# Editorial QA — evening AI architecture lane

## Deliverable

- **Calendar date:** 2026-09-01
- **Lane:** Evening AI architecture
- **Title:** *A High Score Is Not a Shared Standard: How Self-Storage Teams Should Calibrate AI Across Facilities*
- **Author:** Jared Mastroianni
- **Proposed slug:** `ai-score-calibration-across-self-storage-facilities`
- **Thesis:** Identical AI scores or thresholds do not support a common portfolio conclusion until the decision, outcome definition, evaluation population, model version, score meaning, threshold evidence, error costs and release authority are aligned.
- **Practical tool:** `site-ai-score-calibration-review.csv`, a 12-gate site-level review from decision scope through correction and rollback.
- **Body word count:** **1,696**. Counting rule: from the first paragraph after the byline through the last article sentence before the horizontal rule; headings and list text included; title, deck, byline, proposed slug, publication-state note and source footnotes excluded; Markdown control characters and footnote markers removed; whitespace-delimited count with hyphenated compounds counted as one word.
- **Article SHA-256:** `171ceceed3626a639fc30ec8d91c10c6ba841604758ac85bbd756e7d20145cb0`.
- **WordPress body SHA-256:** `090ed35105da1f83bab8c49933493608e063416b723d3e2d36be34c1eee13172`.

## Corpus and duplicate review

- Authority Control Room preflight used exact commit `0c8a9d2557106ff3d7cde6ce478abf38d9f20eba` with 42 calendar rows, 71 editorial/media routes, 328 submission/activity rows and 521 evidence rows.
- The package is distinct from the Missing Denominator Test, reproducible benchmark protocol, Temporal Authority Envelope, Facility Observation Coverage Map, Semantic Drift Register, Facility Intervention Contract, AI decision handback and all current external source-offer premises.
- The proposed canonical is `https://blog.modstorage.com/ai-score-calibration-across-self-storage-facilities/`. No publisher action has occurred.
- Exact title, slug, fictional group and tool-name checks found no prior published, prepared, submitted or assigned work outside this package.
- A normalized screen compared 50 article sentences of 14 or more words with 1,513 current Markdown files outside the package and found zero exact matches.

## Source and factual review

- `source-register.csv` contains six unique primary or authoritative records in 11 consistent columns.
- All six exact URLs returned HTTP 200 on September 1, 2026.
- NIST AI RMF 1.0 is identified as voluntary, sector-agnostic and under revision; it is not represented as certification or a self-storage rule.
- NIST AI RMF Core supports deployment-condition and domain-expert evaluation boundaries without supplying a threshold.
- Google's machine-learning pages are used only for metric and threshold mechanics, not as a self-storage standard.
- scikit-learn documentation supports the definition and evaluation of probability calibration; it is not proof that any model is calibrated.
- NIST SP 1270 is used only for its broad technical-and-human bias boundary.
- Ridgefield Storage Group, all nine facilities, scores, labels, thresholds, work orders and review outcomes are explicitly fictional.

## Voice, claims and structure

- The package is a complete 1,696-word article, not an outline or strategy memo.
- Eight sections move from the operating decision through labels, populations, calibration, thresholds, a fictional site review, the 12-gate tool and the shared-threshold release rule.
- No real facility, customer, employee, work order, account, system, vendor, product, deployment, accuracy, performance, revenue, occupancy, savings, award, coverage or recognition claim is made.
- Jared's roles and tenure are not material and are omitted.
- Model output is treated as analytical evidence, not observed facility fact or authority for consequential action.

## Tool, HTML and image QA

- `site-ai-score-calibration-review.csv` parses as nine consistent columns and 12 unique gates. SHA-256: `2f404fcb9ab983e5a1db6cb50db80a577ad7026a4158dca1802b1d7a64c5b2e5`.
- `wordpress-body.html` parses successfully and preserves the deck, 1,696-word body, eight article headings, lists, six linked source callouts, six source notes and return links. It excludes the H1, byline, proposed slug, local publication-state note and Markdown footnote definitions.
- Public WordPress REST readback confirmed the exact existing category and tag IDs recorded in the publisher manifest; no taxonomy object was created.
- Image 052 is a byte-identical 3,840 × 2,560 RGB PNG. Source and package SHA-256 values both equal `5aa6a224294c7426eba866e92c569e6216ff2ea0d05820e1426977ebebe293e1`.
- Pillow reports only an aspect-ratio metadata field. Raw-byte scanning found no governed AI-system, generator, prompt or workflow identifier.
- Visual inspection confirmed a clean three-person design-review scene with no legibility, anatomy, crop or implied-result defect.

## Current state

Three separate internal editorial lenses scored the package 9.6, 9.7 and 9.8, with visual QA at 9.7. They are internal publication reviews, not external endorsements.

**Truthful state:** Complete local package. No WordPress draft, media object, schedule, publication, canonical activation, search submission, crawling, indexing, coverage or recognition has occurred.