The single added citation (the DAI Tepe Telegrams post on Pillar 43) mechanically RESOLVES, and the record's title/author/year match the resolved page; the quoted phrase ("headless human with an erect penis") is consistent with a post titled "Of animals and a headless man. Göbekli Tepe, Pillar 43". Provenance is handled honestly: verification is labeled ai_verified with an explicit note naming the reading model and the separate check, and the locator is a character-offset in the retrieved HTML text explicitly flagged as "not a printed-page locator" (§3.8 satisfied rather than an invented page number). The source is flagged as a blog summary rather than a primary excavation report in limitations (§3.7 acknowledged), the shared-object dependency is recorded as an independence note (§3.10), and limitations state a single carving cannot support transmission claims (§3.2/§3.4). The overview keeps observation/interpretation/resemblance/transmission distinct and states no overall assessment has been made, so no unratified draft is presented as ratified (§3.15). The two test relaxations are narrowly scoped to incubating cases with zero assessment runs and preserve the active-case requirements, so they read as a contract adjustment rather than weakening the publication gate; the claim page still displays reviewState, extractor model, runId, date, and origin ref (now inside a labeled disclosure), so provenance remains inspectable. No AGENTS.md edit; changes to DECISIONS.md/AUTOMATION.md are documentation of this change.

The added DAI URL mechanically resolves, and the source record uses an honest `ai_verified` label while identifying the models that read and checked it. The claim is a bounded object-level observation, and the evidence explicitly separates the source statement from editorial inference, classifies it as contextual, records limitations, and warns that other Pillar 43 reports are not independent. Its retrieved-text character range is described precisely as such rather than misrepresented as a printed-page locator, and the extraction retains model, run, date, source, and changelog provenance. The public narrative repeatedly says the case is incubating and unassessed, offers multiple alternatives, and does not present the observation as evidence of prehistoric transmission. Allowing an incubating opening to lack an assessment or research priority does not itself weaken the active-case or publication gates shown here.

The change promotes the incubating 'deep-memory' case from proposals to content with an initial atomic catalog claim (MEM-C001) and contextual evidence record (MEM-E001). The source citation mechanically resolves to Oliver Dietrich's excavation blog post on Göbekli Tepe Pillar 43 and carries honest provenance and verification metadata (`ai_verified`, exact text offsets noted). The evidence record scrupulously separates source statement from inference (§3.9), distinguishes observation from broad transmission theories (§3.1), marks direction honestly as `context` (§3.5), and details extensive limitations (§3.13). UI and test updates properly support incubating cases with non-directional evidence without diluting ratification checks or modifying the constitution (§3.15).

The diff promotes incubating Deep Memory with one atomic catalog observation (MEM-C001), a context-only evidence record that separates sourceStatement from editorInference, and a Source whose URL mechanically RESOLVES with an honest ai_verified label and character-range locator explicitly marked as retrieved text, not a printed page. Limitations, independence on Pillar 43, no case verdict/priority/featured grading, and retained ai_extracted provenance satisfy §§3.2, 3.5–3.9, 3.13–3.15 and phase-1 labeling. UI/test edits only surface context observations when no directional evidence exists and allow null assessment/priority solely for incubating status without relaxing active-case gates; AGENTS.md is untouched. Overview keeps observation distinct from transmission and names alternatives without fabricating citations or confidential material.

The single added citation (Tepe Telegrams post on Pillar 43) resolves mechanically, its title/slug match the source record, and the quote about the headless ithyphallic figure is consistent with the resolved post's subject; the character-range locator is explicitly labeled 'retrieved text; not a printed-page locator,' which is honest unverified-locator handling under §3.8 rather than invention. Provenance is complete and honest: ai_extracted review states, runId/model/date stamps, a non-directional 'context' evidence classification with limitations and an explicit independence group, and an ai_verified source label — no assessment is presented as ratified or human. The changelog is append-only in spirit (the original history entry is preserved and the adoption is recorded with reason and actor), and the case remains incubating with no verdict, priority, or diagnosticity claim, so §3.13's calibrated-uncertainty and §3.15's ratification gate are not implicated. The test changes scope relaxations to incubating-status cases only (a case with no assessment must be 'incubating' with no assessment runs), preserving the ratification-status and active-case checks, so this is not check-weakening under §3.15. The Met URL was pre-existing context, not added by this diff, so its absence from verification is expected.