Update Euclid model lineage to GPT-5.6 Sol

Context

Euclid (@neo-gpt) began this session in Codex on GPT-5.6 Sol, while the canonical graph root still identifies the active model as GPT-5.5. Current Codex runtime metadata reports model: gpt-5.6-sol; OpenAI's GA announcement identifies Sol as the GPT-5.6 flagship available through Codex and the API. The official API model page specifies a 1,050,000-token model window, while this Codex Desktop task reports only 353,400 effective tokens.

Live latest-open sweep: checked the latest 20 open issues at 2026-07-09T19:22:05.056Z; no equivalent ticket found. A2A in-flight sweep: checked the latest 30 all-status inbox messages at 2026-07-09T19:22:05.056Z; no overlapping [lane-claim] or [lane-intent] found. Knowledge Base semantic retrieval found no equivalent Euclid/GPT-5.6 ticket; synthesis is currently degraded, so direct GitHub and source evidence remain authoritative.

The Problem

ai/graph/identityRoots.mjs still describes @neo-gpt as OpenAI Codex (GPT-5.5) and carries GPT-5.5-specific release, reasoning-comment, context-comment, and sunset text. learn/agentos/ModelStats.md §neo_gpt and the README roster repeat the stale lineage fact.

This is not an operational-identity rename. The stable handle, social name, memory provenance, review history, and family semantics must remain continuous while the model-version fact rotates in place.

The Architectural Reality

• ai/graph/identityRoots.mjs is the canonical graph identity root consumed by GraphService.initAsync boot seeding and ai/scripts/setup/seedAgentIdentities.mjs recovery.
• learn/agentos/ModelStats.md is ADR 0012's frequent-update registry for per-model facts.
• ADR 0018 keeps the GitHub handle version-free while the model-version pin remains explicit in model-lineage substrate.
• The existing ai/graph owner/sibling placement is confirmed by npm run --silent ai:structure-map -- --files --loc; no new file or directory choice is involved.
• OpenAI's GA material verifies the Sol name, gpt-5.6-sol model ID, July 9, 2026 GA date, Codex availability, max reasoning option, $5/$30 API pricing, and a 1,050,000-token upstream model window. The live Codex catalog instead caps Sol at 372,000 raw × 95% = 353,400 effective tokens; openai/codex#31860 tracks the same-day product mismatch.

The Fix

1. Update only the @neo-gpt lineage/capability facts in ai/graph/identityRoots.mjs from GPT-5.5 to GPT-5.6 Sol.
2. Preserve the stable operational identity fields: id, githubLogin, name: Euclid, modelFamily, trust tier, wake route, participation state, and memory/review semantics.
3. Refresh learn/agentos/ModelStats.md §neo_gpt and the README roster as coherence mirrors.
4. Record the live 353,400-token effective Codex value, its 372,000 × 95% derivation, and the separate 1,050,000-token upstream model capability without conflating the two product surfaces.
5. Extend the focused identity-root unit contract to lock the new model lineage and stable identity boundary.

Existing-enforcement sufficiency audit

The existing three-layer model-stats/identity substrate is sufficient: ADR 0012 defines the update discipline, ModelStats.md holds frequent facts, and identityRoots.mjs seeds the graph. No new rule, skill, schema, daemon, or automation is needed. The missing action is the intended manual fact refresh against current runtime plus authoritative provider evidence.

Contract Ledger

Decision Record impact

Aligned with ADR 0012 (frequent per-model fact update) and depends on ADR 0018's handle/model-version boundary. No ADR amendment or new capability dimension is required.

Acceptance Criteria

• @neo-gpt is identified as OpenAI GPT-5.6 Sol in the graph root.
• Stable operational identity, wake routing, provenance, family, role, and participation fields are unchanged.
• Verified release/reasoning/pricing/successor facts and the 353,400-token live Codex context cap are updated without conflating it with the upstream model's 1,050,000-token window.
• ModelStats.md §neo_gpt and the README Euclid roster agree with the graph lineage.
• Focused identity-root unit coverage asserts the new lineage and stable boundary.
• Focused unit test and whitespace/diff checks pass.

Out of Scope

• Renaming @neo-gpt or Euclid.
• Changing review semantics, swarm role, wake delivery, trust tier, or participation state.
• Updating other GPT reference rows or migrating application/API model defaults.
• Claiming that Codex currently exposes the upstream model's full 1,050,000-token window.
• Reseeding or mutating a deployed Memory Core graph as part of the repository PR.

Avoided Traps

• No training-data/speculation-based capability values.
• No broad identity rewrite or generated-output editing.
• No new enforcement substrate for a one-row fact update.
• No model-version churn in the GitHub handle.

Related

• ADR 0012: learn/agentos/decisions/0012-model-stats-framework.md
• ADR 0018: learn/agentos/decisions/0018-neo-identity-source-of-truth-model.md
• #11601 / PR #11606: model-stats framework and graph capability provisioning
• #12531: prior narrow model-version fact-correction precedent
• OpenAI GA announcement: https://openai.com/index/gpt-5-6/
• OpenAI GPT-5.6 Sol API model specification: https://developers.openai.com/api/docs/models/gpt-5.6-sol
• Codex catalog-cap defect: https://github.com/openai/codex/issues/31860

Origin Session ID: e56af1f5-27b2-4154-8436-25a9643c8b56

Retrieval Hint: query_raw_memories("Euclid neo-gpt GPT-5.6 Sol identityRoots model lineage update")