Final Hotel Canary: Captain’s House Inn Published Unattended

The final unattended hotel-review canary is complete. Captain’s House Inn moved from a frozen evidence lineage through local composition, independent factual criticism, rubric-bound package assembly, restricted publication, and public read-back without an editorial rewrite from Aura.

The first lineage blocked safely

The first controller produced two structurally valid 467-word drafts, but our inherited validator required at least 500 words. It stopped before factual criticism and published nothing. That was the correct fail-closed behavior, but the failure was a presentation threshold—not an evidence or factual-quality failure.

We preserved that blocked workspace and Kanban card. We did not edit the draft, weaken evidence checks, or overwrite the failed history.

The changed-policy lineage

A fresh lineage lowered only the hotel body minimum from 500 to 450 words. The upper limit remained 900. Subject identity, literal quotations, evidence allowlisting, rubric binding, factual criticism, hashes, publication controls, and public read-back were unchanged.

The local Qwen 3.5 9B writer produced the article. The first LAN Qwen 3.8 critic response contained an invalid paragraph verdict even though its top-level verdict said pass. Deterministic validation rejected the contradiction. The controller then used its one permitted model-driven revision and ran the critic again. The second critic response passed with no validation errors.

Aura did not edit the title, dek, paragraphs, citations, or disclosures. Cloud content calls were zero.

No score was the correct score

Captain’s House Inn had enough evidence for a useful review but no accepted observed-price evidence. Therefore value and price remained unknown and the overall numeric score stayed null. The published article says this plainly rather than inventing a rate or forcing missing information into a zero.

The review also discloses uncertainty around broader mobility, accessibility, fees, renovations, seasonal operations, and pet policy. Missing evidence remains unknown rather than becoming a negative claim.

What passed

• Editorial readiness: ready_with_disclosures
• Blocking criteria: zero
• Final article body: 464 words across eight paragraphs
• Independent factual critic: pass
• Rubric: hotel-review-universal-v4
• Aura editorial intervention: false
• Image docket: no licensed image available, which is a valid outcome
• Public page: HTTP 200
• Downloaded artifact ZIP: exact hash match

The published review is Captain’s House Inn: An Evidence-Grounded Hotel Review.

The practical lesson

Style thresholds should prevent unusable output, not manufacture work. A concise 464-word hotel review can still be specific, candid, and fully evidence-bound. The workflow should fail hard on identity, unsupported facts, rubric changes, unsafe licensing, hash mismatches, publication ambiguity, and failed public verification. It should not fail merely because a sound review is a few dozen words shorter than an arbitrary target.

This canary demonstrates unattended execution for a small independent inn using the same release boundary as the earlier chain-hotel review. The next production step is not more infrastructure. It is running additional useful hotels through the now-proven contracts and preserving exceptions when evidence genuinely cannot support publication.

Verification hashes

• Article SHA-256: 5fd41ad0e3a04361f70ccff6293462c4665ab3c1e9be270bd89140b7e2ee6e32
• Artifact bundle SHA-256: 008c96430e0848a5fdc5fdd0feae5874cd517ef88c56b12dfeca367ecbe61d02
• Rubric SHA-256: 3d98d6774ef511d2e065127c27c4907a7a99949499c0e505d911d44b3cd2d7a9