Testing plan: one repository's failure is that repository's

Derived from bugfix.md and design.md, before tasks.md. Authored at test-planning, completed at verification.

Test matrix

#	Type	Applies?	Scope / what it proves	Where it runs
T1	Unit — the GitHub provider	yes	listing() isolates per repository (items from the healthy one, a ScopeFailure for the other, both pull-request and issue failures); "has disabled issues" is permanent: first sighting in failures, then skipped with no gh issue list call while gh pr list still runs, re-probed at cycle 60, renewed silently on a repeat, recovered on success; scope_of; list_work_items still raises; owns untouched	uv run pytest cli/tests/test_poller.py -k "scope or listing or quarantin or disabled"
T2	Unit — the poller core	yes	a Listing with one failure still processes the other items (spawn recorded); poll.scope_error / poll.scope_degraded / poll.scope_recovered emitted with the documented fields; summary.errors, scopes_failed, scopes_skipped, scopes_polled; reconciliation skips sessions in a degraded scope and still runs for healthy ones; a provider without listing() behaves exactly as before	uv run pytest cli/tests/test_poller.py -k "scope or listing or isolat or reconcil"
T3	Unit — heartbeat + status	yes	the three new lastCycle keys round-trip; a heartbeat without them reads; heartbeat_lines prints degraded: per scope and the "no repository was polled" line; a clean cycle prints exactly what it printed before; --format json carries the facts; the exit code is unchanged	uv run pytest cli/tests/test_poll_heartbeat.py cli/tests/test_poll_status.py
T4	Integration (scenario)	yes	Scenario: one repository with Issues disabled does not blind the others — the real provider, the real dispatcher, a two-repository gh double: cycle 1 spawns the healthy item and records the degraded repository; cycle 2 makes no gh issue list call for it, still lists its pull requests, and forwards a new comment on the healthy item; Gherkin-documented	uv run pytest cli/tests/test_poller_integration.py -k disabled
T5	Event catalogue parity	yes	every emitted event type is in EVENT_TYPES	uv run pytest cli/tests/test_eventlog.py
T6	Security / abuse case	yes	A2 (pull requests still listed while issues are skipped), A3 (only the exact message classifies; a 502 stays transient), A4 (a session in a degraded scope is never reconciled)	T1, T2, T4
T7	Lint / typecheck / tests	yes	the commands CI runs	make check
T8	Contract (OpenAPI)	n/a — the status route's row already carries lastCycle as an open object; no schema field changes
T9	UI / visual	n/a — no dashboard change
T10	Performance	n/a — the quarantine saves one gh call per skipped repository per cycle; nothing else changes in cost
T11	Migration / upgrade	n/a — the heartbeat's new keys are additive and an older file reads unchanged (T3)
T12	Manual	n/a — no repository with Issues disabled is reachable from this session; gh's exact message is taken from the ticket's log excerpt and pinned in T1/T4

Scenarios & requirement trace

Row	Requirement(s)	Scenario / case
T1	R1.1, R1.4, R2.1–R2.4, A2, A3	per-repository listing; the quarantine timeline; the strict form
T2	R1.1–R1.4, R2.1, A4	isolation in the core; events; reconciliation at the finer grain
T3	R3.1–R3.4	the heartbeat and the two status formats
T4	R1.1, R2.1, R2.2, R4.1	Scenario: one repository with Issues disabled does not blind the others
T5	R1.2, R2.1	the catalogue

Verification environment

• Repositories: this repository only.
• Services / containers: none — gh is replaced by an injected runner in every test.
• Fixtures & data: canned gh --json answers in the tests; gh's "has disabled issues" message copied from the ticket.
• Credentials: none.
• Bring-up: uv sync · Tear-down: none.
• If bring-up fails: record it under Verification results and escalate.

Evidence plan

• evidence/verification.md — red→green per task, make check output, the matrix results.
• evidence/security-review.md — the abuse-case dispositions.

Activities checklist

• [x] T1 provider unit tests, red first
• [x] T2 core unit tests, red first
• [x] T3 heartbeat and status tests, red first
• [x] T4 integration scenario, red first
• [x] T5 catalogue parity
• [x] T6 abuse cases (in T1/T2/T4)
• [x] T7 make check

Verification results

Recorded at verification in evidence/verification.md: every applicable row (T1–T7) pass; make check clean — ruff, format, pyright, config validation, markdownlint, 2976 tests passed / 1 skipped. T8–T12 n/a as planned.