Visual identity elements — evidence

Approval recorded, 5 October 2026: all 4 recommendations on the review page were accepted. The preparation and consistency notes below are historical evidence, not current pending status. This update records meanings only; it does not repeat the live probes or certify implementation.

Bounded review, 5 October 2026. Four independent additions from B-02 are proposed, not approved. The original question and its remaining visual entries are preserved.

Prior work and scope

Read the complete current Brand & Identity review and the original August 16 dictation, recovered the source B-02 card, and checked the identity extraction, product identity, source-claim and design-scope decisions. The original source spreadsheet was read in the preceding preparation. Extraction authorization is not approval of all extracted terms. Naming rules favor clear names rather than unresolved slash pairs.

The old bundled question treats twenty-two rows as one choice. This packet separates four additions: Topic Color Theme, Gradient System, Shape System and Mascot / Character System. It proposes Themed Color Palette and Brand Character System as clearer names for two of them. These are proposed identity elements, not additions to the whole-piece type list. The other eighteen visual entries remain for separate reconciliation and review. Nothing about their source spreadsheet makes them implicitly approved.

Source inspection

GitHub's current color page was read and its themed application examples inspected in the browser during preparation. The page explicitly distinguishes its general palette, Copilot theme and Security theme. It supports contextual assignment, not classification by hue alone.

The current mascot page was read and its Mona renderings inspected. It names characters, describes usage and distinguishes historical styles. Current spelling is Ducky; the older extraction says Duckie. This packet does not silently turn that discrepancy into an approved alias or repair stored names.

Rendered the existing 73-page GitHub Brand Guidelines 2025 PDF to low-resolution images to locate the relevant sections. Contact sheets were used for navigation, not certification that every page was read. Then rendered and visually inspected full pages 37, 42 and 43 at a readable size. Page 37 describes background-gradient options and directions. Pages 42–43 describe Git-Grid construction and a shape library used in artwork. The reader page includes two dated excerpts so the examples can be seen on a phone. It does not claim that the 2025 rules remain current.

The current illustration webpage was read as a lead, but is not needed to establish the older Git-Grid example. No motion interval was classified. The original PDF, rendered excerpt hashes and observations are recorded in visual-identity-source-observations.json. Selected live queries and local source excerpts are in visual-identity-preflight.json.

Selected system checks

Column inspection preceded registry and guideline queries. Existing registry labels include Gradient, Mascot Character, Gradient Wash and Abstract Gradient in different vocabularies. Those observational labels do not establish an identity-system classification. The selected guideline rows include Color Palette and Mascots sections. Sampled descriptions are stored assertions; their wording was not accepted as independently verified brand truth.

The selected sections have no value in their source-URL column. This is a provenance follow-up, not proof that the wider system holds no evidence. The inspected local onboarding prompt allows training knowledge when extracted content is absent. Its writer accepts an optional source URL and creates unpublished sections. No causal link from that writer to the sampled published rows has been established.

The local consumer hook selects published sections; the profile can derive sections from entity metadata when no sections are available. The local force-replacement branch deletes existing sections before inserting replacements. Those paths need full evidence, publication, deletion and human-correction review. They were read, not executed.

A confirmed local-reader/live-schema mismatch is separately tracked: the local Ask route selects content from brand_guideline_sections, but the saved live column probe has no such column. That block does not inspect the query error before using an empty fallback. This does not prove a deployed incident. The repair audit must inspect the active route, all prompt builders, authorization and coverage reporting before choosing an implementation. No guessed-column substitution was made.

Architecture and alternatives

Keeping all four as loose descriptions is simpler, but weakens explicit comparison of systems. Turning every observed gradient, shape or character into a system overstates the evidence. The recommendation selectively defines identity elements while retaining ordinary visual observations and separating source-stated rules from inferred analysis.

Check relationship direction and multiplicity: one document can define several elements; an element can be demonstrated by several pieces; one piece can demonstrate several elements. A product may have an identity distinct from its company. The source publisher, credited designer, identity owner, depicted character and depicted product are not interchangeable roles. A guideline version, campaign-specific application and long-lived identity also need different scope and time evidence.

Collection should retain the source and relevant part. Extraction should distinguish stated rules from observations. Validation should require the basis for the stronger identity claim. Writes and publication should preserve corrections. Search, trackers and Ask should distinguish explicit guidance from matching examples and link to permitted evidence. Monitoring should add dated changes rather than overwrite the meaning of an old capture.

These are audit requirements, not a physical schema proposal. Full keys, constraints, indexes, access rules, correction ownership, scheduling and deployed end-to-end behavior remain for the comprehensive verification pass. Only evidence that changes these four semantic recommendations blocks this packet; unrelated engineering defects do not restart the decision pass.

Prepared 5 October 2026.

Decision-consistency update — 5 October 2026

The Motion entrance and six verbal/strategy distinctions are now approved in the decision record. The reader page links to those rulings and limits remaining work to unreviewed entries. Its four visual-element recommendations remain pending and unchanged. The original broader identity bundles are not fully approved. Earlier source and system checks retain their preparation scope; this update does not reverify live schema or deployed behavior.