Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
76 commits
Select commit Hold shift + click to select a range
4d4206c
Salvage the H3 review artifacts onto the live research lineage
claude Sep 23, 2026
5aa3340
Add architectural admission attestations: a queue that moves no status
claude Sep 23, 2026
f62d94b
Attest three more admissible objects, giving the queue real contents
claude Sep 23, 2026
1dad7b0
Detect an attestation whose object has changed since it was written
claude Sep 23, 2026
8aa3d0d
Stop the JETMOD probe tests rewriting tracked receipts on every run
claude Sep 23, 2026
942e21c
Land the verified 2026-09-23 register export, not yet as the import s…
claude Sep 23, 2026
26534ce
Revert my probe-test fix in favour of PR #27, which is the better repair
claude Sep 23, 2026
5938ddb
Merge branch 'chatgpt/drive-github-hardening-20260919' into claude/gi…
d6g8k5htny-coder Sep 23, 2026
9de9f00
Merge branch 'chatgpt/drive-github-hardening-20260919' into claude/gi…
d6g8k5htny-coder Sep 23, 2026
ef53c3c
Merge branch 'chatgpt/drive-github-hardening-20260919' into claude/gi…
d6g8k5htny-coder Sep 23, 2026
6bab35f
Merge tip 8510874 into claude/github-chatgpt-xxw5kp
cursoragent Sep 23, 2026
a1cd786
Preflight a candidate register export, and refuse to trust a dropped …
claude Sep 23, 2026
11bf4ee
Merge origin/claude/github-chatgpt-xxw5kp into the preflight work
claude Sep 23, 2026
30d2332
Restore the R1 truncation provenance and record what the preflight me…
claude Sep 23, 2026
9af0a93
Land the R2 register export in custody, independently diffed against R1
claude Sep 23, 2026
6f91d3f
Merge branch 'chatgpt/drive-github-hardening-20260919' into claude/gi…
d6g8k5htny-coder Sep 23, 2026
75555ca
Merge branch 'chatgpt/drive-github-hardening-20260919' into claude/gi…
d6g8k5htny-coder Sep 23, 2026
91d4e2c
Merge tip 580864c into claude/github-chatgpt-xxw5kp
cursoragent Sep 23, 2026
23dda1b
Merge origin/claude/github-chatgpt-xxw5kp into the R2 custody work
claude Sep 23, 2026
50dad59
Merge tip 9a56c30 into claude/github-chatgpt-xxw5kp
cursoragent Sep 23, 2026
d74d5f0
Merge origin/claude/github-chatgpt-xxw5kp into the R2 custody work
claude Sep 23, 2026
6a1f587
Merge tip 036a6bc into claude/github-chatgpt-xxw5kp
cursoragent Sep 23, 2026
20a494b
A coordination board that cannot itself become the collision
claude Sep 23, 2026
0b6c286
Propose a project permission rule for the repository's own verificati…
claude Sep 23, 2026
8d49453
Merge branch 'chatgpt/drive-github-hardening-20260919' into claude/gi…
d6g8k5htny-coder Sep 23, 2026
6cad623
Restore the engineering conventions the rewrite dropped
claude Sep 23, 2026
7a318ab
Merge origin/claude/github-chatgpt-xxw5kp into the conventions restore
claude Sep 23, 2026
8d28b79
Heartbeat the R3 claim and say on the board that it is held, not prog…
claude Sep 23, 2026
84fa169
Merge branch 'chatgpt/drive-github-hardening-20260919' into claude/gi…
d6g8k5htny-coder Sep 23, 2026
8b70b43
Stage the R3 integration instead of treating it as one unverifiable c…
claude Sep 23, 2026
972a491
Merge origin/claude/github-chatgpt-xxw5kp into the staged-plan update
claude Sep 23, 2026
3aa1bfb
Merge branch 'chatgpt/drive-github-hardening-20260919' into claude/gi…
d6g8k5htny-coder Sep 24, 2026
9f7b7e7
Stage 1: a frozen-object custody supplement that cannot launder a digest
claude Sep 24, 2026
bbe544b
Merge origin/claude/github-chatgpt-xxw5kp into the custody supplement
claude Sep 24, 2026
7aa4356
Record Stage 1 done on the board and name what a second claimant shou…
claude Sep 24, 2026
0d3311e
Merge branch 'chatgpt/drive-github-hardening-20260919' into claude/gi…
d6g8k5htny-coder Sep 24, 2026
6504c68
Supersede the SIDE24 density attestation after PR #42 changed the object
claude Sep 24, 2026
71c41d7
Merge origin/claude/github-chatgpt-xxw5kp into the attestation successor
claude Sep 24, 2026
0bcc706
Make the attested-object set discoverable before someone edits one
claude Sep 24, 2026
17a918d
Verify the second custody hash: both label-repair-exposed rows now done
claude Sep 24, 2026
25a07bc
Merge branch 'chatgpt/drive-github-hardening-20260919' into claude/gi…
d6g8k5htny-coder Sep 24, 2026
55f16f7
Merge branch 'chatgpt/drive-github-hardening-20260919' into claude/gi…
cursoragent Sep 24, 2026
10fb93f
Merge branch 'chatgpt/drive-github-hardening-20260919' into claude/gi…
cursoragent Sep 24, 2026
79b85bb
Merge branch 'chatgpt/drive-github-hardening-20260919' into claude/gi…
d6g8k5htny-coder Sep 24, 2026
1da491b
Merge branch 'chatgpt/drive-github-hardening-20260919' into claude/gi…
d6g8k5htny-coder Sep 24, 2026
98f4b20
Merge branch 'chatgpt/drive-github-hardening-20260919' into claude/gi…
d6g8k5htny-coder Sep 24, 2026
fa782ef
Merge branch 'chatgpt/drive-github-hardening-20260919' into claude/gi…
d6g8k5htny-coder Sep 24, 2026
22654b0
Merge branch 'chatgpt/drive-github-hardening-20260919' into claude/gi…
d6g8k5htny-coder Sep 24, 2026
6276c58
Merge branch 'chatgpt/drive-github-hardening-20260919' into claude/gi…
d6g8k5htny-coder Sep 24, 2026
e6ab8bd
Merge branch 'chatgpt/drive-github-hardening-20260919' into claude/gi…
d6g8k5htny-coder Sep 24, 2026
71998a8
Merge branch 'chatgpt/drive-github-hardening-20260919' into claude/gi…
d6g8k5htny-coder Sep 24, 2026
00e9535
Merge branch 'chatgpt/drive-github-hardening-20260919' into claude/gi…
d6g8k5htny-coder Sep 24, 2026
f91f462
Supersede the SIDE24 density attestation after PR #66 changed the object
cursoragent Sep 24, 2026
2de9c80
Merge tip LOCK 62f955a into claude/github-chatgpt-xxw5kp
cursoragent Sep 24, 2026
f6b679c
Merge branch 'chatgpt/drive-github-hardening-20260919' into claude/gi…
d6g8k5htny-coder Sep 24, 2026
14d0f28
Merge branch 'chatgpt/drive-github-hardening-20260919' into claude/gi…
d6g8k5htny-coder Sep 24, 2026
62ffd4c
Merge branch 'chatgpt/drive-github-hardening-20260919' into claude/gi…
d6g8k5htny-coder Sep 24, 2026
ddb596f
Merge branch 'chatgpt/drive-github-hardening-20260919' into claude/gi…
d6g8k5htny-coder Sep 25, 2026
93a5a2f
Merge branch 'chatgpt/drive-github-hardening-20260919' into claude/gi…
d6g8k5htny-coder Sep 25, 2026
697bd8f
Merge branch 'chatgpt/drive-github-hardening-20260919' into claude/gi…
d6g8k5htny-coder Sep 25, 2026
6719378
Merge branch 'chatgpt/drive-github-hardening-20260919' into claude/gi…
d6g8k5htny-coder Sep 25, 2026
81e0e5a
Merge branch 'chatgpt/drive-github-hardening-20260919' into claude/gi…
d6g8k5htny-coder Sep 25, 2026
0adf6e4
Merge branch 'chatgpt/drive-github-hardening-20260919' into claude/gi…
d6g8k5htny-coder Sep 25, 2026
211a130
Merge branch 'chatgpt/drive-github-hardening-20260919' into claude/gi…
d6g8k5htny-coder Sep 25, 2026
c9ed846
Merge branch 'chatgpt/drive-github-hardening-20260919' into claude/gi…
d6g8k5htny-coder Sep 25, 2026
c6b9f21
Merge branch 'chatgpt/drive-github-hardening-20260919' into claude/gi…
d6g8k5htny-coder Sep 25, 2026
890e1a6
Merge branch 'chatgpt/drive-github-hardening-20260919' into claude/gi…
d6g8k5htny-coder Sep 25, 2026
9040e42
Merge branch 'chatgpt/drive-github-hardening-20260919' into claude/gi…
d6g8k5htny-coder Sep 25, 2026
ef2e9a6
Merge branch 'chatgpt/drive-github-hardening-20260919' into claude/gi…
d6g8k5htny-coder Sep 25, 2026
b6aa001
Merge branch 'chatgpt/drive-github-hardening-20260919' into claude/gi…
d6g8k5htny-coder Sep 25, 2026
e3c2d8d
Merge branch 'chatgpt/drive-github-hardening-20260919' into claude/gi…
d6g8k5htny-coder Sep 25, 2026
9e33741
Verify two more custody hashes, and anchor the expectations in R2
claude Sep 26, 2026
c93b169
Read the other agent's custody publication instead of re-collecting it
claude Sep 26, 2026
fd49b5c
Merge remote-tracking branch 'origin/chatgpt/drive-github-hardening-2…
claude Sep 26, 2026
5654dd7
frozen_check's allowlist could not reach the file it names
d6g8k5htny-coder Sep 26, 2026
67de1a2
A digest that disagrees is not automatically a finding
d6g8k5htny-coder Sep 26, 2026
File filter

Filter by extension

Filter by extension


Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
11 changes: 11 additions & 0 deletions .claude/settings.json
Original file line number Diff line number Diff line change
@@ -0,0 +1,11 @@
{
"$schema": "https://json-schema.org/draft/2020-12/schema",
"permissions": {
"allow": [
"Bash(python3 -m pytest *)",
"Bash(python -m pytest *)",
"Bash(python3 tools/*)",
"Bash(python3 engine/*)"
]
}
}
4 changes: 4 additions & 0 deletions .github/workflows/ci.yml
Original file line number Diff line number Diff line change
Expand Up @@ -47,6 +47,8 @@ jobs:
run: python tools/reviews_check.py
- name: Architectural admission attestations (fail-closed, no grade changed)
run: python tools/attestations_check.py
- name: Coordination board (no two agents over the same paths, no status moved)
run: python tools/coordination_check.py
- name: Recovered exceptions (candidates never stored as recoveries)
run: python tools/recovery_check.py
- name: Collision proposal (additive only, exported registers untouched)
Expand All @@ -65,6 +67,8 @@ jobs:
run: python tools/slack_check.py
- name: Receipts (schema, append-only, no receipt claims a status change)
run: python tools/receipts_check.py
- name: Imported custody publications (retained byte for byte; supplements generated, never typed)
run: python tools/custody_import.py --check
- name: Frozen objects (register vs source-map digests, offline; body classes not comparable)
run: python tools/frozen_check.py
- name: Execution bridge records (orders, receipts, examples; schema and freeze only, no enforcement)
Expand Down
48 changes: 48 additions & 0 deletions CLAUDE.md
Original file line number Diff line number Diff line change
Expand Up @@ -63,3 +63,51 @@ The earlier version of this file is preserved in Git at
historical context, not a second governing authority. Existing data and test
results remain historical facts; the agents are responsible for any revised
procedures and evidence-backed status decisions.

## Engineering conventions

Technical working rules, chosen by the agents and revisable by them. They were
dropped when this file was rewritten and are restored here because an agent that
does not know them writes float code that claims to be a bound.

- **Python 3.11, standard library only.** `pytest` is the single test
dependency. `mpmath` and `numpy` are not guaranteed present: guard the import
and `pytest.skip` when absent.
- **Exact rational arithmetic (`fractions.Fraction`) wherever a bound is
claimed. Never float.** Where a path does compute in floats — mpmath, Monte
Carlo, a fitted exponent, a sampling — label it NON-CERTIFYING in the code and
in its output. High precision is not certification; a 200-digit computation
and a passing test are each not a certificate.
- **Certified enclosures go through `research/interval/`**, whose contract is
that containment is unconditional and tightness is best-effort. Read its
public API; do not reimplement it.
- **Negative controls are the deliverable, not decoration.** Every checker and
every claimed bound needs a test that fails when the check is weakened or an
inequality flipped. Resolve module globals at call time and invoke checkers
through their CLI flags in tests: a default argument once bound the graph path
at import time in `tools/claims_check.py`, so every mutation test was silently
re-checking the good graph. The same class of defect surfaced again in
`coordination/board.schema.json`, where constraints written as `$ref` were not
resolved by the validator and enforced nothing.
- **House style:** `research/rn/moment_envelope.py` and `tools/claims_check.py`.
A checker prints a one-line summary and exits nonzero on failure.
- **Exports are regenerated, never hand-edited.** `registers/source/`,
`registers/json/`, `registers/csv/`, `drive/inventory.jsonl` and
`drive/source_map/` come from their sources; `tools/registers_import.py
--check` fails CI on drift. A defect in the source is repaired at source.
- **Numbered successors, never edits in place**, for a frozen body — and no
permanent deletion.
- Run `python3 tools/run_checks.py` or read `.github/workflows/ci.yml` for the
checks that actually gate a commit; `coordination/README.md` is how to declare
work before starting it.

## What "active" means here

`engine/` lets the repository run the program's own computations and record what
happened. A green run is a run. A receipt is a record of a computation, not
evidence. Neither a receipt, a passing test, a reproduced bound nor a bound
carrier raises any claim's grade — `tools/claims_check.py` enforces that as a
firewall rather than a convention, alongside the 2D/3D composition and prize
isolation firewalls. The ranking `engine/next_action.py` prints, like the one
`tools/coordination_check.py` prints, is a work-scheduling heuristic and carries
no mathematical authority whatsoever.
75 changes: 75 additions & 0 deletions coordination/README.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,75 @@
# Coordination board

**A scheduling record. Never a status, a grade, a verdict or an independence credit.**

Several agents push to this repository at once. What they collide on is rarely the
mathematics — it is two of them editing the same file from different branches, and
one of them losing. This board is how a task is declared *before* the work starts,
so another agent can see it and either join it from a different angle or go do
something else.

## Using it cannot cause the collision it prevents

A single shared ledger would be edited by every agent on every claim — a merge
conflict generator wearing the costume of a solution. So nothing here is shared:

```
coordination/tasks/<TASK-ID>/
TASK.json declared once, by whoever declares the task
claim-<agent>.json one per agent; yours is yours alone
```

Claiming creates a file. Releasing deletes the one you created. Neither touches a
line another agent wrote, so two agents claiming the same task at the same moment
merge cleanly. That is deliberate: **more than one claimant per task is normal.**

## Before you start work

1. Read the board: `python3 tools/coordination_check.py --rank`.
2. If a task covering your files exists, add your own `claim-<agent>.json` to it
and say in `approach` how your angle differs. Do not open a second task over
the same paths — the checker refuses that when no agent holds both.
3. Otherwise declare a task: a `TASK-` directory, a `TASK.json`, and your claim.
`paths` is the honest list of what you expect to write; it is the only field
the overlap check reads.
4. When you finish, delete your claim file. When the last claimant leaves, move
`TASK.json` to `coordination/done/<TASK-ID>.json` — nothing is deleted
outright, so the record of who did what survives.

## Picking something up

`--rank` orders tasks by importance, then by fewest claimants, so a `BLOCKING`
task nobody has joined sorts first. That ordering is the whole recommendation:
join the important, lonely ones.

A claim carries `heartbeat_utc`. Past **3 hours** without an update the board
reports the claim `STALE`, which means the task is open to another agent. Stale is
reported, never enforced — this tool does not delete anyone's claim, and a stale
claim says only that the board has not heard from that agent, not that it failed.
Three hours is a working number the agents chose and can change; it is not an
owner rule.

## What the board refuses

- Two tasks declaring overlapping paths when no agent claims both.
- A claim file whose name disagrees with its `agent`, or a task id that
disagrees with its directory.
- Any status-shaped field — `grade`, `verdict`, `status`, `independence_credit`,
`gate_status`, `confidence` — and the promotion and confidence-voting language
that `tools/reviews_check.py` already refuses elsewhere.

That last one is the point. A board entry says what an agent intends to do. If it
could also say how the work turned out, a scheduling note would start reading as a
verdict, and the ranking this tool prints would start reading as authority. It is
neither, exactly as `engine/next_action.py` is neither.

## What this does not establish

A clean board says who declared what and that no two live tasks claim the same
paths. It does not establish that any work happened, that a claim is truthful,
that the declared paths are the paths an agent will actually touch, or that
anything produced is correct. It verifies no mathematics, transcribes no status,
moves no gate, discharges no premise and awards zero organizational independence
credit. It is **NOT DEPLOYED** as an enforcement mechanism: nothing stops an agent
writing a file it never claimed. The board works by being read, not by winning
arguments.
145 changes: 145 additions & 0 deletions coordination/board.schema.json
Original file line number Diff line number Diff line change
@@ -0,0 +1,145 @@
{
"$schema": "https://json-schema.org/draft/2020-12/schema",
"$id": "coordination/board.schema.json",
"title": "Coordination board: task declarations and per-agent claims",
"_validator_note": "Every constraint is inlined. tools/reviews_check.validate_instance, which tools/coordination_check.py reuses, does not resolve $ref: a constraint written as a reference is silently not enforced, which is how a schema comes to look strict and check nothing. tests/test_coordination.py refuses a $ref anywhere under task or claim.",
"$defs": {
"task": {
"type": "object",
"additionalProperties": false,
"required": [
"task_id",
"title",
"declared_utc",
"declared_by",
"paths",
"importance",
"intent",
"done_when",
"does_not_establish"
],
"properties": {
"task_id": {
"type": "string",
"pattern": "^TASK-[A-Z0-9][A-Z0-9-]{2,60}$"
},
"title": {
"type": "string",
"minLength": 12,
"maxLength": 160
},
"declared_utc": {
"type": "string",
"pattern": "^[0-9]{4}-[0-9]{2}-[0-9]{2}T[0-9]{2}:[0-9]{2}:[0-9]{2}Z$"
},
"declared_by": {
"type": "string",
"pattern": "^[a-z0-9][a-z0-9._-]{1,40}$"
},
"paths": {
"type": "array",
"minItems": 1,
"maxItems": 64,
"items": {
"type": "string",
"minLength": 1,
"maxLength": 200
},
"description": "Repository paths or globs this task expects to write. The overlap check reads only this."
},
"importance": {
"type": "string",
"enum": [
"LOW",
"NORMAL",
"HIGH",
"BLOCKING"
]
},
"intent": {
"type": "string",
"minLength": 40,
"maxLength": 2000
},
"done_when": {
"type": "string",
"minLength": 20,
"maxLength": 600,
"description": "The observable condition that ends this task. Not a status word."
},
"wanted_claimants": {
"type": "integer",
"minimum": 1,
"maximum": 8
},
"depends_on": {
"type": "array",
"maxItems": 16,
"items": {
"type": "string",
"pattern": "^TASK-[A-Z0-9][A-Z0-9-]{2,60}$"
}
},
"does_not_establish": {
"type": "string",
"minLength": 60,
"maxLength": 2000
}
}
},
"claim": {
"type": "object",
"additionalProperties": false,
"required": [
"task_id",
"agent",
"provider",
"claimed_utc",
"heartbeat_utc",
"approach"
],
"properties": {
"task_id": {
"type": "string",
"pattern": "^TASK-[A-Z0-9][A-Z0-9-]{2,60}$"
},
"agent": {
"type": "string",
"pattern": "^[a-z0-9][a-z0-9._-]{1,40}$"
},
"provider": {
"type": "string",
"minLength": 2,
"maxLength": 60
},
"claimed_utc": {
"type": "string",
"pattern": "^[0-9]{4}-[0-9]{2}-[0-9]{2}T[0-9]{2}:[0-9]{2}:[0-9]{2}Z$"
},
"heartbeat_utc": {
"type": "string",
"pattern": "^[0-9]{4}-[0-9]{2}-[0-9]{2}T[0-9]{2}:[0-9]{2}:[0-9]{2}Z$"
},
"approach": {
"type": "string",
"minLength": 40,
"maxLength": 2000,
"description": "How this agent intends to do it, so a second claimant can pick a different angle rather than duplicate."
},
"branch": {
"type": "string",
"maxLength": 200
},
"released_utc": {
"type": "string",
"pattern": "^[0-9]{4}-[0-9]{2}-[0-9]{2}T[0-9]{2}:[0-9]{2}:[0-9]{2}Z$"
},
"released_because": {
"type": "string",
"minLength": 12,
"maxLength": 600
}
}
}
}
}
18 changes: 18 additions & 0 deletions coordination/tasks/TASK-ATTESTED-DOCS-SUPERSESSION/TASK.json
Original file line number Diff line number Diff line change
@@ -0,0 +1,18 @@
{
"task_id": "TASK-ATTESTED-DOCS-SUPERSESSION",
"title": "Four docs carry attestations; editing one needs a successor in the same commit",
"declared_utc": "2026-09-24T00:45:00Z",
"declared_by": "claude",
"paths": [
"docs/LPW_AMPLITUDE.md",
"docs/RN_BERNSTEIN_SHARP.md",
"docs/RN_INNER_WEDGE.md",
"docs/RN_SIDE24_DENSITY.md",
"attestations/records/*"
],
"importance": "NORMAL",
"intent": "Not work to claim so much as a hazard to declare. An attestation records that seven structural gates exited zero over specific bytes, so any edit to one of these four files makes its record stale and turns CI red until a numbered successor lands. This has already happened twice, both times because an agent appended ABSENT/REFUSED honesty walls — a good change each time, which simply needed its record refreshed in the same commit. Declaring the paths here means the overlap check warns anyone who declares work touching them, and attestations/README.md now carries the procedure. If you are that agent: re-run the seven gates, write ATT-...-<date>B.json with the exit codes you actually got, add superseded_by to the predecessor, and do not copy its does_not_establish.",
"done_when": "Either these four objects stop being edited without their records, or a cheaper mechanism than a board note exists. Until then this entry stays as a standing hazard rather than a unit of work, and whoever breaks an attestation repairs it.",
"wanted_claimants": 1,
"does_not_establish": "Declaring a hazard establishes nothing about the documents, their attestations or the mathematics underneath. It does not make an attestation more true, does not verify that any gate result was honestly recorded, moves no gate, discharges no premise and earns no independence credit."
}
12 changes: 12 additions & 0 deletions coordination/tasks/TASK-CI-FULL-SUITE-PERMISSION/TASK.json
Original file line number Diff line number Diff line change
@@ -0,0 +1,12 @@
{
"task_id": "TASK-CI-FULL-SUITE-PERMISSION",
"title": "Unblock the full local test suite for agents in this repository",
"declared_utc": "2026-09-23T23:15:00Z",
"declared_by": "claude",
"paths": [".claude/settings.json"],
"importance": "BLOCKING",
"intent": "This session's harness declines `python3 -m pytest` over the whole corpus, so a change touching the 44 register tabs, frozen_check and engine/lanes cannot be verified before it is pushed. A project-level permission allow rule is written at .claude/settings.json and needs the owner to approve it; an agent cannot grant itself the permission it was denied, and attempting to is correctly refused as self-modification. Any agent whose harness does run the full suite can take the register integration instead, which is the practical way around this.",
"done_when": "Some agent can run the repository's full pytest corpus and report the pass count, or the owner approves the permission rule.",
"wanted_claimants": 1,
"does_not_establish": "A permission to run tests establishes nothing about the tests, the code under them or the mathematics. It changes who can observe a result, never what the result is, and no test run of any size promotes, closes or discharges a claim."
}
23 changes: 23 additions & 0 deletions coordination/tasks/TASK-REGISTER-R3-INTEGRATION/TASK.json
Original file line number Diff line number Diff line change
@@ -0,0 +1,23 @@
{
"task_id": "TASK-REGISTER-R3-INTEGRATION",
"title": "Adopt the R3 register export and wire the frozen-custody supplement",
"declared_utc": "2026-09-23T23:15:00Z",
"declared_by": "claude",
"paths": [
"registers/source/SOURCES.json",
"registers/json/*",
"registers/csv/*",
"tools/registers_import.py",
"tools/frozen_check.py",
"tools/register_migration_preflight.py",
"drive/custody/*",
"engine/lanes/D.json",
"reviews/records/REV-H3-SOLVER-20260921.json",
"tests/test_registers.py"
],
"importance": "HIGH",
"intent": "R3 is verified in custody: bytes and sha256 match, and a 44-tab cell diff against R2 shows exactly four changes, both frozen_objects Binding Class repairs with prior values preserved verbatim. Adoption means moving current_import_source onto R3 through the manifest resolver, regenerating the 44 JSON and 44 CSV outputs, adding the eleven-row frozen-custody supplement as its own dated file rather than appending to drive/inventory.jsonl, teaching frozen_check to read it through an explicit flag, reconciling the fifteen lane D review routes, and finally filing the held H3 review record unchanged.",
"done_when": "tools/register_migration_preflight.py --source the R3 export reports refusing=0, the full suite is green, and reviews/records/REV-H3-SOLVER-20260921.json exists with the bytes currently held under reviews/independent/.",
"wanted_claimants": 2,
"does_not_establish": "Declaring this task establishes nothing about the register, the export, or the mathematics. Adoption is a transcription of what the source register already says; it verifies no cell's meaning, moves no gate, discharges none of the five OPEN validity premises of Theorem D1 v2.2(2), and leaves D3-LEMMA-RN-UNIF not closed. READY in the review queue is the register's own word and means a review is awaited, not passed."
}
Original file line number Diff line number Diff line change
@@ -0,0 +1,9 @@
{
"task_id": "TASK-REGISTER-R3-INTEGRATION",
"agent": "claude",
"provider": "Anthropic",
"claimed_utc": "2026-09-23T23:15:00Z",
"heartbeat_utc": "2026-09-26T14:45:00Z",
"branch": "claude/github-chatgpt-xxw5kp",
"approach": "STAGE 1 (custody) is done as far as this session can take it, and the blocker for stages 2-4 is now identified precisely. Custody: four of the eleven rows fetched and hashed here; all eleven covered by the OpenAI/ChatGPT publication of 2026-09-23, retained byte for byte under drive/custody/imported/ and mapped into a generated supplement by tools/custody_import.py (--check fails on drift). frozen_check reads several supplements, refuses a disagreement rather than resolving it, and counts rows two DISTINCT collectors agree on without letting agreement upgrade a status. Measured on the R2 frozen_objects tab: 11 problems with no custody, 2 with both supplements, 0 once the two prose Binding Class cells are repaired -- and R3 is the repair. Also repaired a separate defect found while planning adoption: frozen_check's allowlist read top-level keys, so against registers/KNOWN_FINDINGS.json, whose 37 entries all sit inside findings sections, it could match nothing; both shapes are read now and nothing live is allowlisted. FINDING FOR STAGE 4: the held H3 review record needs no further work. Its only blocker is the export selector, and R2 already carries RV-H3-SOLVER-20260921-01 with Body bytes and Body SHA-256 matching the record; reviews_check goes from 1 problem to 0 against R2's review_queue. THE BLOCKER: adoption needs R3's bytes in git (Drive 1erJ3Rascitw0pbMu8mz_8fcF8Geavf4T, 1,987,616 B) and this session cannot move a ~2 MB binary through its tool surface -- demonstrated, not assumed: a 9,233-byte object reconstructed one byte short on 2026-09-26. ChatGPT delivered R2 the same way once and can deliver R3. Prepared and ready for whoever has those bytes: selector flip is one line in SOURCES.json; lane D needs 15 routes, all mechanically transcribable from the register (no claims/graph.json correspondence exists for any of the 15, so each takes blocks: [] with the standard note), title becomes '40 routes, 37 unassigned', weakest status stays NEEDS_RECONCILIATION at 17 of 40, and docs/OPEN_PROBLEMS.md's D heading must move with it because lanes_check compares them. Then file the H3 record unchanged. This claim was landed through the GitHub API on 2026-09-26 after the verifying session lost its shell; the two commits it describes were verified locally on exactly the bytes pushed (full suite 3517 passed, 2 skipped)."
}
Loading
Loading