medlab

cases48

1. 2026-11-02 A workaround written down is assumed to still be the design

   now fails loudly None mechanical. The paragraph now describes what the tool does, and the neighbouring section says the same thing, which is the only check this document has.

   Human ComprehensionEvolutionDocumentation DriftContract Tests

2. 2026-11-02 A review verdict belongs to the commit you are looking at

   now fails loudly A posted verdict names the head it read, so the server binds it to that commit, and a standing refusal names the commit it was given at rather than the commit under merge.

   InvariantsHuman ComprehensionContract TestsDecision Provenance

3. 2026-11-02 A review on your own pull request is assumed to be invalid

   now fails loudly The refusal covers verdicts only: an author may post a comment review on their own pull request, and a test requires that path to exist, so the handback has a channel.

   InvariantsHuman ComprehensionIndependent ReviewContract Tests

4. 2026-11-02 A list read at its page size is the oldest page, not the record

   now fails loudly Both readers page until a page comes back short, bounded at twenty pages of fifty and failing closed at the bound; tests assert that an approval on a later page is found and a refusal on a later page still blocks.

   Human ComprehensionInvariantsContract TestsExample-Based Tests

5. 2026-11-02 A record that fills its last page has not been shown to continue

   now fails loudly Both readers ask one page past the bound: an empty page means the record ended there and a non-empty one means it did not, so a record of exactly a thousand reviews is read rather than refused.

   InvariantsTest EffectivenessBoundary SensorsContract Tests

6. 2026-11-02 A model's quotation of a sentence is the sentence

   now fails loudly The fidelity check compares words after folding characters that carry no meaning, stores the source's own characters when a sentence matches that way, and reports a validator refusal as a refusal rather than as malformed output.

   InvariantsHuman ComprehensionSchema ValidatorProperty-Based Testing

7. 2026-11-02 A word is the first substring that matches it

   now fails loudly Cloze construction selects a whole source word by position, and a test requires that a target which is a substring of a longer word is refused rather than cut.

   Human ComprehensionInvariantsExample-Based TestsMetamorphic Testing

8. 2026-11-02 A commit's CI result is assumed to be reachable through a pull request

   now fails loudly `forge checks` answers for a pull request number, a revision, or the branch's head, and refuses an argument it cannot read as either rather than guessing.

   Human ComprehensionInvariantsContract TestsExample-Based Tests

9. 2026-10-26 A page that leaves no receipt is assumed not to have fired

   now fails loudly The probe logs what the messaging service answered for every page, accepted with its message id or refused with the reason, so a page's delivery is a fact instead of an inference.

   Human ComprehensionRuntimeSynthetic MonitoringDecision Provenance

10. 2026-10-24 Cutting between words preserves a sentence's meaning

    now fails loudly Regression tests require the full selected context and its governing modal to survive cloze construction.

    InvariantsTest EffectivenessRuntime InvariantsMutation Testing

11. 2026-10-16 An OAuth session renews itself before its refresh token lapses

    now fails loudly The harness renews its model-provider refresh token before it lapses (or lengthens the token's TTL), and a probe alerts when the token nears expiry, so the daily expiry boundary no longer emits an ERROR span.

    RuntimeInvariantsSynthetic MonitoringObservability Events

12. 2026-10-14 A migration transaction also locks SQLite connection setup

    now fails loudly A test holds the initialization lock and asserts that a second initializer cannot create the database before the lock is released.

    RuntimeIntegration Tests

13. 2026-10-14 A file readable by the build owner is readable by the runtime user

    now fails loudly The image build reads packaged public assets after switching to its unprivileged runtime user.

    BehavioralIntegration Tests

14. 2026-10-14 A shared photo always arrives as image data

    now fails loudly A native iOS test supplies a camera filename and requires a stored image after the recovery picker.

    BehavioralIntegration Tests

15. 2026-10-14 Successful signing implies valid workflow control flow

    now fails loudly Reject repeated control-flow group identifiers before signing and test balanced block nesting in the generated artifact.

    BehavioralIntegration Tests

16. 2026-10-14 Valid structured output preserves the source vocabulary

    now fails loudly A sparse normalization fixture must still deliver every short-source label to extraction, and phrase lemmas cannot collapse onto one base word.

    BehavioralStructuralSchema ValidatorIntegration Tests

17. 2026-10-14 A running shell script already has its remaining source

    now fails loudly The runner executes an in-memory snapshot of its source before starting simulator work.

    BehavioralIntegration Tests

18. 2026-10-14 A recurring failure warrants a new task in each delivery window

    now fails loudly identical failure shapes coalesce into one task per six-hour horizon instead of one per fifteen-minute window, with a running occurrence count, and a shape that reappears after a quiet horizon is flagged as a recurrence

    RuntimeInvariantsObservability EventsIncident CorrelationRuntime Invariants

19. 2026-10-14 A successful retry removes its failure alert

    now fails loudly A failure-to-recovery test requires capture identity and removal of only that capture's active alerts.

    BehavioralIntegration Tests

20. 2026-10-14 A failed preflight command stops the next shell command

    now fails loudly A deployment-script test injects a busy worker and requires termination before the build command.

    BehavioralIntegration Tests

21. 2026-10-14 A self-hosted CI runner has no dependency on the public internet

    now fails loudly A scheduled probe resolves the action registry hostname and reports the DNS error by name; a fallback public resolver removes the single point of failure.

    RuntimeStructuralSynthetic MonitoringObservability Events

22. 2026-10-14 An image decoder that works on desktop also handles phone photos

    now fails loudly Browser tests disable the bitmap decoder and require a resized JPEG upload, while undecodable images must show an actionable error without sending a request.

    BehavioralIntegration Tests

23. 2026-10-14 Shifting selected new cards leaves suspended cards unchanged

    now fails loudly A real-backend promotion test asserts that suspended and reviewed cards retain their scheduling fields.

    BehavioralIntegration Tests

24. 2026-10-14 A local sync server covers hosted download routing

    now fails loudly A bootstrap contract test requires the shard endpoint returned by the sync handshake to reach the download call.

    RuntimeIntegration TestsContract Tests

25. 2026-10-11 A tool returning output means the command succeeded

    now fails loudly a regression requires exit code 1 to remain a command failure when the tool error flag is false

    RuntimeBehavioralDistributed TracesIntegration Tests

26. 2026-10-11 A trace span whose status is ERROR records a station failure

    now fails loudly the telemetry relay drops every error span and error log record carrying a tool-call marker, so a routine non-zero exit no longer reaches the specimen queue; a tool call that crashed is dropped by the same rule, which is the hole the last section names

    RuntimeDistributed TracesObservability Events

27. 2026-10-11 Every exported error occurrence warrants a separate incident task

    now fails loudly regression tests reject historical exports and consumer tool errors, group repeated failures, and enforce a persisted rolling delivery budget

    RuntimeInvariantsObservability EventsRuntime Invariants

28. 2026-10-11 A queued notification can run with another channel's tools

    now fails loudly a queue regression requires different channel origins to remain queued for their own turns while same-origin follow-ups still fold

    RuntimeInvariantsIntegration TestsRuntime Invariants

29. 2026-10-11 A deploy that misses its host once is a failed publish

    now fails loudly the deploy retries a dropped connection up to three times and fails fast on any other error

    RuntimeDistributed TracesSynthetic Monitoring

30. 2026-10-11 A developer's Python installation supplies the CI test dependencies

    now fails loudly CI installs the pinned Python test requirements and verifies the YAML module before building or running the suite.

    BehavioralIntegration Tests

31. 2026-10-11 A CI runner carries the same tools as a developer's machine

    now fails loudly make test preflights the suite's declared Python dependencies and fails before the demos when one is missing

    RuntimeInvariantsSmoke TestsDependency Graph

32. 2026-10-10 An Ubuntu host label was safe on a macOS runner

    now fails loudly runner-label-check reads active runner config and pages when a Darwin host runner lacks a matching macOS :host label

    RuntimeInvariantsSynthetic MonitoringRuntime Invariants

33. 2026-10-10 A restarted worker is assumed to have recovered its queue

    now fails loudly a stalled queue alarms while the worker's message log is still advancing, and a recovery nudge must not report complete while older pending work remains

    RuntimeInvariantsSynthetic MonitoringResource TelemetryRuntime Invariants

34. 2026-10-10 A required tool is assumed to be installed on whichever machine runs the job

    now fails loudly the step that checks for a required tool installs it when it is absent, instead of only adding its presumed directory to the search path

    ChangeStructuralSmoke Tests

35. 2026-10-10 The token file was assumed to update the running service

    now fails loudly switchboard-watch restarts on /metrics 401, retries authentication, pages if the retry fails, and clears blind-alert state after a successful scrape

    RuntimeInvariantsSynthetic MonitoringRuntime Invariants

36. 2026-10-10 A busy worker is assumed to be making progress

    now fails loudly a liveness check declares a worker wedged when its message store stops advancing while it is running and busy, then restarts it and pages the operator

    RuntimeInvariantsResource TelemetryRuntime InvariantsSynthetic Monitoring

37. 2026-10-07 A placeholder written for a reader was a broken link the site's own checker followed

    now fails loudly a build gate fails when generated text carries a literal <...> placeholder inside an http(s) URL, so a placeholder cannot ship as a 404-ing link on the site's own domain

    StructuralTest EffectivenessPre-Promotion Invariant GatesSnapshot Tests

38. 2026-10-07 A running worker is assumed to be listening to its queue

    now fails loudly a scheduled liveness check alarms within minutes when a queue's oldest pending item has gone unclaimed by its owning worker past a short limit

    RuntimeInvariantsSynthetic MonitoringResource TelemetryRuntime Invariants

39. 2026-09-27 An empty backup suffix makes `sed -i` portable

    now fails loudly a gate refuses any `sed -i` whose argument is not a `.bak` suffix, so both the GNU-only no-suffix form and the macOS-only empty-suffix form fail the suite before they reach CI

    BehavioralEvolutionExample-Based TestsPre-Promotion Invariant Gates

40. 2026-09-24 A restart policy that fires on exit restarts a hung worker

    now fails loudly a per-harness run timeout now bounds a persistent worker's lifetime, killing and respawning a wedged process and re-arming the timeout every lifetime

    RuntimeInvariantsRuntime InvariantsResource Telemetry

41. 2026-09-21 `sed -i 'edit' file` means "edit in place" on every machine

    now fails loudly `sed -i` with no backup suffix is GNU-only; the deck test now uses the portable `sed -i.bak` + `rm -f` form, which edits in place identically on GNU and BSD sed

    BehavioralExample-Based TestsPre-Promotion Invariant Gates

42. 2026-09-19 Hardcoding a "recent" date into a test fixture keeps it recent

    now fails loudly the telemetry collector accepts SUBLAYER_NOW to freeze its clock, and the test pins it to the fixtures' own day, so the window is deterministic and cannot age out

    BehavioralTest EffectivenessExample-Based TestsPre-Promotion Invariant Gates

43. 2026-09-13 Moving a repo layout's root is one mechanical change

    now fails loudly A test that silently skips because a path does not exist is a failed test; environment checks that mask defaults are recorded as overrides, and the defaults themselves are tested without them.

    ChangeBehavioralIntegration TestsPre-Promotion Invariant Gates

44. 2026-09-12 YAML frontmatter with a colon-space in a scalar value is a mapping

    now fails loudly Every SKILL.md frontmatter is parsed as YAML by a test; a frontmatter block that does not parse fails CI.

    StructuralRuntimeSchema Validator

45. 2026-09-12 Renaming a repo and updating its topology entry is one operation

    now fails loudly The deploy gate refuses when the topology file and the filesystem disagree about which repos exist; a worker with a schedule attached is not a worker — it is skipped by collect and never rendered as down.

    ChangeInvariantsPre-Promotion Invariant GatesStatically Checked Invariants

46. 2026-09-12 An AI reviewer that reports an approval has posted one

    now fails loudly A review counts only when the forge API shows it submitted with a verdict; until the tool submits, the merge record is the evidence and the reviewer's own result is not.

    ChangeHuman ComprehensionIndependent ReviewSecond-Agent Review

47. 2026-09-06 Looking up a DNS name before creating it is harmless

    now fails loudly Absence is probed only against the authoritative nameserver; the user's own resolver is queried last, and only to read what it already believes.

    RuntimeChangeSynthetic Monitoring

48. 2026-09-06 Validating a web server's config only reads it

    now fails loudly Config validation runs as the service user, and a smoke request against the new host is what declares the deploy done — not the validator's exit code.

    ChangeRuntimePre-Promotion Invariant GatesSmoke Tests