# Final editorial QA — evening AI architecture

## Delivery facts

- Calendar date: 2026-08-27 (Europe/Zurich)
- Slot: evening AI architecture
- Title: *The AI Decision Build: Version the Whole Operating Path, Not Just the Model*
- Proposed slug: `ai-decision-build-version-whole-operating-path`
- Thesis: a model identifier cannot govern an AI-assisted operating decision; the releasable unit must bind the model, instructions, retrieval, facility context, semantics, policy, tools, human review, evaluation, activation, and observation as one versioned decision build.
- Practical tool: `ai-decision-build-manifest.csv`, a 16-control release gate
- Article body word count: 1,940 words
- Counting rule: tokens containing letters or digits; internal apostrophes and hyphens remain one word; title, deck, byline, role line, disclosure, endnotes, and HTML boundary comments excluded

## Refresh and duplicate check

- Automation memory: absent before this first run; no prior completion record.
- Current-day output directory: absent before this run; no morning, midday, or evening package existed for 2026-08-27.
- Other daily lane memories: no memory files were present for the morning operator or midday systems lane.
- Editorial calendar: `/Users/jared/Documents/Codex/2026-08-22/this-is-a-continual-editorial-execution/outputs/editorial-calendar.csv` refreshed; 28 scheduled rows reviewed.
- Live public corpus: `https://jaredmodstorage.github.io/writing/` and its live sitemap refreshed on 2026-08-27; 30 article canonicals were visible, including current correction-propagation, semantic-drift, facility-intervention, facility-observation, human-review, counterfactual, temporal-authority, policy, memory, observability, and event-spine work.
- modSTORAGE public corpus: search results and the known owned canonical for *When Automation Should Stop* reviewed; no title, slug, or practical-tool collision found. The attempted standard WordPress post-sitemap route returned HTTP 404, so search discovery and the known canonical were used rather than treating that sitemap route as complete.
- Active or submitted manuscripts: the Facilities Management Advisor operational-memory submission, Commercial Property Executive held change-impact manuscript, Inside Self-Storage event-bus manuscript sent 2026-08-27, IFMA FMJ accepted comparison-contract revision, and local architecture packages through the semantic-drift and correction-propagation work were screened.
- Exact duplicate result: no matching title, proposed slug, thesis, fictional scenario, or 16-control decision-build manifest found.
- Closest published article: *Observe the Decision Chain, Not Just the API Call* records which versions contributed to one trace. This article governs the frozen multi-component configuration that is eligible for evaluation and release before that trace exists.
- Closest unpublished/assigned work: the facility change-impact graph maps release blast radius; the semantic-drift register governs meaning changes; the correction-propagation contract governs downstream remediation. This article does not reuse their primary tool or thesis.

## Originality check

- No copied passages, quotations, customer anecdotes, deployment descriptions, or performance claims.
- The “AI decision build” framing, sixteen-field manifest, seven-gate release sequence, and `ADB-17`/`ADB-18` scenario were authored for this package.
- The SLSA build-provenance structure is identified as an analogy and bounded from decision authority and operating truth.
- The fictional two-facility scenario and all identifiers are explicitly labeled fictional.
- Quantified local-corpus screen: 217 Markdown files checked after excluding builds, dependencies, renders, and this package; no exact sentence match of 14 or more words; highest six-word-shingle overlap was 0.101% against the facility change-impact paper. This is a local editorial collision test, not a universal plagiarism finding.

## Claim-boundary check

- No customer, partner, integration, product-availability, certification, adoption, award, revenue, occupancy, maintenance, safety, accuracy, productivity, or financial-result claim is made.
- Model output remains a recommendation, not an observed facility fact.
- Software provenance, evaluation, authorization, activation, technical activity, provider acceptance, source readback, correction, and operating outcome remain separate states.
- NIST guidance is described as voluntary and the AI RMF 1.0 revision state is disclosed.
- OpenTelemetry stability and repository-move limitations are disclosed.
- The image is labeled AI-generated editorial content and not documentary evidence.

## Voice and style check

- Byline and verified roles are correct: Chief Operating Officer of modSTORAGE; CEO and Founder of Facily.ai.
- No tenure statement, fake quotation, self-aware drafting language, generic AI boosterism, or repeated disclaimer cadence.
- Direct operator consequences follow each technical control.
- Body length is within the requested 1,800–2,500-word range.

## Publication-state ledger

- Draft: complete and publication-ready locally.
- Publisher handoff: not performed.
- Published: no.
- Public canonical: none assigned.
- Indexed: no evidence.
- Third-party coverage: none claimed.
- Recognition or award: none claimed.

## Final editorial disposition

**PASS — complete local editorial package.** The article, tool, source register, governed image record, originality review, claim-boundary review, and final QA are present. All seven source URLs returned HTTP 200 through their current canonical or DOI resolution on 2026-08-27. The packaged image is byte-identical to the governed master at SHA-256 `5aa6a224294c7426eba866e92c569e6216ff2ea0d05820e1426977ebebe293e1`. The truthful state is publication-ready draft only.