Repository navigation
Conversation
|
CREATOR: Published the #35 planning checkpoint at
|
aram356
left a comment
There was a problem hiding this comment.
REVIEWER: REQUEST_CHANGES
GitHub refused the formal event with HTTP 422: Review Can not request changes on your own pull request. This is the one-account COMMENT fallback preserving the requested-changes verdict and inline finding, not a formal changes-requested state or approval.
Planning self-review of 5e2e6c7fc4d5757e0c95e646b1fb3125ea126c53 for #35: one P2 inline finding. The deadline contract needs a clock-refreshing host checkpoint and a turn-frozen-clock preparation control before implementation starts.
Reviewed the complete two-document diff and relevant existing protocol/merge, account/privacy/usersync, configuration, resource/lifetime/deadline, SDK and test-ownership surfaces. No other substantive finding was established. New storage/work reservations and provider qualification remain explicit later gates; this review does not approve them.
Verification evidence: GitHub Actions, not an independent reviewer suite run. All 25 hosted checks pass at this head. Inspected output from secure job 134860158817, exit 0:
$ cargo test --locked -p auctioneer-exchange --test source_preflight --test disclosure_policy_diff --test projection_scrub_bound --test privacy_policy_composition
test result: ok. 57 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out; finished in 0.14s
$ cargo test --locked -p auctioneer-bidder -p auctioneer-exchange -p auctioneer-core --all-targets
test result: ok. 309 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out; finished in 2.97s
These are selected actual summaries from the existing matrix, not new lookup tests or a combined full-suite count. No local Cargo/tests/fmt/Clippy/CI scripts or mutation experiments ran, per the user's standing instruction. No source changes or thread resolutions were made.
One-account mode: acting login and PR author are both aram356. The detached reviewer-owned worktree was checked clean and removed after the source review; creator branches/worktrees remain untouched. Next actor: CREATOR to amend the documents and reply in the inline thread, followed by REVIEWER verification. No approval or merge authorization is given.
aram356
left a comment
There was a problem hiding this comment.
REVIEWER: REQUEST_CHANGES — GitHub blocked this review event: HTTP 422 "Review Can not request changes on your own pull request". The acting login and the PR author are both aram356 (one-account mode), so this verdict is recorded as a COMMENT. It is not a formal change request.
Planning review of 5e2e6c7fc4d5757e0c95e646b1fb3125ea126c53 for #35. The two documents are careful and their claims about the code hold up. I am asking for changes because two contracts are left open where an implementer would have to guess, and five more gaps would surface during implementation or at issue closure.
Findings, one thread each
- blocking Endpoint URL-safety rules do not carry the existing configured-endpoint protections, and the development transport for tests is undefined.
- blocking A template-supplied
tmaxhas no defined effect on the bidder-visible timeout, and cannot raise the timeout above the host default. - should-fix The no-disclosure rule conflicts with final-ingress diagnostics that quote values and impression IDs.
- should-fix The no-cache design is not reconciled with the cache, cold/warm, host-store and hot-path measurement wording in #35, #53 and #78.
- should-fix The format-2 hard cut has no deployment order or rollback path.
- should-fix The Task 1 inventory misses fixtures that depend on the unknown-publisher fallback.
- should-fix The Task 9 checkpoint omits the layout-witness owners.
- nit The Task 6 checkpoint omits the exchange owner of
BidRequest::validate(conditional). - nit Secret-store outages are unclassified in the failure table.
- nit
Cache-Control: no-storeis not marked as a client check or a service duty. - nit A non-string account selector result changes a pinned behavior without appearing in section 1.1.
What I checked and found correct
- Every existing file, test target, script and Cargo feature the plan names exists at this head (69 files and 71 test targets, checked by script against
git ls-tree). The five files the plan creates do not exist yet. - Section 1.1's description of current behavior: synchronous
prepare_auction, requiredBidRequest.idandImp.id, the inlineaccountsmap, default fallback for an unknown named account, and account discovery before typed decoding. - The six default account paths and their order match
ACCOUNT_ID_PATHS. - The response-cap order in section 3.2 matches
resolve_policy_budgets.AccountConfighas no response-cap fields. root::projectdropsext.prebid.storedrequest, so the root guard does have to run before projection, as the #232 refresh says.InvalidImpFirstPartyDatais code 15 and HTTP 500. Account-disabled is HTTP 503 for auctions and HTTP 400 on the sync routes.resolve_tmaxtreats absent and zero as the host default and clamps only the deadline.- At the pinned SDK
e9ffaa8:ResponseDecoding::Original, last-write-winsstream_responseandmax_response_bytes,into_bytes_bounded_until, declared-length checks before the head is returned, removal of Connection-nominated headers, no redirect following on plainsend,OutboundCachePolicy::Bypass, and theResponseTooLarge,PayloadTooLargeandBadGatewayvariants. - The acceptance commands correspond to the steps in
test.yml.
Process gates. Section 9 records the user's approval of the spec for planning on 2026-11-09, and the plan has the writing-plans header. The PR description reports advisory spec and plan reviews. It does not name the two Superpowers skills, which the handoff should.
PR description, no thread. It calls auction_data "the existing top-level" policy. No auction_data exists at this head, and the spec adds it. It also still says the head's checks are running.
Verification
- GitHub reports 25 of 25 check runs successful at
5e2e6c7fc4d5757e0c95e646b1fb3125ea126c53. Those are GitHub Actions' executions. I did not open their logs, because the diff is two documentation files (+1,225 lines) and no test exercises them. - My own checks were read-only: the existence script above, and reads of the code and the pinned SDK at the lines linked in each thread.
- I ran no Cargo, tests, fmt, Clippy or CI scripts locally, under the user's standing instruction. No full-suite run of mine exists for this head. What counts as the full suite for this documentation-only PR is the user's decision and has to be settled before any approval.
Threads. This is the first review. No earlier threads existed, so there was nothing to resolve.
Next actor. CREATOR, for a disposition on each of the 11 threads.
|
REVIEWER: Correction to review 5961063456. Its summary says it was the first review and that no earlier threads existed. That was true at my last readback and false when it was submitted.
Next actor is unchanged: CREATOR, for a disposition on each thread. |
|
CREATOR: Ready for document re-review at
|
aram356
left a comment
There was a problem hiding this comment.
REVIEWER: REQUEST_CHANGES — GitHub blocked this review event: HTTP 422 "Review Can not request changes on your own pull request". The acting login and the PR author are both aram356 (one-account mode), so this verdict is recorded as a COMMENT. It is not a formal change request.
Re-review of c84c7753dae67a38bfefa99b7f279493d82e958e for #35. I accepted and resolved 11 of the 12 threads. One remains open: secret-store classification. Its fix keys on an SDK error type that the secret read never returns. My original finding named that type, so the error started with me.
Accepted and resolved
- Frozen-clock checkpoint during expansion, the other REVIEWER session's finding, verified here.
- Endpoint URL-safety rules
- Template
tmaxand the bidder-visible timeout, with all six table rows recomputed againstresolve_tmax. - Diagnostics on expanded documents
- No-cache decision and issue criteria, with #35, #53, #78 and #245 read back.
- Hard-cut deployment and recovery, with the mismatch behavior checked against the code.
- Unknown-publisher fallback inventory
- Layout-witness owners in Task 9
- Exchange and bidder owners in Task 6
Cache-Control: no-storeduty- Non-string selector change in section 1.1
Still open
- Secret-store outcomes. Re-key the rows on
SecretErrorand the twoOkoutcomes of the bounded read. Classify a too-large token as a permanent configuration failure rather than overload, and name the token allowance. Details are in the thread.
Delta checked. git diff 5e2e6c7 c84c7753dae67a38bfefa99b7f279493d82e958e touches only the two documents (+211/-12). 5e2e6c7 is an ancestor of the new head, so nothing was force-pushed. git diff --check 5e2e6c7 c84c7753dae67a38bfefa99b7f279493d82e958e exited 0 with no output in my run. I read every changed hunk and found no other defect.
Verification
- GitHub reports 25 of 25 check runs successful at
c84c7753dae67a38bfefa99b7f279493d82e958e. Those are GitHub Actions' executions. I did not open their logs, because the diff is documentation only. - My own checks were read-only: the delta and
git diff --checkabove, the issue and dependency readbacks, and reads of the code and the pinned SDK at the lines linked in each reply. - I ran no Cargo, tests, fmt, Clippy or CI scripts locally, under the user's standing instruction.
- What counts as the full suite for this documentation-only PR is still the user's decision, and it has to be settled before any approval.
Next actor. CREATOR, for the secret-store thread.
|
CREATOR: Re-review requested for The pair now uses the actual Executed read-only inventory passed: 2 documents, 2 local links, 88 command templates/94 adapter-expanded commands, 71 named existing targets, no errors; Full acceptance remains the entire unchanged 25-check hosted matrix with no docs-only exemption. Current-head state at this read: {"IN_PROGRESS":8,"SUCCESS":1,"QUEUED":16} across 25 checks; pending jobs are not passes. Prior-head successes are not credited to this repair. No local Cargo, tests, fmt, Clippy, CI scripts or mutations ran. PR is OPEN and non-draft against main. Planning phase for #35; no implementation, issue closure or merge claimed. One-account mode: acting login and PR author |
aram356
left a comment
There was a problem hiding this comment.
REVIEWER: APPROVE — GitHub blocked formal approval: HTTP 422 "Review Can not approve your own pull request". The acting login and the PR author are both aram356, so this verdict is recorded as a COMMENT.
Planning review of 214d099cbe5690c3f5677800e8c21adfc9b423ed for #35. All 12 review threads are resolved, and the full hosted matrix passes at this head.
Threads
- 11 were accepted and resolved in the previous round.
- Secret-store outcomes is accepted and resolved this round. The rows now key on the SDK's actual
SecretErrorand the twoOkoutcomes. Their 500/503 split matches the SDK's own conversion, a too-large token is no longer overload, and the 4,096-byte cap is named.
This round's delta. git diff c84c775 214d099cbe5690c3f5677800e8c21adfc9b423ed touches only the two documents, with 44 insertions and 13 deletions. c84c775 is an ancestor of the new head. git diff --check c84c775 214d099cbe5690c3f5677800e8c21adfc9b423ed exited 0 with no output in my run. I found no other defect in the changed hunks.
Full suite. The user chose the hosted GitHub Actions matrix as this documentation-only PR's full prescribed suite.
- GitHub reports 25 of 25 check runs successful at
214d099cbe5690c3f5677800e8c21adfc9b423ed. These are GitHub Actions' executions, not mine. - I fetched every job's log with
gh api --allow-escape-sequences repos/IABTechLab/auctioneer/actions/jobs/<id>/logs. Each job checked outHEAD is now at 10a8bec Merge 214d099cbe5690c3f5677800e8c21adfc9b423ed into d26fcf7f566db6d1dce2e1847d79c9f44581edb8. - Across all jobs, no
test result:line was anything butok, and no test printedFAILED.
| Job | Summaries | Summed passed / failed / ignored |
|---|---|---|
cargo test (workspace-default) |
246 | 6,680 / 0 / 19 |
cargo test (workspace-insecure) |
243 | 6,673 / 0 / 19 |
cargo test (secure) |
190 | 4,882 / 0 / 11 |
cargo test (cli) |
25 | 1,224 / 0 / 3 |
cargo test (compiler conformance schema) |
14 | 1,046 / 0 / 0 |
cargo test (evidence-producers) |
2 | 2 / 0 / 0 |
The sums overlap across jobs and are not a workspace total. The ignored counts are pre-existing ignored cases in those binaries.
The workspace-default job ran:
cargo test --workspace --all-targets --no-fail-fast -- --skip tests::selected_schema_source_producer_records_exact_real_dispatch_without_qualification --skip tests::controlled_native_acceptance_requires_all_producers_but_never_completes_the_full_catalogSelected verbatim result lines:
workspace-default: test result: ok. 1229 passed; 0 failed; 9 ignored; 0 measured; 0 filtered out; finished in 1.75s
workspace-default: test result: ok. 801 passed; 0 failed; 3 ignored; 0 measured; 0 filtered out; finished in 46.34s
secure: test result: ok. 684 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out; finished in 0.50s
compiler schema: test result: ok. 686 passed; 0 failed; 0 ignored; 0 measured; 6 filtered out; finished in 325.22s
evidence-producers: test result: ok. 1 passed; 0 failed; 0 ignored; 0 measured; 18 filtered out; finished in 763.10s
evidence-producers: test result: ok. 1 passed; 0 failed; 0 ignored; 0 measured; 18 filtered out; finished in 319.97s
The remaining jobs passed too: the three other compiler-conformance lanes, config build failures, the fanout conformance suite and its Clippy, both schema-source probes, all four application artifacts, fmt + Clippy, and the Cloudflare, Fastly and Spin guest checks and package builds.
I ran no Cargo, tests, fmt, Clippy or CI scripts locally, under the user's standing instruction. No check was dismissed.
Scope of this verdict. It covers the specification and plan at 214d099cbe5690c3f5677800e8c21adfc9b423ed. It does not authorize implementation, a new resource reserve, deployment or a merge. Several maintainer decisions are recorded by CREATOR, and I did not see those answers myself:
- the no-cache policy;
- the hard cut;
- the 4,096-byte secret cap and its error mapping.
One-account mode. The acting login and the PR author are both aram356. This COMMENT is not a formal approval and does not make the PR merge-ready. The next step is CREATOR's merge question to the user. Only the user can decide to merge without a formal approval, as a one-time exception for this PR at this head, and CREATOR records that exception on the PR before merging.
PRfection phase: plan
Refs #35. Implementation ownership remains with #53, #77 and #78; this planning PR closes none of them.
Scope
Specify one API-backed path for stored OpenRTB request/impression templates and account policies, shared by standalone handlers and the settings-first embedded auction API. This PR contains only the specification and implementation plan. It changes no production code, workflow, dependency, workload limit or resource ceiling.
Contract
auction_datablock to the one runtime operator policy. Configure the complete endpoint, namespace, secret reference and JSON Pointer selectors.Review Repairs
The first repair at
c84c7753dae67a38bfefa99b7f279493d82e958ewas re-reviewed: REVIEWER accepted and resolved eleven threads. The last thread correctly identified that its secret-read mapping still named config extraction errors. That remaining contract is corrected at214d099cbe5690c3f5677800e8c21adfc9b423ed; it awaits REVIEWER acceptance. Earlier accepted repairs remain:The remaining secret-read thread now has the exact
SecretError/successful-read mapping, independently checked 4,096-byte backend/value caps, fixed-500 oversize handling and separate boundary/deadline/non-disclosure controls in Task 5. Eleven threads were resolved by REVIEWER, not CREATOR; the last remains open for REVIEWER. Whole-pair advisory re-review approved this correction, not formal independent PR approval.Decisions And Gates
On 2026-11-18 the user explicitly approved “the 4-KiB cap and error mapping”: both token value and guest-visible backend read are capped at 4,096 bytes; Internal/missing/invalid/oversized results are fixed 500, Unavailable is 503, and DeadlineExceeded retains typed expiry. This logical input contract does not approve production activation or a new reserve; retained token/read/prefix/header-copy overlap remains in the separate resource review.
On 2026-11-18 the user approved no Auctioneer cross-request cache for the initial delivery and requested a future configurable-cache ticket. The acceptance criteria of #35, #53 and #78 now reflect fresh initial snapshots, request-pinned coherence, no stale/negative fallback and retained failure/privacy/resource/measurement controls. Optional configurable caching is tracked separately in #245, a Backlog Story under epic #13, dependent on #53 and not blocking uncached delivery. This closes the pending policy/issue-alignment decision, not the implementation or reviewer-acceptance gates.
Skills used:
superpowers:brainstorming,superpowers:writing-plans, andsuperpowers:receiving-code-review, through the PRfection CREATOR workflow. The written design/publication approval is recorded in the spec; the 2026-11-18 no-backwards-compatibility instruction is recorded separately. Independent resource review, explicit new-reserve approval, implementation authorization and merge approval remain distinct.Verification
Current repair head:
214d099cbe5690c3f5677800e8c21adfc9b423ed. Only the two planning documents changed: 44 insertions and 13 deletions.Executed read-only documentation verification, exit 0:
The first repair head
c84c775passed all 25 hosted checks in Run Tests 44354797840 and companion workflows. Inspected secure-job output at that prior head, GitHub Actions exit 0:Those are selected real prior-head summaries, not a combined test count, new lookup coverage or current-head verification. Fresh full CI is pending at
214d099cbe5690c3f5677800e8c21adfc9b423ed. The prescribed suite is the complete unchanged 25-check hosted matrix, including all feature/compiler/producer/guest/provider/lint owners; no docs-only exemption. No local Cargo, tests, fmt, Clippy, CI scripts or mutation runs were executed. Current-head reviewer acceptance and approval remain outstanding; passing CI alone is not merge authorization.Handoff
Planning phase for #35; next actor is REVIEWER for the secret-read disposition and current-head hosted CI. No implementation, qualification, issue closure or merge is claimed. The hard cut remains mandatory during the separately reviewed implementation; no legacy reader or compatibility implementation is restored here.
One-account mode: acting login and PR author
aram356. Branchprfection/35-stored-objects-plan; retained worktree.worktrees/prfection/35-stored-objects-plan. Spec and plan links above bind this exact head. Temporary task patch/payload files are removed at handoff. The open planning worktree stays for re-review; unrelated worktrees and previously reported unowned files are not removed.