Validate 1997 Toyota RAV4 4WD multi-component job planning

[lucronn]

Goal

Make this exact request work end to end for the canonical vehicle 1997 Toyota RAV4 2-door 4WD 2.0L:

oil pump, water pump, timing belt, and power steering pump replacement procedure

Technical acceptance status

• Resolve year, make, model, drivetrain, and vehicle/provider identity to one canonical vehicle record.
• Translate natural-language component terminology into safe AutoData component intents.
• Read the vehicle article list first and fetch only the matching article-detail and labor resources needed for the request.
• Normalize fetched articles, preserve source evidence and guide images, and persist idempotently.
• Calculate combined labor using stable operation IDs so shared work is counted once.
• Produce a structured combined procedure with source article IDs and evidence IDs on every step.
• Persist the combined procedure as an immutable derived article/revision.
• Repeat the same request from the local database without repeating AutoAPI calls.
• Fail closed when source content is unavailable or unauthorized; never fabricate procedure, labor, or vehicle facts.
• Preserve complete removal and reassembly/installation coverage in the consumer guide and PDF.
• Display the exact selected vehicle variant in the consumer response and PDF.
• Detect provider-summary artifacts in the consumer evaluator.
• Classify falsely complete replacement responses without both phases as blocking fail.

Current measured verification

• Local implementation sequence ends at fa76df0; runtime fixes are 256a42f and 87ca230.
• Full local suites pass: developer 74 passed, 2 subtests; autonomy 37 passed; ingestion 386 passed, 3 skipped, 12 subtests; Go API and shared-contract suites pass.
• Fresh positive report-only matrix: 5/5 cases pass at score 100 with zero findings and zero issue actions. Report SHA-256: 33539182a7b02eaaf3df09716bf744bef3ef8208ad817e8d8e1138db80d1b038.
• Fresh negative battery case: aggregate fail, expectation_met: true, no PDF, and a blocking high-severity missing-phase finding. Report SHA-256: fd7de16d67d87aa87a1088211804c598ac05caea36d96369d82f2ca4939e681d.
• The source oil/water pump PDFs were visually audited page-by-page; the generated RAV4 guide contains the full installation sequence through step 127.
• The RAV4 PDF cover visibly identifies 1997 Toyota Truck RAV4 2-Door 4WD L4-2.0L (3S-FE).

Open acceptance item

• Human technician review of stored RAV4 source evidence and procedure content. Automated readiness confirms structure, source linkage, consumer rendering, and fail-closed behavior; it does not approve safety-critical content.

Tracking

• Plan: docs/superpowers/plans/2026-09-15-negative-completeness-gate.md
• Vehicle-label plan: docs/superpowers/plans/2026-09-15-vehicle-specific-guide-label.md
• Consumer-agent plan: docs/superpowers/plans/2026-09-15-consumer-agent-artifact-detection.md
• Project: AutoData Portfolio

Safety boundary

No source credentials or user-owned sample data/, output/, or tmp/ artifacts were committed. No GitHub push was performed.

[lucronn]

Checkpoint — 2026-09-10

• Local ingestion tests: 225 passed, 12 subtests passed.
• Exact query parser verified for 1997 Toyota RAV4 4WD: oil pump, water pump, timing belt, and power steering pump resolve to four canonical component intents.
• Deterministic normalized-catalog replay produced status=ready, four selected article IDs, total_labor_hours=9.0, overlap_hours=0.0, and four procedure steps.
• Dashboard/API request reached http://127.0.0.1:8080/job-plans but the real source fallback is currently blocked because AutoAPI is not listening on 127.0.0.1:3000; the API returned 502 and did not fabricate source facts.
• GitHub verification: Autonomous Verification run 34489152451 passed at commit 1d2fc670e060b0be9e67ce8498946ddc0addb597.

The issue remains In Progress until the authorized AutoAPI service/session is available and the first-ingest plus second-request cache-hit smoke is measured.

[lucronn]

Live retry — 2026-09-10

• AutoAPI process is now reachable: GET http://127.0.0.1:3000/healthz returned 200 / {"status":"ok"}.
• The first catalog request, GET /v1/api/years, returned 502 upstream_auth_failed: Authenticated MOTOR session could not be validated.
• The exact RAV4 dashboard request therefore still returns 502 ingestion dependency failed before vehicle/article hydration. No vehicle facts, labor, or procedure were fabricated.

Next live checkpoint requires refreshing the authorized MOTOR session used by the AutoAPI service, then rerunning the same request.

[lucronn]

Live AutoAPI job-plan verification

Verified locally on commit da3570e686a303b491a1e17e8af53ed9d3b03cf9 using the running Compose stack and authorized local AutoAPI service.

• Vehicle: 1997 Toyota RAV4 4WD (US)
• Query: oil pump, water pump, timing belt, and power steering pump replacement procedure
• Cold path: ready; targeted Motor traversal; 1 vehicle family; 1,098 normalized catalog records; 4 procedure bodies and 4 matching labor resources fetched.
• Selected procedure IDs: P:552127574, P:553007043, P:552818463, P:552314723.
• Labor: 16.0 standalone hours, 0.0 overlap hours, 16.0 total hours, confidence 1.0.
• The shared Engine Oil Pump Drive Belt R&R step is published once and linked to both oil-pump and power-steering-pump source articles/evidence.
• The composed procedure revision was persisted with source article IDs and evidence IDs.
• Warm path: ready, cache_hit: true, source.mode: derived_article_cache, same 16.0-hour result, with no AutoAPI fallback.

Implementation details fixed in this delivery:

• AutoAPI model routing uses the displayed make name and resolves engine-specific provider IDs from the YMME response.
• Procedure (P:) rows are kept distinct from labor (L:) rows and joined by normalized title.
• AutoAPI mainOperation.laborTime and included-operation shape is normalized into the labor contract.
• Provider trim/engine display suffixes no longer create a duplicate canonical vehicle family.
• Replayed source snapshots receive distinct immutable model/powertrain IDs.
• Persisted combined procedures are checked before any source fallback, keeping repeat lookups fast.

Remote verification: Autonomous Verification run 34493636840 is still running its live Compose smoke.

[lucronn]

Mercury-2 compiled overlap procedure verified

The dashboard was previously showing the persisted deterministic revision. The live generation path was corrected and verified on commit 1ab46820f8100f99d0d1150064e9b7d09c93e13b.

For 1997 Toyota RAV4 4WD and the multi-component request for oil pump, water pump, timing belt, and power steering pump replacement:

• Cold lookup: llm_status: generated
• Procedure generation: mercury-2
• Procedure title: Vehicle Repair Procedure
• Five LLM-worded steps were published.
• The shared Engine Oil Pump Drive Belt R&R step appears once with both oil_pump and power_steering_pump components.
• Every generated step has valid fetched source article IDs and evidence IDs.
• Labor remains 16.0 total hours with 1.0 confidence.
• Warm lookup: cache_hit: true, source.mode: derived_article_cache, procedure.generation: mercury-2, five steps, and no AutoAPI fallback.

The underlying fixes include the official Inception API default (https://api.inceptionlabs.ai/v1), evidence validation for labor-operation evidence, regeneration of incomplete deterministic/LLM cache revisions, and preservation of shared included operations during composition.

[lucronn]

Verified after dashboard fix on commit 7bd1264: the 1997 Toyota RAV4 request for oil pump, water pump, timing belt, and power steering pump returns a cached Mercury-2 compiled procedure with 5 steps, including the shared Engine Oil Pump Drive Belt R&R step covering oil_pump + power_steering_pump. Labor is 16.0 total hours with 0.0 additional overlap hours. The dashboard now renders every procedure step (no four-step truncation) and labels the result Generated by mercury-2. Local Go tests pass; remote CI is running for this commit.