Repository navigation
Conversation
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: bd5223deee
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.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 9db91af7bf
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".
Propose the Amendment-6/7 material pre-authority correction of ADR-0012 as a PROPOSED Route-B candidate (#958), closing R-9DB-01 in source. - ADR-0012: Status APPROVED -> PROPOSED with no current approver or approval date. Every activation epoch references exactly one immutable route-contract revision with a canonical semantic signature; eligibility requires epoch membership and a match to that revision; continuous producer-version coverage with expand/contract and cross-route fan-out; durable IN_EPOCH_CONTRACT_MISMATCH anomalies; field-by-field historical-enrollment validation; atomic database-level revision equivalence; deterministic materializer compatibility; per-revision route health; safe revision withdrawal and route retirement. Superseded route-level-contract wording is reconciled across sections 3.2, 3.3, 3.9, 3.15, 7-12. - decision register: ADR-0012 row PROPOSED; Last updated 2026-10-06. - CHANGELOG: one new PROPOSED correction entry; history preserved. - implementation report: candidate stage, Amendments 6-7 history and candidate validation evidence. ADR-0008, ADR-0010, ADR-0013 and shared control files unchanged. Architecture proposal only: no runtime, migration, provider, deployment, release or activation change.
Record the CTO exact candidate-content approval of 2026-10-06 for the Amendment-6/7 Route-B candidate (head b596536, ADR-0012 blob cd897d5) as durable decision state, as the single approval-state commit Route B permits (#958). - ADR-0012: Status PROPOSED -> APPROVED, Approved by: Javi, CTO, Approval date: 2026-10-06, and the leading section 12 approval-state paragraphs; no architecture text changes (normalized comparison with the candidate blob shows no difference). - decision register: ADR-0012 row approval state only. - CHANGELOG: route-contract correction entry approval-state wording only. - implementation report: approval-state stage, no-drift proof and validation evidence. ADR-0008, ADR-0010, ADR-0013 and shared control files unchanged. Architecture approval only: no runtime, migration, provider, deployment, release or activation change.
Report-only control-record recovery of the Amendment-6/7 approval-state report (#958). The approval-state authorization erroneously required the report to copy the GitHub identifier of the CTO content-approval record into repository source; repository-authoritative Route-B doctrine (docs/engineering/decisions/README.md) prohibits that and takes precedence. - implementation report only: remove the copied identifier, record the content approval through owning issue #958, approval date 2026-10-06, candidate head b596536 and candidate ADR-0012 blob cd897d5, replace the deviation framing with the reconciled control fact, and record the recovery and its validation. ADR-0012 (a7d95ca), its approval state, approver and date, the decision register and CHANGELOG are unchanged. No runtime, migration, provider, deployment, release or activation change.
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 22bd94dc95
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".
Report-only correction of the control-record recovery report (#958, PR #960) after the configured final-head automated review of the recovery head 22bd94d. Control adjudicated its two findings on #958: the topology finding rested on a non-head object and was rejected against the authoritative PR head/parent graph (no lineage change); the transient review-thread-state finding was accepted. - implementation report only: stage-time review-thread inspections are rephrased as historical stage evidence, the live assertion about the number and resolved state of PR #960 review threads is replaced by durable wording that points to PR #960 as the authority for current thread state, the final-head review outcome is recorded substantively without review identifiers, and the correction and its validation are recorded as the current stage. ADR-0012 (a7d95ca), its approval state, approver and date, the decision register and CHANGELOG are unchanged. No runtime, migration, provider, deployment, release or activation change.
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 9b4bb1b022
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".
Two-path correction after the configured final-head automated review of the transient-state correction head 9b4bb1b (#958, PR #960). Both findings were accepted by control; the CTO classified the ADR change as a FACTUAL CLARIFICATION of prior approved ADR-0012 blob a7d95ca. - ADR-0012: replace the unconditional "exactly one durable disposition exists for each (event_id, route_key)" sentence in section 3.2, section 3.5 and invariant 8 with the at-most-one uniqueness rule, each paired with the already-selected rule that an eligible route obligation remains owed until it receives exactly one disposition. No-disposition cases (ineligible events, in-epoch mismatch, refused enrollment, approved divergence), validated enrollment, duplicate prevention, job idempotency, header approval fields and section 12 are unchanged; no architecture decision changes. - implementation report: replace statements about which live PR #960 lifecycle gate is pending with the durable lifecycle model (GitHub, through #958 and PR #960, owns current gate state), record the clarification and its validation, and advance the stage record. Decision register and CHANGELOG unchanged. No runtime, migration, provider, deployment, release or activation change.
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: eb3129f13a
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.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: e2588f7b7e
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.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: e41f377202
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".
| no slot is consumed; | ||
| - case 2, eligible backlog: an eligible obligation may temporarily have no | ||
| disposition while owed and visible (line 360; undispositioned eligible-event | ||
| backlog, line 552); | ||
| - case 3, in-epoch contract |
There was a problem hiding this comment.
When a producer records an in-epoch contract mismatch concurrently with withdrawal of the only selectable revision that could validly enroll it, withdrawal can observe no anomaly, pass the section 3.2 strand check, and commit before the event transaction records the anomaly; the required revision is then unavailable and the anomaly has no approved resolution path. Fresh evidence in this revision is this explicit deferral of the strand-check timing question, while the new atomic-order rule covers revision selection but not anomaly creation. Require anomaly creation and other mutations of the strand predicate to share a database-atomic order with withdrawal, or require a commit-time recheck, so invariant 39 can actually hold.
AGENTS.md reference: AGENTS.md:L350-L361
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
CTO/control disposition: VALID / HIGH / BLOCKING as R-E41F-01. This is a MATERIAL ARCHITECTURAL CORRECTION: the current withdrawal/selection ordering does not make withdrawal's strand-prevention predicate stable against concurrent creation of an unresolved mismatch anomaly (or another already-defined mutation that can turn the predicate unsafe). The implementation report explicitly deferred this timing question. Specification Amendment 10 is recorded on #958 as comment 6066621583. This thread remains unresolved pending fresh independent CRITICAL specification review and a separately authorized Route-B correction.
1 participant
THOTH-ASYNC-01-ADR-01 - ADR-0012 shared async architecture, Amendment 9 final approval-state source
Programme: #957
THOTH-ASYNC-01Owning task: #958
THOTH-ASYNC-01-ADR-01Risk: CRITICAL
This PR carries the current Route-B approval-state version of ADR-0012 under
docs/engineering/decisions/README.md.The architecture decision is APPROVED, but it is not yet repository-authoritative.
Repository authority still requires this exact approved version to pass the remaining
review and merge gates and become reachable from
develop.PR Draft/Ready state, checks, review/thread state, merge authorization and merge state
remain GitHub-owned lifecycle evidence rather than repository-source lifecycle records.
Current exact source identity
Amendment 9 Route-B approval lineage
The configured final-head review of the preceding Amendment-8 approval-state head
identified
R-E258-01: coalescing attachment to an existing not-yet-started jobcould race the first worker claim/start transition. A router could observe the job
as coalescible, the worker could start it, and the router could then attach a new
route disposition from its stale observation.
Specification Amendment 9 is recorded on #958 as comment
6046180884.Its fresh independent CRITICAL specification review is recorded as APPROVED in
#958 comment
6046871411.The content-approved Amendment-9 candidate was:
The one bounded Route-B approval-state commit is the current head:
Independent control inspection proved the final ADR is architecture-equivalent to
candidate blob
0fc33df9...after removing only the authorized approval-staterepresentation.
Effective specification
The current architecture is governed by the #958 base specification plus
Specification Amendments 1-9 and their associated review/control records.
Amendment 9 adds one durable ordering boundary between:
For every logical job kind that permits many-route-to-one-job coalescing:
later attempts never reopen coalescing;
boundary or equivalent database-enforced order;
The architecture defines both race orders:
represent every accepted pre-seal disposition/input under the kind's approved
coalescing semantics;
is committed, the
(event_id, route_key)slot remains unused, and the routeobligation remains owed/visible through the existing error/attention-recovery path.
A rolled-back first claim does not durably seal the job.
The exact schema, SQL, lock and state encoding remain implementation-specification
details. CURRENT_STATE / REVISION_BOUND semantics, job identity, route-disposition
uniqueness, Amendment-8 withdrawal/selection ordering, target-scoped effect
serialization, claim-token fencing and external-write safety remain unchanged.
Current finding disposition
The thread is intentionally not resolved by this metadata reconciliation.
What ADR-0012 continues to establish
ADR-0012 defines the shared PostgreSQL-backed asynchronous event/route/job architecture,
including:
route_key;(event_id, route_key)disposition uniqueness;IN_EPOCH_CONTRACT_MISMATCHanomalies;Repository and programme boundaries
thothowns the generic durable engine and canonical PostgreSQL state.Domain-specific provider/dissemination semantics remain in their owning repositories.
In particular,
thoth-disseminationcontinues to own dissemination behaviour.Cross-repository implementation remains separately decomposed, authorized and reviewed.
No downstream repository may guess an unmerged upstream contract.
Scope and effects
This PR is an architecture/documentation decision.
It does not authorize:
Architecture approval is not implementation authorization.
Cumulative PR footprint
Against
feature/worker @ b44303c214498baf76c9d9a0a7cab374377f9cbe,the PR footprint remains exactly:
ADR-0008 and ADR-0010 carry only the bounded partial-supersession metadata already
approved in this PR. ADR-0013 and shared decision-process/control doctrine are unchanged.
Remaining gate sequence
No merge, programme reliance, runtime implementation, migration, deployment, release or
production activation is implied by this metadata reconciliation.
Tracks #957 and #958.