Gap Scan — 2026-07-02 (models-scoped, OpenAI GPT)

On-demand model-robustness-check run (user-requested), distinct from the day's scheduled Gap-Scan. Scope: all OpenAI model pages in models/ plus a models-scoped gap hunt for missing OpenAI GPT models.

Scanned

Part A targets: models/gpt-4-family, models/o1-o3, models/gpt-oss, models/gpt-53-codex, models/gpt-54-thinking, models/gpt-55, models/gpt-56. Part B: wiki-wide broken [[models/...]] link analysis (mainspace), named-but-pageless OpenAI models, missing family versions, current-release check.

Gaps actioned (7 of 7 found)

New pages created (live)

• GPT-5 Family (OpenAI) — the GPT-5 → GPT-5.2 span (GPT-5, Aug 7 2025; GPT-5.1, Nov 12 2025; GPT-5.2, Dec 11 2025) had no page while GPT-5.3–5.6 each have one; [[models/gpt-5]] was broken (muse-spark.md) and GPT-5.1/5.2 are referenced substantively across concepts/, companies/, and the system-card baselines of the 5.3/5.4 pages. 3,417 words, confidence medium. Depends on queued sources: GPT-5 System Card; GPT-5.2 System Card update.

Pages expanded (live — Part A deepen, Operation: Update)

• GPT-4 Family (OpenAI) — 709 → 2,912 words; confidence high (kept); 3 corrections; 3 foundational sources queued.
• OpenAI o-series (o1 → o4-mini) — 815 → 3,676 words; high (kept); release-date and benchmark-provenance corrections; 3 sources queued.
• gpt-oss (OpenAI open-weight models) — 547 → 2,633 words; high (kept); gpt-oss-safeguard and MFT methodology added; 2 sources queued.
• GPT-5.3 Codex — 641 → 2,289 words; medium (benchmarks outside the 3-month window); Spark characterization corrected.
• GPT-5.4 Thinking — 459 → 3,760 words; medium; release date corrected 2026-03-01 → 2026-03-05; misattributed Harvey quote fixed; 1 source queued.
• GPT-5.5 ('Spud') — 601 → 2,931 words; medium → high (4+ current sources; fast-decay facts within window); 3 sources queued.
• GPT-5.6 (Sol, Terra, Luna) — 1,224 → 4,084 words; medium (6-day-old staggered limited preview); citation misattributions fixed; 1 source queued.

Link aliases fixed

• [[models/gpt-5]] → [[models/gpt-5-family]] (models/muse-spark.md, 2 instances)
• [[models/gpt-5-5-instant]] → [[models/gpt-55]] (concepts/ai-progress-measurement.md)
• [[models/gpt-5-5]] → [[models/gpt-55]] (companies/openai.md)
• [[models/claude-sonnet-4.5]] → [[models/claude-sonnet-45]] (sources/farahany-class-12-red-teaming-ai.md)
• [[models/sora-2]] → [[models/sora-veo]] (sources/farahany-class-14-when-anyone-can-fake.md)
• [[entities/openai-preparedness-framework]] → [[sources/openai-preparedness-framework]] (standards/nist-ai-rmf.md, 2 instances)

Queued — foundational sources (15 INGEST tasks, all authenticity-verified on canonical hosts)

• GPT-4 Technical Report (arxiv.org/abs/2303.08774) — INGEST-gpt-4-technical-report-2023.md
• GPT-4o System Card — INGEST-gpt-4o-system-card-2024.md
• GPT-4.5 System Card — INGEST-gpt-4-5-system-card-2025.md
• OpenAI o1 System Card — INGEST-openai-o1-system-card-2026-07-02.md
• o3 / o4-mini System Card — INGEST-o3-o4-mini-system-card-2026-07-02.md
• Deliberative Alignment paper (arxiv.org/abs/2412.16339) — INGEST-deliberative-alignment-paper-2026-07-02.md
• GPT-5 System Card — INGEST-gpt-5-system-card-2025.md
• GPT-5.2 System Card update — INGEST-gpt-5-2-system-card-update-2025.md
• OpenAI CoT-controllability evaluation — INGEST-openai-cot-controllability-eval-2026.md
• UK AISI GPT-5.5 cyber evaluation — INGEST-uk-aisi-gpt-55-cyber-evaluation-2026.md
• OpenAI "Where the goblins came from" — INGEST-openai-where-the-goblins-came-from-2026.md
• METR Frontier Risk Report Feb–Mar 2026 — INGEST-metr-frontier-risk-report-feb-mar-2026.md
• GPT-5.6 Preview System Card (deploymentsafety.openai.com/gpt-5-6-preview) — INGEST-gpt-56-preview-system-card-2026-06-26.md
• gpt-oss-safeguard Technical Report — INGEST-gpt-oss-safeguard-technical-report-2026-07-02.md
• Estimating Worst-Case Frontier Risks of Open-Weight LLMs — INGEST-estimating-worst-case-frontier-risks-open-weight-llms-2026-07-02.md

Corrections in other pages

• OpenAI — GPT-5.5 ship date corrected April 26 → April 23, 2026 (TechCrunch, CNBC, and OpenAI's launch post all dated Apr 23); the coupled "same day" ChatGPT Images 2.0 claim corrected to its verified April 21, 2026 date (openai.com/index/introducing-chatgpt-images-2-0/).

Non-gaps confirmed (alias-resolution discipline)

o3-pro, deep-research variants → covered by OpenAI o-series (o1 → o4-mini); GPT-4.1, GPT-4.5 → GPT-4 Family (OpenAI); GPT-5 Pro, GPT-5.5 Instant → GPT-5 Family (OpenAI) / GPT-5.5 ('Spud'); Sora 2 → Sora and Veo (Video Generation Models); GPT-5.x Codex variants → covered within GPT-5 Family (OpenAI) and GPT-5.3 Codex (no separate Codex-line page warranted).

Authenticity-verification failures

None. (Several unverifiable claims were excluded from pages rather than cited — per-page details in the agents' reports, summarized in the log entry.)

Deferred backlog

None models-scoped. Escaped-pipe table-cell wikilinks ([[models/gpt-55\|…]] etc.) register as false positives in naive broken-link scans; noted for the lint skill's scanner.

One-line summary

All 8 OpenAI GPT model pages now at Wikipedia depth (8 pages, ~25,700 words total, 1 created live), 7 link aliases fixed, 2 factual date corrections, 15 foundational sources queued for review — the OpenAI ingest queue is now substantial and worth a dedicated ingest cycle.