Repository navigation
Description
Residuals from the 2026-07-16 semantic audit of decidesk's 88 done specs. Fixed items landed in fix-inert-seeds (#138) and done-spec-fixes; everything below is deliberately not fixed there, because each needs a product decision an audit shouldn't make on its own.
Headline: decidesk is mostly healthy
decidesk is the origin corpus of the gate-6 orphan-auth class, so the 2026-04 trio was re-verified first:
| Method | Status |
|---|---|
isTransitionAllowed() |
Genuinely wired — MeetingService:176, DecisionLifecycleService:229, DecisionTransitionGuard:216 |
requiresChairAuthorization() |
Genuinely wired, fail-closed — MeetingService:191, DecisionLifecycleService:133 & :245 |
validateQuorum() |
Legitimately superseded — QuorumService deleted in #164; isQuorumRequired() / checkQuorum() wired |
No fabricated pass exists anywhere in the app. lib/Service + lib/Lifecycle were grepped for => true / return true near always / for now / TODO / placeholder / stub / temporary / assume / simplif — zero hits. Recording the negative so absence of a report isn't mistaken for absence of a look.
Residual 1 — three orphaned capabilities (zero production callers)
| Capability | Evidence |
|---|---|
QuorumVerificationService::computeQuorum() |
DI-registered (Application.php:772), 5 green unit tests, only tests call it. Textbook green-tests-dead-feature. |
DecisionNotificationService::notifyOnPublish() |
DI-registered (Application.php:583). Zero callers anywhere, not even tests. Possibly superseded by the declarative x-openregister-notifications dialect (ADR-031) — unconfirmed. |
NotificationPreferenceService::createPreference() |
Zero callers anywhere. |
Each needs the same product decision: is the feature wanted? If yes, wire it and prove it live. If no, delete it and drop the spec claim. Guessing would replace a false done with a false fixed. Until then their specs should not read as live.
Residual 2 — settings#load has no route (gate-14)
SettingsController::load() calls loadConfiguration(force: true) — decidesk's "re-import the configuration" recovery path — but it has no entry in appinfo/routes.php, so it is unreachable. Felt directly during this audit: proving the seed fix required occ app:enable because the endpoint that exists to do exactly that cannot be called.
Also unrouted: settings#index, settings#create, preferences#getPreference, preferences#setPreference.
Residual 3 — two OpenRegister defects (upstream, found by live-verifying)
importSeedData()fails all seeds on one stale register id. It resolves its target as$configuration->getRegisters()[0]and callsregisterMapper->find()unguarded; a stale id throws, the outer catch swallows it asSeed-data import failed, and every seed silently fails — even when a valid register sits later in the same list. Hit live: config 150 recorded[2409,18], 2409 dangling, register 18 valid and ignored. Suggest guarding the find and trying each recorded register.importFromJson()'s early-return skips seed import entirely, contradicting the comment at ImportHandler.php:3029 that promises the version-equal path still "checks seedData". The comment describes intended behaviour the code does not implement.
Residual 4 — hydra gate-runner truncates its own suite
scripts/run-hydra-gates.sh:2803 — [ -n "${HYDRA_GATE_PR_BODY}" ] under set -u with the var unset ⇒ unbound variable, aborting before gates 50-57. It only fires when gate-49's log is non-empty, so it silently truncates the suite exactly on the apps that have findings — including skipping the orphan-capability gates. First run here reported "8 failed" and never ran gates 50-57; with HYDRA_GATE_PR_BODY="" the real total is 12. Suggest ${HYDRA_GATE_PR_BODY:-}.
Residual 5 — pre-existing accessibility gate failures
Gates 32 (semantic-controls, 5), 38 (skip-link, 5), 40 (form-label-association, 10), 43 (table-headers, 2), 45 (prefers-reduced-motion, 1), 49 (controller-exception-translation, 3), 46 (spec-anchor-existence, 54), 53 (effective-manifest-crossref — ajv unresolvable, refused to run fail-open). Unrelated to spec semantics; listed for completeness.
Coverage caveat
The 88 done specs were audited by defect-shape sweep (gates + targeted greps + supersession checks), not by reading all 88 end-to-end and driving each through the UI. A spec whose capability is wired and whose dialect is canonical would pass this audit while still being wrong about its behaviour. The verdict below is "no evidence of the six shapes", not "all 88 verified live".
Gemigreerd van Codeberg — oorspronkelijk Conduction/decidesk#139, geopend door rubenvdlinde op 2026-07-16.
Activity
github-actions commentedon Aug 22, 2026
Thank you for opening this issue! It has been assigned to @rubenvdlinde for review.
What happens next:
Triage labels:
triage:accept-feature— accepted as a new featuretriage:accept-bugfix— accepted as a bug fixtriage:needs-info— more information neededtriage:wontfix— will not be addressedtriage:duplicate— duplicate of another issuerubenvdlinde commentedon Sep 25, 2026
Still true at
b7c056ed:QuorumVerificationService::computeQuorum()andgetAttendanceReport()have no caller outside their own class and unit test. The capability matrix in #1372 (rowsliv-02,bod-14) hit the same gap from the user side: no page shows quorum status and a body's quorum rule does not change which votes may open. Filed together with two other uncalled services in #1380.