# Originality and claim-boundary review ## Selected work - **Title:** *The Alias Is Not the Model: Prove Which AI Runtime Touched Facility Work* - **Proposed slug:** `ai-model-alias-runtime-identity-self-storage` - **Thesis:** A friendly model name, endpoint or alias does not establish which runtime actually produced an AI-assisted recommendation. Consequential workflows should preserve requested, resolved and approved runtime identities; bind them to the decision packet; and reduce authority when identity or eligibility cannot be proved. - **Practical tool:** `ai-runtime-identity-receipt.csv`, a 45-field template with one blank operator row and five explicitly fictional records covering an eligible canary, an ineligible fallback, a low-consequence approved fallback, an unavailable resolved identity and a model match invalidated by a prompt change. - **Fictional example:** Copper Vale Storage and every facility, model, provider, router, system, person, record, timestamp, classification, test and outcome in the article or tool are invented. ## Duplicate preflight - The `2026-09-06/evening-ai-architecture` directory was absent before this run. The same-date morning camera-outage and midday data-freshness packages already existed and were preserved. Exactly one evening article was created. - A preliminary human-review-capacity direction was rejected before drafting because ED-024 is the published canonical *Designing a Human Review Queue That Does Not Become a Bottleneck*, MED-029 is a sent external article proposal on qualified human-review capacity, and MED-034 is a scheduled guest conversation on the human-review queue. No version of that candidate remains in this package. - A broad backup-model substitution title was also rejected after external screening surfaced current generic articles using closely related fallback and substitution-gate language. The selected work is narrower: request-time runtime identity, mutable aliases, resolved-model evidence and authority degradation when the serving identity cannot be established. - Recent evening articles cover whole-path decision builds, pre-execution authority, semantic content boundaries, command arbitration, reversibility, score calibration, alert-to-work-order handoff and failed-work recovery. The selected thesis addresses hidden runtime resolution during routing, fallback and alias reassignment. - The nearest governed work is *The AI Decision Build*, which versions a release's complete operating path. This article begins at inference time and distinguishes the model requested, the model reported as serving the request and the model approved for that exact task and authority mode. A matching model identity still does not prove that a changed prompt, retrieval build, tool schema or policy remains eligible. - Current Authority Control Room readback covered 46 calendar records (SHA-256 `9f18ed50f4f846239c051262ea42579dfca478331d9ec4c129307af715c0b6ac`), 75 editorial/media records (`efdb48315eee013304090d7f6a55a515fc4f968e531c117a64c1520f3077daee`), 358 submission records (`1becc54e236a1c3c0aaf8f242eab628af15ce36c40f90461b4943542695c6083`), 343 published-asset records (`f21c8b30b9ce213b35d2c7b8ffe824bbe9674a012a9bb13de36df9a949d46ede`) and 551 evidence records (`ed4f2d97c78f6b1572b9ec7d0d694a6d3681c556c2eaa85c3340889a859bfe8d`). None contained the selected title, slug, Copper Vale scenario, practical-tool filename or runtime-identity-receipt thesis. - The live Jared sitemap returned HTTP 200 with 58 canonical URLs and SHA-256 `539bd602bece313a102d85150d2b1d3965595661198133856503427ccf58c7b4`. The live modSTORAGE post sitemap returned HTTP 200 with 139 post canonicals and SHA-256 `fe0a85a05ad2f43f7cd663b38047b75d83a0f6d66885df7e0d8090745feb39d1`. Neither contained the selected slug. - Both proposed owned URLs returned HTTP 404, and the WordPress exact-slug REST query returned an empty array. Exact public searches for the title, slug and tool filename found no matching work. - Public technical writing does discuss mutable model aliases and model routing. The package makes no claim to invent aliases, runtime metadata, model registries, shadow tests, canary releases or request-versus-response telemetry. Its distinct contribution is the self-storage operating method, three-identity control, authority-mode release sequence, fictional scenario and 45-field runtime receipt. ## Local originality screen - The bounded editorial-execution corpus included 290 unique-content Markdown files after excluding this package, dependency directories, Git data, compiled distributions, build and render output, archives, generated authority-site copies and duplicate control-room worktree copies. - Zero normalized body-sentence matches of 14 or more words were found across 69 target sentences. - The article contained 2,240 unique normalized six-word shingles. The maximum overlap with another file was four shingles, or 0.179%, against the August 29 semantic-firewall article. The shared phrases were generic NIST and fictional-example language, not a reused thesis, structure, scenario or practical tool. - Generic-language scanning found no use of `in this article`, `delve`, `leverage`, `game-changer`, `seamless`, `transformative`, `as an AI`, `let's dive`, `it is important to note`, `in today's fast-paced`, `unlock the power`, `navigate the complexities`, `real-world example` or `imagine a world`. ## Claim boundaries - The article does not claim a real model, alias, router, provider, fallback, facility, employee, customer, note, work order, runtime mismatch, release, incident, deployment, performance result or operating outcome. - It does not claim that Jared, modSTORAGE or Facily.ai operates the proposed architecture or achieved reliability, safety, customer-service, financial, occupancy, adoption, latency, quality, certification, award, coverage or recognition from it. - Model, endpoint and alias configuration are not represented as action-time runtime proof. Provider-reported runtime metadata is preserved as a reported value, not independent proof of behavior or correctness. - Schema validation, API success, trace capture, model identity match, shadow completion, canary traffic and release approval remain separate states from semantic correctness, decision authority, action execution, governing readback and reconciliation. - NIST AI RMF 1.0 is voluntary and currently being revised. NIST AI 800-4 is a cross-sector monitoring landscape report that says the field's practices and terminology remain nascent. Neither source certifies the proposed receipt or prescribes a self-storage architecture. - Google model-cards research, MLflow, JSON Schema, AWS SageMaker and OpenTelemetry sources support only the documented research concepts, registry mechanics, schema vocabulary, deployment techniques and development-state telemetry fields. They do not validate the article's self-storage method or prove an implementation. - The runtime receipt is an adaptable operating tool, not a security standard, privacy assessment, provider contract, model evaluation certification, legal opinion or authorization to process customer, payment, access, employee or facility data. ## Assistance, imagery and rights - AI assistance supported research synthesis, drafting and editorial QA. Legitimate source, authorship, image ownership and assistance disclosures are preserved. - Governed image 052 is disclosed as a Jared-owned AI-generated editorial illustration, not documentary evidence. It was visually reviewed and copied byte-for-byte from the high-resolution master. - Prior use in the August 27 decision-build and September 1 score-calibration packages is disclosed. No prior title, body, fictional scenario or practical tool is reused. - No author portrait was used. The approved navy-blazer v2 portrait was not needed, and the rejected black-turtleneck portrait was not used. **Truthful state:** Current governance authorizes one reversible publication to Jared's personal authority site after exact source, rendered, accessibility, deployment and public-readback gates pass. This review does not establish a repository commit, deployment, live publication, search submission, crawling, indexing, ranking, coverage or recognition.