This is a housekeeping cover-image swap with no citations (verification tooling found none) and no edit to AGENTS.md. The image record is honestly labeled: source `generated`, license and credit state it is AI-generated editorial artwork made for the founder's lab site and brought here unchanged, and the model field says plainly 'OpenAI built-in Imagegen (model identifier not exposed)' rather than inventing a model name — the opposite of fabricated provenance (§3.8). The history entry records what changed, why, who acted, the channel and date of the founder's direction, the upstream repo, commit 5dd8e043 and an md5, satisfying §3.14 and the runId/model/date stamping in §3.15/§4. Founder-supplied material enters under the permission rule; the direction to use it, recorded with date and channel, is that permission, and the image is published as editorial art depicting an idea, not as evidence. The style-table row and script tone set are updated to match the picture actually used, so the documentation does not misdescribe the asset. The alt text and prompt are mutually consistent (bird above a shore, layered earth), so I see no internal contradiction that would indicate invention.

The replacement is recorded with specific provenance: repository, commit, content hash, generation date, model disclosure, run ID, full prompt, operator, and dated founder direction. The founder’s recorded request to bring the image here supplies permission under §3.15, while the metadata clearly labels it as AI-generated editorial artwork rather than evidence or a human-produced assessment. The history preserves the retired cover’s run ID and explains the change instead of silently rewriting the record, satisfying the relevant version-history requirement. No citation failure, confidential-source exposure, constitution edit, or weakening of publication checks appears in the diff.

The pull request replaces the Deep Memory editorial cover image with an AI-generated artwork previously created for the founder's lab site, following documented founder direction. The image metadata in images.yaml is properly updated to reflect its AI-generated status, model origin, run date, and prompt, fully complying with image styling guidelines and provenance requirements (§3.8, §4). An append-only history entry is added to history.yaml documenting the swap, its reason, and the retirement of the prior cover record. No core claims, evidence, or citations are modified, nor are any constitutional checks weakened.

The diff only swaps the Deep Memory cover binary, its images.yaml register entry, matching tone rows in IMAGE_STYLE.md and generate-case-art.mjs, and an append-only history line. Provenance is stamped honestly (lab origin and commit, md5, style-v2, runId, full prompt, model labeled as identifier-not-exposed rather than invented, credit that it is AI editorial artwork brought over at founder request). Founder direction is recorded with date and channel per §3.15; AGENTS.md is untouched; no citations, confidence material, or check-weakening appear. Generated cover remains non-evidential editorial art as IMAGE_STYLE requires.

The change swaps a cover image under a founder direction that is recorded with date and channel ('in session 2026-09-09') in the history entry, which under §3.15 is exactly how founder-supplied material becomes publishable — the direction to use his own lab-site image, so recorded, is the permission. Provenance is preserved exactly (source repo, commit 5dd8e043, md5, original run id, model label, full prompt), satisfying §3.8 and §3.14; the image stays labeled as AI-generated editorial artwork depicting an idea, not evidence, so §4's honesty rules hold. The edit to docs/IMAGE_STYLE.md is not an edit to AGENTS.md, so §3.15's constitution rule is not triggered, and no check is weakened. Nothing in the diff is internally inconsistent; the unverifiable session and commit hash are carried-over locators for post-training-cutoff facts, not positive evidence of fabrication.