Skip to content

docs(api-review): Goldpath.Approvals and Goldpath.FileExchange under the [Experimental] decision (G4 API-5, #394) - #576

Open
omercelikdev wants to merge 2 commits into
mainfrom
agent/394-api-review-approvals-fileexchange
Open

omercelikdev wants to merge 2 commits into
mainfrom
agent/394-api-review-approvals-fileexchange

Conversation

@omercelikdev

@omercelikdev omercelikdev commented Nov 11, 2026 •

Copy link
Copy Markdown
Collaborator

What and why

API-5 of the 1.0 public-API review. It gives every live ledger line of the two experimental packages a verdict, under owner decision Q4: the packages ship in 1.0 with their surface [Experimental].

What was measured (at main@c9d23ae): 635 live lines, 376 in Approvals and 259 in FileExchange. The issue's 462 was measured at 0186ca3, before #479 and #480 grew Approvals.

The verdicts:

  • 588 experimental, under GPEXP001 (Approvals) or GPEXP002 (FileExchange).
  • 24 remove.
  • 23 rename: one record, GoldpathQuarantinedRow → GoldpathFileQuarantinedLine (added at e822209 from the verifier's finding 2).
  • No keep (the gate refuses it here).

The removals:

  • The two EF stores (11 lines each): GoldpathEfApprovalStore<TContext> and GoldpathEfFileLedger<TContext>.
    • Each is registered only by the TContext overload of AddGoldpathApprovals / AddGoldpathFileExchange, behind its store interface.
    • No other public line, template or sample names either (measured).
    • So unlike Bulk's GoldpathBulkFileStore<TContext>, which a public constructor takes, nothing forces them to stay public. API-14 adds InternalsVisibleTo for the three test projects that construct them.
  • The two metrics classes (GoldpathApprovalsMetrics, GoldpathFileExchangeMetrics): public static classes with no public member. This applies API-3's metrics policy.

The rename: FileExchange has three types for one quarantined line: GoldpathFileQuarantineRow (the EF entity), GoldpathQuarantinedRow (the record the ledger's GetQuarantineAsync and GoldpathFileRailResult.Quarantined return) and GoldpathFileQuarantineInfo (the admin DTO). The record alone lacks the File prefix its siblings carry, and it takes the …Row suffix this review reserves for EF entities. The 23 rows are its 18 own lines plus the 5 lines whose signatures name it. Each row has the full target line and a reason that says why the rename comes before the T22 lift. It is listed in the README's findings for API-13, so the owner can sign or overrule it.

The experimental rows carry a reason per category. Each was checked against the code:

  • The EF rows: the adopter's migrations name them, which is why Bulk's and Archival's entities were kept.
  • GoldpathApprovalRequest's five set accessors: the engine moves state on the request it read, and a custom store outside the assembly materializes the same object. Neither init nor an internal setter serves both.
  • The model extension: it waits on API-3's spelling finding.
  • The escalation job: its type name is its job key.
  • The in-memory stores: registered with TryAdd, and constructed by adopters' tests.
  • The seams: IGoldpathApprovalStore, IGoldpathFileLedger and IGoldpathFileLedgerQueries. The last is the read half that the admin endpoints require, as the code checks.
  • The non-generic admin services: they read through IGoldpathApprovalStore / IGoldpathFileLedgerQueries, so a TContext parameter would bind nothing (method Q7).
  • The default reason claims only what was checked: Q1, Q2, and Q3 (measured: every token is last and named cancellationToken, and neither package has a ct).

Also in this PR:

  • The record's §4 states the lift condition once, linked to T21 and T22, with their proofs quoted as written. It sits after the diagnostic ids' explanation.
  • api-review-1.0/README.md "Findings for API-13" answers the issue's four questions (EF rows and plumbing, generic or not, singular or plural) and lists the rename for the owner.
  • ga-gate-status.md G4 now reads 2,136 of 3,349.
  • scripts/api-review-gate.sh --selftest resets its second fixture (Goldpath.FileExchange) to a header-only table in every scratch copy, as API-4 did for Caching. Its "experimental package under its own id" case appends a row for every live line, and the real table turned those into duplicates, so the selftest went red. The checks of the real tables are unchanged.

Reproduce

git fetch origin agent/394-api-review-approvals-fileexchange && git checkout agent/394-api-review-approvals-fileexchange
G="./scripts/api-review-gate.sh --require-complete --package Goldpath.Approvals --package Goldpath.FileExchange"
$G; echo rc=$?                                   # green, rc=0 (376/376, 259/259)
./scripts/api-review-gate.sh --selftest          # green
./scripts/api-review-gate.sh                     # report: 2136 of 3349; Approvals 364 exp / 12 remove; FileExchange 224 exp / 23 rename / 12 remove

# RED 1, main's tables: 0 of 635
git checkout origin/main -- docs/strategy/api-review-1.0/Goldpath.Approvals.tsv docs/strategy/api-review-1.0/Goldpath.FileExchange.tsv
$G; echo rc=$?                                   # rc=1: "Goldpath.Approvals: no verdict for a live line: …"
git checkout HEAD -- docs/strategy/api-review-1.0/

# RED 2, a planted keep (either table)
F=docs/strategy/api-review-1.0/Goldpath.FileExchange.tsv
python3 -c "import sys;p=sys.argv[1];L=open(p).read().split('\n');c=L[5].split('\t');c[1]='keep';c[2]='';L[5]='\t'.join(c);open(p,'w').write('\n'.join(L))" $F
./scripts/api-review-gate.sh; echo rc=$?         # rc=1: "… 'keep' is not a verdict there"
git checkout $F                                  # (same with Goldpath.Approvals.tsv)

# RED 3, main's selftest against these tables (the copy must sit in scripts/: the gate finds the repository root from its own path)
git show origin/main:scripts/api-review-gate.sh > scripts/.gate-main.sh && bash scripts/.gate-main.sh --selftest; rm scripts/.gate-main.sh
#   "api-review-gate selftest FAILED: - an experimental package under its own id went red: ['…Goldpath.FileExchange.tsv:261: duplicate row …"

# RED 4, a rename with its target blanked
python3 -c "import sys;p=sys.argv[1];L=open(p).read().split('\n');i=[k for k,l in enumerate(L) if '\trename\t' in l][0];c=L[i].split('\t');c[2]='';L[i]='\t'.join(c);open(p,'w').write('\n'.join(L))" $F
./scripts/api-review-gate.sh --require-complete --package Goldpath.FileExchange   # "…tsv:36: a 'rename' needs a target (the full new ledger line)"
git checkout $F

I ran RED 1–4 and saw each one as quoted.

Acceptance criteria (#394), with evidence

  • One row per live line in both TSVs, each experimental (with its package's id), rename or remove, and no keep. --require-complete is green for both; a planted keep is red (RED 2).
  • Every rename and remove carries a reason that says why it must happen before the 1.x lift. Each remove reason says what replaces the line and why no adopter needs it. Each rename reason says that from the lift the name is SemVer-bound and only a major could change it.
  • The lift condition is written once in the record's §4, linked to T21 and T22.
  • --require-complete --package for both is green, and report mode is green for the whole tree.
  • No file under packages/, templates/ or tools/ changes (git diff --stat origin/main).
  • The ga-gate-status.md G4 row shows the new count: 2,136 of 3,349.
  • Merge tier A. The data files are tier A, but scripts/api-review-gate.sh changes too (the selftest fixture reset, the same fix API-4 made), so the PR is tier C. The fix is needed because without it gates is red.

Pre-PR self-check

  • Every criterion is met with evidence, or stated plainly (the tier).
  • Ledgers: the CHANGELOG [Unreleased] entries (the review under Added, with the rename and the counts; the selftest fix under Fixed), the record's §4, the README's findings, and the G4 row in ga-gate-status.md only. goldpath-ga-readiness.md is untouched. No new deferral needs an open-threads row.
  • XML docs: not applicable. No code file changes.
  • Honesty: every category reason was checked against the code. The admin-service reason now names IGoldpathFileLedgerQueries, the constructor's parameter.
  • Wired / contracts / bench: not applicable (docs data).
  • Deep self-review: no runtime change. The verdicts are proposals; API-14 applies them. The rename is source-breaking only for a caller that names the record, and it is cheaper now than after the lift. The selftest change is confined to the scratch copy. Edge cases, concurrency, migrations, performance and code quality are not applicable.

Gates run (at e822209)

Gate Result
api-review-gate.sh (report), --require-complete --package Goldpath.Approvals --package Goldpath.FileExchange, --selftest green ×3
docs-freshness.sh, ledger-check.sh, schema-honesty.sh, skills-parity.sh, template-pins.sh, surface-freeze-gate.sh rc 0 ×6
dotnet build not run: no code changed

What I could not verify

  • Q4 (nullable annotations), line by line. I did not audit all 635 lines' annotations against the code. The default reason does not claim it.
  • Whether EF compiled models are a scenario Goldpath supports. This bears on the cross-module entity question in the findings.
  • CI: main's own oracle job is red (not required), so a red oracle here needs reading before it is attributed to this PR, which changes no code.

Closes #394

🤖 Generated with Claude Code

… line under the [Experimental] decision (G4 API-5, #394)

635 live lines at c9d23ae (376 Approvals, 259 FileExchange; the issue's 462
was measured before #479/#480): 611 experimental under GPEXP001/GPEXP002,
24 remove, no keep (GA readiness Q4), no rename needed.

- remove: the EF stores GoldpathEfApprovalStore<TContext> and
  GoldpathEfFileLedger<TContext> - registered only by the TContext overload of
  AddGoldpath..., behind the store interface; no other public line, template or
  sample names them (InternalsVisibleTo for the tests that construct them, in
  API-14). The metrics classes - public static classes with no public member.
- experimental, with a reason per category: the EF rows (the adopter's
  migrations name them, as Bulk's and Archival's entities), the request's set
  accessors (the engine and a custom store both need them), the model
  extension (waits on API-3's spelling finding), the escalation job (its name
  is its key), the in-memory stores, the seams, the non-generic admin services.
- The record's §4 states the lift condition once, linked to T21 and T22. The
  README answers the issue's four questions for API-13. ga-gate-status G4:
  2,136 of 3,349.
- api-review-gate.sh --selftest resets its second fixture (FileExchange) to a
  header-only table in every scratch copy, as it does Caching since API-4: the
  experimental-package case appends a row per live line, which a real table
  turned into duplicates.

Proof: --require-complete for the two packages is red at main (0/635) and
green here; a planted keep in either table is red ("'keep' is not a verdict
there"); --selftest green.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018AFdNcUHVZCtRVq4DLoW71
@omercelikdev omercelikdev added the agent-pr PR opened by the overnight agent label Nov 11, 2026
@github-actions github-actions Bot added the tier:C ADR-0014 Tier C: decisions, releases, gates, policy, agent layer; the owner merges label Nov 11, 2026
@omercelikdev

Copy link
Copy Markdown
Collaborator Author

Verifier verdict (run 2026-11-12)

Head verified: b3b7657c7181fd3c5461dabb9cfa609d1651c968 · Base: main @ c9d23ae5020cbd4a87540dd58e9c20253764d52b (= merge-base) · Tier C (owner merges).

Verdict: changes requested. 2 MINOR, 3 NIT. Both MINORs are about how accurate the verdict tables' reason text is. Every count, gate and red→green claim reproduced exactly.

Findings

  1. MINOR: the admin service's reason names the wrong interface. docs/strategy/api-review-1.0/Goldpath.FileExchange.tsv:139–143 (the 5 GoldpathFileExchangeAdminService rows) say "it reads through IGoldpathFileLedger, not the app's context". The constructor takes IGoldpathFileLedgerQueries (packages/Goldpath.FileExchange/GoldpathFileExchangeAdminService.cs:54,57; the TSV's own line 140 shows (…Options! options, Goldpath.IGoldpathFileLedgerQueries! queries)). The record is what the owner signs in API-13, so its reasons have to hold against the code. Fix: in those 5 reasons, change IGoldpathFileLedger to IGoldpathFileLedgerQueries. The Q7 argument still holds as written.

  2. MINOR: GoldpathQuarantinedRow gets the default "named like its siblings (Q2)" reason, but its naming was never actually examined. See Goldpath.FileExchange.tsv:69–78,210–212… (13 lines). FileExchange has three types for one quarantined line:

    • GoldpathFileQuarantineRow: the EF entity;
    • GoldpathQuarantinedRow: the record that IGoldpathFileLedger.GetQuarantineAsync returns (GoldpathFileRailEngine.cs:23,30);
    • GoldpathFileQuarantineInfo: the admin DTO.

    The record lacks the File prefix every other FileExchange data type carries. It also takes the …Row suffix, which this same review uses to mean "EF entity". The issue's premise is that a rename is cheap now and a major-only change after the T22 lift. So this is the one line where "nothing to change" needs either a stated reason or a rename row with a target. Fix: give the 13 lines a specific experimental reason that says why the name stays, or rename them with the target line. Either way, list it in the README's "Findings for API-13" so the owner sees it.

  3. NIT: docs/rfc/goldpath-1.0-api-review.md:105–111. The new lift-condition paragraph sits between the GPEXP001/GPEXP002 list and its continuation ("There is one id per package…", line 113), so that sentence now reads detached from the ids it explains. Consider moving the paragraph below line 115.

  4. NIT: docs/strategy/api-review-1.0/README.md now ends with a blank line (\n\n); main's ended with a single \n.

  5. NIT, no action needed: the issue asks for merge tier A. The PR is tier C because scripts/api-review-gate.sh changes. The PR says so, and the change is necessary (see RED 3 below).

Steps 1–6

# Step Result / evidence
1 Scope vs #394 ✓ Both TSVs have one row per live line: Approvals 371 Shipped + 5 Unshipped = 376, FileExchange 259 + 0 = 259. Every row is experimental/remove, with no keep and no rename; experimental targets are GPEXP001/GPEXP002. Every remove has a reason. The §4 lift condition links T21/T22 and matches their PROOF columns (open-threads lines 32, 35). The G4 count is updated. No file under packages/, templates/ or tools/ changes. Only deviation: the scripts/ change (NIT 5).
2 Rules / ledgers ✓ CHANGELOG [Unreleased] has Added (line 624) and Fixed (line 958). The ga-gate-status G4 row is correct: 1,501 + 635 = 2,136 of 3,349, the report total measured at head. No open-threads row added, so there is no T-number collision. Experimental reasons are mandatory under the record's §4 vocabulary, and they are present. XML docs and PublicAPI: n.a., since no code changes.
3 Confidentiality ✓ grep -inE over the diff and commit message for the listed names, \bBTK\b, factoring/corpay, secrets and internal hosts: 0 hits.
4 Reproduce ✓ --require-complete --package Goldpath.Approvals --package Goldpath.FileExchange: green, 376/376 and 259/259. Report mode: green, 2136 of 3349. --selftest: green. RED 1 (main's tables): "635 finding(s)", red. RED 2 (planted keep, line 6 of each TSV): red with the Q4 message, in both files. RED 3 (main's gate script --selftest against these tables): FAILED with the quoted duplicate row at Goldpath.FileExchange.tsv:261. Gates docs-freshness, ledger-check, schema-honesty, skills-parity, template-pins and surface-freeze-gate: rc 0 for all. dotnet build: n.a., since no project changes.
5 CI Required checks: gates, integration, kit and golden manifests (PR subset) are green. merge-policy is red by design: its log says "Tier C … without owner-approved". oracle is red, but it is not required and is red on base c9d23ae too. The failure is an Oracle connection error in NotificationFacts.The_renewal_mail_actually_lands… (OracleCommunication.DoConnect), so this PR did not cause it.
6 Charter n.a.

Deep review (a–j)

Area Result
a Security ✓ No runtime change. The two proposed removals shrink the surface. No new dependencies.
b Correctness of claims Findings 1–2. Checked as true: the EF stores are registered only by the TContext overloads (GoldpathApprovalsExtensions.cs:45, GoldpathFileExchangeExtensions.cs:46), behind the interfaces. Outside the ledgers and the review files, no template, sample or package names them. Only the module RFCs mention them, historically. The constructors are in Approvals.Tests + IntegrationTests and in FileExchange.Tests, which matches the InternalsVisibleTo lists in the reasons. Both metrics classes have only internal members. In-memory stores are TryAddSingleton. Escalation-job keying: GoldpathJobDefinition(Type) uses the type name. GoldpathApprovalRequest has exactly 5 .set lines (DecidedBy, PendingRole, PendingSince, Reason, Status). Q3: every CancellationToken parameter is last and named cancellationToken, with 0 ct in either package; the delegate Func<TRow, CancellationToken, Task> is not a parameter. IGoldpathFileLedgerQueries is enforced by MapGoldpathFileExchangeAdmin (…AdminEndpoints.cs:23). The EF-entity rows cover all of each *Row class's members (24/28). Q4 (nullability) was not audited, and the PR says so.
c Concurrency n.a. (data and docs)
d Data / migrations n.a.
e Performance n.a.
f Compatibility ✓ The verdicts are proposals, applied in API-14. The removals break only an adopter that constructs GoldpathEf…<TContext> directly, which the PR states.
g Test quality (gate change) ✓ Mutating the fixture reset to (X,) only (Caching not reset) turns the selftest red. Main's version (P only) turns it red (RED 3). Real-table checks are unchanged, and planted faults in the real tables still go red (probes below).
h Code quality ✓ The selftest change mirrors API-4's reset, and its comment explains why. NIT 3 is in the RFC.
i Observability n.a.
j Honesty Finding 1 (a reason contradicts the code). The PR's "What I could not verify" section is accurate.

Edge-case probes (throwaway, not pushed)

Probe Result
Duplicate an Approvals row red: "duplicate row (first at …:10)" ✓
Delete one FileExchange row under --require-complete red: "no verdict for a live line" ✓
GPEXP001 on a FileExchange row red (rc 1) ✓
Blank the reason of a remove row red: "a 'remove' needs a reason" ✓
Fixture reset mutated to FileExchange only selftest red ✓
Gate under LC_ALL=tr_TR.UTF-8 green. The tr_TR locale is not installed here, so it fell back to C; the gate is Python with no culture-sensitive casing. Not a true tr-TR run.

Owner: fix the two reason texts (Finding 1 is a one-word change in 5 rows; Finding 2 is a reason or a rename for one record type), then re-verify. This PR is tier C, so the owner merges.


Generated by Claude Code

@omercelikdev omercelikdev added the changes-requested Verifier found problems; the worker fixes these first label Nov 12, 2026
…rQueries, and GoldpathQuarantinedRow is a rename (#394)

Verifier findings on #576:
- The five GoldpathFileExchangeAdminService reasons named IGoldpathFileLedger; its
  constructor takes IGoldpathFileLedgerQueries.
- GoldpathQuarantinedRow carried the default reason without its naming being examined.
  It is now a rename to GoldpathFileQuarantinedLine (the File prefix its siblings carry,
  not the Row suffix the review reserves for EF entities): its 18 own lines and the 5
  lines whose signatures name it, each with the full target line and a reason saying
  why before the T22 lift. Listed in the README's findings for API-13; counts moved in
  the CHANGELOG, the README and the G4 row (588 experimental, 24 remove, 23 rename).
- NITs: the lift-condition paragraph moves below the ids' continuation; the README
  ends with one newline again.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018AFdNchYxTtB7tJ7kURQXf
@omercelikdev

Copy link
Copy Markdown
Collaborator Author

Addressed verifier findings: 1 (MINOR), 2 (MINOR), NIT 3, NIT 4. Head: e822209.

  • Finding 1. The 5 GoldpathFileExchangeAdminService reasons now say "it reads through IGoldpathFileLedgerQueries", which matches the constructor's parameter. Checked with grep -c "IGoldpathFileLedgerQueries, not the app", which returns 5.
  • Finding 2: rename, not a kept reason. I could not find a reason to keep the name that holds up.
    • GoldpathQuarantinedRow → GoldpathFileQuarantinedLine. Its 18 own lines are covered, and so are the 5 lines whose signatures name it: GoldpathFileRailResult's constructor, Quarantined and Deconstruct, IGoldpathFileLedger.GetQuarantineAsync, and the in-memory ledger's GetQuarantineAsync. The EF ledger's line stays remove.
    • Each of the 23 rows carries the full target line and a reason that says why the rename has to happen before the T22 lift.
    • It is listed in the README's "Findings for API-13", so the owner can sign it or overrule it.
    • Counts are updated in the CHANGELOG, the README and the G4 row: Approvals 364 experimental / 12 remove; FileExchange 224 experimental / 23 rename / 12 remove. The total is still 2,136 of 3,349.
  • NIT 3. The lift-condition paragraph now sits after "There is one id per package … API-11 (G4 11/14: Mark Goldpath.Approvals and Goldpath.FileExchange [Experimental] with documented diagnostic ids, and opt every in-repo consumer in #400) applies them."
  • NIT 4. The README ends with a single \n again.
  • PR body: refreshed for the new counts, the rename, and RED 4.

Pre-PR self-check (the whole PR at e822209):

  • Criteria: each is met with evidence. The tier is C because of the gate script, as stated in the body.
  • Ledgers: CHANGELOG [Unreleased], README findings, the record's §4 and the ga-gate-status G4 row are updated. goldpath-ga-readiness.md is untouched, and no new deferral is created.
  • XML docs: n.a., since no code changes.
  • Honesty: the reasons are checked against the code, including the corrected interface name. Every rename target is the actual new line text.
  • Wired / contracts / bench: n.a.
  • Deep review: n.a. for runtime areas, since this is data only. Gate behaviour is proven:
    • --require-complete is green.
    • The selftest is green.
    • A rename row with its target blanked → red: "a 'rename' needs a target".

Gates:

  • api-review-gate (report, --require-complete, --selftest): green.
  • docs-freshness, ledger-check, schema-honesty, skills-parity, template-pins, surface-freeze-gate: rc 0.

Generated by Claude Code

@omercelikdev

Copy link
Copy Markdown
Collaborator Author

Verifier verdict (run 2026-11-12, daytime)

Head verified: e8222094eb70a38402528ebf809b10af6f9a2575. Merge base: c9d23ae. It merges cleanly onto main@90f78d3 (git merge-tree). The delta since my last verdict (b3b7657) is one commit, 5 files.

Verdict: no blocking findings. 0 BLOCKER, 0 MAJOR, 0 MINOR, 1 NIT. Tier C (gate script): the owner merges.

Previous findings

  • MINOR 1 (wrong interface in a reason): fixed. All 5 GoldpathFileExchangeAdminService reasons now name IGoldpathFileLedgerQueries.
  • MINOR 2 (GoldpathQuarantinedRow waved through): fixed as a rename to GoldpathFileQuarantinedLine, and the README lists it under "Findings for API-13" for the owner to sign or overrule. I checked it mechanically:
    • PublicAPI.Shipped.txt has 24 lines that name GoldpathQuarantinedRow, and the TSV covers all 24: 23 rename and 1 remove (the EF ledger's GetQuarantineAsync).
    • For all 23 renames, the target is exactly the live line with the type name replaced. I compared each one with a script, so these are exact string matches.
    • The new name collides with nothing in packages/.
  • NIT 3: fixed. The lift paragraph now follows "There is one id per package … API-11 (G4 11/14: Mark Goldpath.Approvals and Goldpath.FileExchange [Experimental] with documented diagnostic ids, and opt every in-repo consumer in #400) applies them."
  • NIT 4: fixed. The README ends in a single \n.

Steps 1–6

# Step Result / evidence
1 Scope ✓ The delta is reasons, the rename rows, and the counts in CHANGELOG / README / G4 / RFC §4. No package, template or tool file changes.
2 Rules / ledgers ✓ The counts reconcile. FileExchange: 224 exp + 23 rename + 12 remove = 259. Approvals: 364 + 12 = 376. Total: 588 + 24 + 23 = 635. G4 is still 2,136 of 3,349. There is no new open-threads row, so no T-number can collide.
3 Confidentiality ✓ The denylist grep over the diff against main and over the commit messages: 0 hits.
4 Reproduce ✓ api-review-gate.sh --require-complete --package Goldpath.Approvals --package Goldpath.FileExchange: green. --selftest: green. Report mode: green. A rename with a blank target goes red (…tsv:36: a 'rename' needs a target). Gates docs-freshness, ledger-check, schema-honesty, skills-parity, template-pins, surface-freeze-gate: rc 0. No project changes, so dotnet build is n.a.
5 CI ✓ Required checks are green: gates, integration, kit, golden manifests (PR subset), CodeQL, review. merge-policy is red by design (Tier C). oracle (not required) has 2 of 73 red. One is NotificationOracleTests.The_renewal_mail_actually_lands… (Oracle connect failure), the same red as on main and on #575. This PR changes data and docs only.
6 Charter n.a.

Deep review

Dimension Result
a Security n.a. Data and docs only. The proposed removals shrink the surface.
b Correctness of the record ✓ The rename rows were checked mechanically, as above. The corrected reason matches the constructor (GoldpathFileExchangeAdminService.cs:54,57).
c–e Concurrency / data / performance n.a.
f Compatibility ✓ The rename is a proposal; API-14 (#403) applies it before the T22 lift, as the RFC paragraph says.
g Test quality ✓ For the gate: the blank-target plant goes red, and the selftest is green. See the NIT for what the gate does not check.
h Code quality ✓
i Observability n.a.
j Honesty ✓ The reasons now hold against the code, and the README's finding says the rename is the owner's to sign.

Edge-case probes (throwaway edits to the TSV, all restored; git status clean)

# Probe Result
P1 Blank the target of a rename row red ✓
P2 Point the type's rename target at an existing type, Goldpath.GoldpathFileQuarantineRow (the EF entity) green: the gate accepts a rename onto a name that already exists
P3 Make a rename target identical to its own line, which renames nothing green: the gate accepts it
P4 Every live line that names the type has a row, and every target is the substituted line ✓ 24/24 covered, 23/23 exact

NIT (optional, not this PR's change)

  1. P2 and P3 show that scripts/api-review-gate.sh checks only that a rename target is non-empty. It does not check that the target differs from the line, or that it names no type already on the surface. This PR's data is correct (P4), and the validation predates it (API-4), so it does not block. This PR is the first to use rename, so a follow-up issue for the gate would be worthwhile before API-14 applies renames. Suggested checks: target ≠ line, and the new type name is not already declared in the package's PublicAPI files.

Labels: changes-requested removed, agent-verified added. Tier C: the owner merges, after owner-approved.

Owner: safe to merge. Summary: no runtime change. The 1.0 API review record now gives a verdict for every public line of Goldpath.Approvals and Goldpath.FileExchange ([Experimental] under GPEXP001/GPEXP002). It proposes two EF-store removals and one record rename (GoldpathQuarantinedRow → GoldpathFileQuarantinedLine) for you to sign in API-13.


Generated by Claude Code

@omercelikdev omercelikdev added agent-verified Independently verified by the verifier agent: ready for owner merge and removed changes-requested Verifier found problems; the worker fixes these first labels Nov 12, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

agent-pr PR opened by the overnight agent agent-verified Independently verified by the verifier agent: ready for owner merge tier:C ADR-0014 Tier C: decisions, releases, gates, policy, agent layer; the owner merges

Projects

None yet

Development

Successfully merging this pull request may close these issues.

G4 5/14: Review every Goldpath.Approvals and Goldpath.FileExchange public-API line (462) under the [Experimental] decision

2 participants