Execution Log: end-to-end PDLC scenario tests

Append-only log for issue-217. Ticket: #217.

How this session ran the loop

One cloud session, one pass, no human at the other end — the same posture as issue-208/209/211, with the same two consequences a reviewer should hold:

1. phase-selection was not run as a gate. The session was started by the ticket itself; there was nobody to tick the checklist. Phases assumed: the full spec chain, verification, self-review. brainstorming and the opt-in design-critic-review were not taken — the ticket already states the shape (scenario runner + fixture sets at existing seams), and no second model was available.
2. The chain was authored before the code, but approved by nobody. The artifacts are a proposal to ratify, not a locked chain; status: draft on all four says so.

Phase transitions

Phase	Entered	Reviewed/approved by	Notes
phase-selection	2026-08-12	—	Not run as a gate; see above
requirements-definition	2026-08-12		requirements.md — 4 requirements, 4 NFRs, 3 abuse cases. Risk tier 3: test-only change, but the traces it pins become the reference for how the process may behave
design	2026-08-12		design.md — a data-driven scenario runner over the real runtime; closed step vocabulary; fakes at transport seams only; four deliberate absences
test-planning	2026-08-12		testing-plan.md — 8 rows in scope, 7 n/a with reasons
tasks-breakdown	2026-08-12		tasks.md — 7 tasks
implementation	2026-08-12		Built. Tasks 1–6 complete
verification	2026-08-12		Testing plan executed in full: 15 e2e/meta tests green in 0.5s; whole suite 1886 passed + 1 skipped (+15 over baseline, incl. the self-review addition); lint, format, types, markdown, config validation clean
needs-review	2026-08-12		Handed to the PR

Pull requests

PR	Scope / tasks	Status
claude/github-issue-217-tnm2ek	the whole work item	open, awaiting human approval

Progress entries

2026-08-12 — orientation

Read the ticket, the harness config, the skill, and the issue-209 chain for conventions; mapped the graph runtime and the existing integration-test fakes before designing (two exploration passes over cli/the_loop/graph/ and cli/tests/). Baseline suite: 1873 tests collected.

2026-08-12 — the shape of the change

Three seam findings shaped the design:

• Discovery is pinned, and the pin is a sensitive path. the-loop scenarios reads testing.integrationTestGlobs, which this repo pins to the non-recursive cli/tests/test_*_integration.py. A test module inside test_pdlc_e2e/ would be invisible to it, and widening the glob means editing .the-loop/harness-config.yaml — a declared sensitive path this work item has no reason to touch. So the test module lives beside the fixture directory (test_pdlc_e2e_integration.py), which the existing glob already matches.
• The seams were already there. the_loop.graph.integrations.resolve is the one patch point every graph test uses (and the source comments say so — hooks bind it at call time for exactly this reason); the ask/reply tests define the poster/TmuxRunner patch shape; the event log configures to any path. No new mocking layer was needed (R3.1), and no production code changed (NFR2).
• riskTier does not gate anything in the runtime. The tier lives in front matter and on the WorkItem, but no shipped hook reads it — tier scaling is declared at phase-selection by a human, not computed. The trivial-tier scenario therefore asserts the declared-skip mechanics (provenance, onlyWhenSkipped re-targeting), not a tier-driven branch that does not exist.

Documentation

Doc	Change
docs/capabilities/testing-and-contracts.md	Current behaviour: the e2e process-conformance suite (what is real, what is faked, what is asserted); issue-217 history row
cli/tests/test_pdlc_e2e/README.md	New — the fixture-set format, step vocabulary and expectation keys, written so adding a scenario requires reading nothing else (R4.3)

README.md, the docs site and SKILL.md are untouched with reason: no command, config key, workflow or user-facing behaviour changed — the deliverable is an internal test suite, documented where tests are documented (the capability doc and the suite's own README).

Capability docs

testing-and-contracts.md — updated in this PR (see the table above). No other capability's behaviour changed; process-graph.md is untouched because the suite tests the graph's documented behaviour without changing any of it.

Verification results

In testing-plan.md § Verification results, with evidence in evidence/verification.md.

Review cycles

Cycle	Type (self/critic/security)	Reviewer	Outcome	Link
1	self	the-loop (this session)	new finding — the module-level Requirement: in the test file's docstring shifted every scenario's requirement attribution by one row in the-loop scenarios (the extractor carries the last-seen Requirement: onto the next Scenario:). Moved each Requirement: above its Scenario:; attribution verified in the table output	test_pdlc_e2e_integration.py
2	self	the-loop (this session)	new finding — R1.3's "the execution log mirrors the transitions" was asserted only as section presence, not as the walk's checkpoints. Added the executionLogEntries expectation (ordered log-entry markers) and re-ordered the happy path so the log exists before the phases that checkpoint into it	runner.py, happy-path/scenario.yaml
3	self	the-loop (this session)	new finding — the node-trace and label-trail divergence reporters were never exercised on a failing path, so a bug there would let a regression pass silently. Added a meta-test that forges a phase-skipping event log and asserts both reporters fire naming the divergence. A fourth pass (requirements trace R1–R4 → mechanism → test; docs-altitude sweep for stale "no e2e" claims) found nothing new — stopped per reviews.stopOnNoNewFindings	test_pdlc_e2e_integration.py
4	critic	—	not run. reviews.critics is empty in this repo's harness config and no second harness was available to this session
5	security	—	mechanism-level review against the requirements' 3 abuse cases: fixture playback is inert data behind a closed step vocabulary (refusal tested); fakes sit at transport seams while authorization/loop-prevention run production code (positively asserted by the loop-prevention scenario); hermeticity is a raising tripwire, not a convention (tested). Test-only change, no new attack surface, no production code touched. Risk tier 3 — no named human sign-off mandated; the PR approval gate stands	requirements.md § Security considerations