewrote an append-only entry in place.
docs/DECISIONS.mdsays that about itself, and I changed a 2026年08月09日 decision to its opposite while leaving the heading claiming the original. The body is now WITHDRAWN with the heading kept — so the sequence has no silent gap and no edited entry pretends to be the original — and a new dated entry records the supersession, including why the earlier decision was made: it solved a real failure and traded it for a worse one. - OB-002 is back beside the
phase-1-statustoken, with enough of its required action stated there to act on. Flipping that token to DONE can no longer strand a direction whose name the reader cannot recover. - The bounded-cycle rule is restated in full, all three terminal outcomes included, so the
MERGED_P9_OPENderivation below it is checkable without reaching for anything outside this repository. docs/campaigns/is empty and nothing points at it now — the six files you named are all fixed. Your search was right and mine was too narrow: I had searched for the two file names, not for the directory.
The Russian sentences in PROJECT_STATUS.md and README.ru.md were edited through UTF-8 and read back afterwards; one of them also needed rewording rather than substitution, because dropping the path left the clause ungrammatical.
tests/docs and tests/governance are green at 365 passed, and the derived documentation pair is regenerated to a fixed point.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: b851b660c4
i️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
P1 Badge Restore the original decision before superseding it
Fresh evidence at this head is that the replacement now explicitly says the 2026年08月09日 body was removed; retaining only its heading and a summary still rewrites this append-only log and erases the original claims about the workspace, register, and byte-exact migration evidence. Keep that entry intact and add the 2026年08月20日 supersession after it, as the repository contract requires.
AGENTS.md reference: AGENTS.md:L52-L58
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
P2 Badge Retain the P9 success criteria in the roadmap
Fresh evidence at this head is that the added summary still defines only MERGED_P9_OPEN, while the conclusion below invokes the unavailable archived P9 definition and the roadmap later uses that outcome to decide what may start. The deleted P9 rule additionally required host migration, native workflow binding, and a successful post-fast-forward protected run before yielding MERGED_CHECKPOINT; without those success criteria, a repository-only reader cannot verify when this residual is closed, so retain that boundary alongside this summary.
AGENTS.md reference: AGENTS.md:L44-L50
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
P1 Badge Make the privacy boundary match what remains published
For anyone with access to this repository, this new privacy claim is not true: the current file still publishes verbatim owner development directives and working-relationship rules in the preceding entries (for example, the PR and publication instructions at lines 330-387), and the deleted obligations and campaign archive remain retrievable from the parent commit with ordinary Git history. If these materials are genuinely private, deleting them only from the tip is insufficient; otherwise narrow this decision so it does not assert that retained and historically accessible material is unpublished.
AGENTS.md reference: AGENTS.md:L546-L549
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
P3 Badge Remove the dangling preposition in the README
When a reader reaches the primary English project overview, this edit now reads "campaign records in archived outside this repository," which is grammatically incomplete and makes the archive-location statement unnecessarily hard to parse. Remove the surviving in or rephrase the parenthetical as "campaign records archived outside this repository."
Useful? React with 👍 / 👎.
Owner, 2026年08月20日: no internal names in this repository, only explicit understandable ones. An identifier like OB-012 names nothing to a reader who cannot open the register it belongs to, and that register is not published here. Worse, it LOOKS like a name, so a reader stops there instead of asking what the thing is. Seven places in ROADMAP.md and the two documentation-skill files now say what the direction requires -- moving the build host and binding the workflow natively; using Simplified Technical English for project documentation; building the in-repository adaptation skill -- and none of them says a number. The trigger comment beside phase-1-status keeps the full statement of what it defers, which is what stops the token being flipped and the work stranded. That was the reason the number was there, and the statement does the job the number could not.
test1card
commented
Aug 20, 2026
@codex review
Head is 3d16c56fdb54d76372900d1ae33446d023507226. This head reverses part of the previous round on the owner's instruction, and the reversal is worth stating plainly rather than letting it look like drift.
Your finding asked me to keep the OB-002 identifier beside the phase-1-status trigger so completing that phase could not silently strand the deferred direction. I did. The owner then directed that this repository carry no internal names at all, only explicit understandable ones.
The finding's concern is still met, by the other half of that fix. An identifier like OB-002 names nothing to a reader who cannot open the register, and the register is not published here — so it looks like a name and stops the reader from asking what the thing is. What stops the direction being stranded is the full statement of what it defers, which is beside the trigger and stays. Seven places now say what the direction requires — moving the build host and binding the workflow natively; using Simplified Technical English for project documentation; building the in-repository adaptation skill — and none says a number.
What this change does NOT touch, measured on this branch: the OC- row keys of docs/OPEN_CELLS.md (224 occurrences across 26 files, and the freshness guards match on the literal row prefixes), and the internal codename that appears in eight tracked paths and about 195 occurrences, all of which the machinery reads. Both are the same direction and both are larger changes with a sequencing decision behind them.
tests/docs and tests/governance are green at 365 passed, and the derived documentation pair is regenerated to a fixed point.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 3d16c56fdb
i️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
P2 Badge Record the no-identifier decision consistently
At this head, the latest ROADMAP change follows the owner's instruction to remove internal obligation identifiers, but this authoritative owner decision still requires agents to retain those identifiers wherever a deferred direction exists. The conflict is already concrete: repository-wide search still finds the now-unresolvable OB-006 in docs/OPEN_CELLS.md, tests/release/test_whole_tree_artifact_freshness.py, and several tools/ci_* comments. Because future agents must treat this log as a durable source of owner direction, they can legitimately reintroduce the identifiers that this round removed; append a superseding decision that records the no-identifiers instruction and replace the remaining obligation references with explicit descriptions.
AGENTS.md reference: AGENTS.md:L52-L58
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
P3 Badge Split the changed skill boundary into short instructions
When this edited bullet is treated as new or changed project documentation, it violates the documentation contract it is meant to enforce: the sentence contains 46 words, while the Core Rewrite Rules in this same file limit instructions to 20 words and require one instruction per sentence, and ROADMAP.md states that new and changed English text must conform immediately. Split the prohibition from its repository-specific rationale so the checked-in binding skill does not demonstrate a direct counterexample to its own required form.
Useful? React with 👍 / 👎.
Five review findings, and one of them says my own text was untrue. THE PRIVACY CLAIM WAS WRONG, AND I CHECKED IT RATHER THAN ARGUING. Removing a file from the current tree does not remove it from the repository: the campaign archive is still 1721 lines away through ordinary git history, and after this merges those commits stay ancestors of master. Entries above the new one also still publish owner directions verbatim. So the decision now says what it actually is -- a decision about what is WRITTEN here from now on -- and states plainly that closing the rest would take a history rewrite or a change of repository visibility, both of which are outward acts and belong to the owner. THE APPEND-ONLY ENTRY IS RESTORED INTACT. I had removed its body and kept the heading, which still rewrites an append-only log. It is back byte-faithful, with the supersession appended after it where the contract says supersession goes. That created a real conflict with the dead-path guard, because the restored entry names a file this same change removes. **I did not edit the record to make the check pass** -- an append-only log whose entries are rewritten to satisfy a check stops being a record. The entry is fenced with the byte-faithful markers the documentation skill already defines, and the guard now skips those fences for the dead-path scan and for nothing else. Proved both ways: a dead path in ordinary text still reddens; the same path inside a fence does not. THE NO-IDENTIFIER INSTRUCTION IS NOW A DECISION, not only an edit. Without that, this log still told a future agent to keep identifiers, so the next agent could legitimately put back what this round removed. The entry also records what the retrofit has NOT reached and how large it is, so the remaining identifiers read as work in progress rather than as a contradiction. THE ROADMAP KEEPS THE WHOLE TERMINAL BOUNDARY. The summary defined only the open outcome, so a repository-only reader could see the conclusion but not the criteria that close it. All three conditions are stated. AND TWO SMALLER ONES: the README sentence lost a preposition and read "campaign records in archived outside this repository"; and the skill's own exemption bullet was a 46-word sentence inside a file whose rules limit an instruction to 20 words and one instruction per sentence. It is four short instructions and a separate rationale now. `tests/docs` is green at 68 passed.
test1card
commented
Aug 23, 2026
@codex review this pull request at head 3e3b770285fcdc4eae074f1f16fd641ca7a67d13.
This head has been sitting without a review request since it was pushed; the last review names an earlier commit. Nothing has changed on the branch since. Zero review comments are anchored to this commit.
This is the privacy repair: it removes the internal register and the campaign archive from the public tree. Please check that nothing it leaves behind still names internal-only material, and that no reference it removes was load-bearing for something the repository still reads.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 3e3b770285
i️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
P2 Badge Remove the remaining private scratchpad dependencies
In a fresh or public checkout, this revised startup block still directs agents at line 159 to read scratchpad/montana/exec/implementation_agent_doc_requests.md, while a repository-wide search finds 11 tracked references to scratchpad/montana and no corresponding tracked files; production docstrings and the refactor report likewise point readers to that private campaign layer. Consequently the repair still names internal-only material and makes unavailable external records inputs to repository work; replace those references with the necessary in-tree rationale or remove them.
AGENTS.md reference: AGENTS.md:L76-L80
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
P2 Badge Correct the measured campaign-name inventory
At this exact head, git grep -Il Montana returns 26 tracked paths and a case-insensitive search returns 51, not eight; examples include PROJECT_STATUS.md, AGENTS.md, production modules, tests, and tools. Because this newly recorded measurement scopes the remaining privacy retrofit, understating it can cause the cleanup to stop while many campaign-name references remain, and the claim also lacks the exact command and immutable-object binding required for a current count.
AGENTS.md reference: AGENTS.md:L445-L450
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
P2 Badge Add a regression that prevents republishing the private files
If a later change restores docs/OBLIGATIONS.md or the campaign archive, this modified test remains green because it checks only whether referenced paths exist; once either private file is restored, its references resolve, and no other repository guard rejects the publication. The change describes the prior publication as a confirmed, machine-testable privacy mistake, so add a negative-control regression that rejects those private paths or equivalent tracked register/archive material rather than relying on deletion at this one head.
AGENTS.md reference: AGENTS.md:L358-L365
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
P2 Badge Keep the binding documentation scope in the repository
For a repository-only documentation change, this rewrite makes the governing scope unavailable: the deleted OB-012 row was the only place specifying that client-facing and agent-layer English were governed while Russian text, verbatim evidence, archives, and changelog history were exempt. The remaining roadmap still says new text conforms "per the obligation" and later points readers to "its disposition list," but neither source is now readable here, so an agent cannot determine what must be rewritten or protected from rewriting. Retain that public scope and its dispositions in an authoritative in-tree document without the private register identifier.
AGENTS.md reference: AGENTS.md:L44-L58
Useful? React with 👍 / 👎.
This repository is an instrument. It publishes the instrument and nothing else. All tracked references to the private working area are removed, including tool inputs and outputs and the code, test and documentation mentions of it. The correction row is bound to an immutable head rather than to a moving count: at 3e3b770 the case-sensitive count is 26 and the case-insensitive count is 51. A documentation-gate regression now refuses the restoration of the removed register, the campaign archive, or any private reference, so this cannot come back by accident. The roadmap states the binding English-documentation scope and its exemptions in the tree, rather than leaving them implied. Deletion does not undo history: the removed content stays readable through git history, and closing that needs a history rewrite or a visibility change, which are the owner's to make. Written with AI assistance.
test1card
commented
Aug 27, 2026
@codex review this pull request at head 1647b2f63781067a8f4f73ebc0ed07b5f8fd0834.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 1647b2f637
i️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
P1 Badge Stop granting coverage without the preserved review report
This run still sets whole_file=True, so when a current file matches the recorded commit, the loop adds it to whole_file["fable"] even though the evidence field now confirms that no report is available for verification. Running this ledger can therefore grant whole-file coverage to the 26 assistant/reporting paths from an unavailable transcript; remove the run from coverage, as was done for WORKTREE_REVIEW_RUNS, or validate a persisted hash-bound report before applying it.
AGENTS.md reference: AGENTS.md:L163-L165
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
P2 Badge Reject renamed copies of the private documents
Fresh evidence at this head is that the new publication guard checks only the two former exact path names. If the obligation register or campaign archive is restored as, for example, docs/directions.md, and it contains no old scratchpad reference, _private_publication_violations() returns no violation, so the regression stays green despite republishing the same private material. Add a negative control and a content/root-level invariant that survives renaming rather than treating the deleted filenames as the privacy boundary.
AGENTS.md reference: AGENTS.md:L361-L365
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
P2 Badge Restrict the verbatim exemption to immutable records
Applying this substitution to every tracked Markdown file creates an unrestricted opt-out from the dead-path guard: a newly added or edited document can wrap a misspelled backtick path in the generic verbatim fences, and the entire block disappears before validation. Nothing checks that the block is an immutable historical record rather than current instructions, so the gate can remain green while actionable documentation points to a missing file; constrain the exemption to the intended append-only source/section and add a negative control for fenced current prose.
AGENTS.md reference: AGENTS.md:L361-L365
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
P2 Badge Keep audit ledgers in the evidence review lane
Moving the generated ledger to .audit-run changes its classification without updating _review_lane: because the path ends in .md and .audit-run/ is not recognized, both this output and the dual-review ledger are now assigned to R09 (canonical product/operator documentation) instead of the former R10 historical/generated-evidence lane. The generators explicitly inventory these output paths, so their review maps now misstate non-authoritative audit evidence as canonical documentation; classify .audit-run/ as evidence before the generic Markdown rule.
AGENTS.md reference: AGENTS.md:L76-L80
Useful? React with 👍 / 👎.
Uh oh!
There was an error while loading. Please reload this page.
This repository is an instrument utility. The working relationship, the plans and the development directives are private and are not published here.
Removed, chosen by measurement rather than by feel. Both documents have ZERO references from
tests/,governance/and.github/, so they are the part of this direction that costs nothing to carry out today:docs/OBLIGATIONS.md(17 782 bytes) — the register of owner directions. Removed rather than emptied, because an emptied register is still a statement about how the owner works.docs/campaigns/MONTANA_CAMPAIGN_ARCHIVE.md(137 300 bytes) — campaign coordination history.Both files were copied out byte-exactly and compared by hash before either was removed. Nothing is lost; it is simply not published.
Every pointer to them is removed as well, and no replacement says where they went — naming the place is the same mistake in a smaller font. Fourteen pointers across
PROJECT_STATUS.md,README.md,ROADMAP.md,docs/DECISIONS.md,docs/MONTANA_IMPLEMENTATION_AGENT_SPEC.mdand the two ASD-STE100 skill files now read "the obligations register", "kept outside this repository", or nothing at all. The obligation identifiers stay, because a reader who meets OB-012 in a comment needs to know it is a real thing with a name; what disappears is the path. Thephase-1-statustoken that a trigger depends on is untouched, and its comment still forbids rewording it.What is deliberately NOT in this change, and it is the larger half. The same direction reaches
ROADMAP.md,AGENTS.md,docs/DECISIONS.md,docs/OPEN_CELLS.md,docs/CLAIM_CORRECTIONS.mdandgovernance/agent_preventions.yaml. Those carry 27 references from tests, governance and continuous integration between them, and they are what the required checks read. Moving them is a separate change with a sequencing decision behind it, and it is not smuggled into this one.Documentation only, no source change.
tests/docsandtests/governanceare green at 364 passed, the derived documentation pair is regenerated to a fixed point, and a repository-wide search for the removed paths returns nothing.This change was prepared with AI assistance.