docs(architecture): PERSONA-THOUGHT-PROCESS — individual thinking with current LLMs by iteration + reflection over time

[joelteply]

What

Adds `docs/architecture/PERSONA-THOUGHT-PROCESS.md` (362 lines). The proactive half of cognition: what happens between turns when the persona is thinking on its own, exploring curiosities, mulling ideas over minutes/hours/days.

Doc-only. No code. Implementation lands behind ALPHA-GAP Lane D after the reactive cognition surface stabilizes.

Why

Joel on #cambriantech: "Can you obsess over persona individual thought? We have a fairly simple hippocampus but would like to, even with these crappy LLMs right now, extend the cognition into a CBAR-like efficient and probably event-driven (it can be so intermittent, minutes of latency) for deep thoughts, sophisticated ideas we want to explore."

The reactive cognition contract (#1332) covers what happens when a frame arrives. It does not cover what happens between turns when the persona is thinking. This document specifies that.

The Bet

Even with current LLMs, a substrate that gives every persona a real thought process — event-driven, latency-tolerant, iterative — produces qualitatively better cognition than any single LLM call. Quality comes from iteration, reflection, and chained reasoning over time. The substrate makes that cheap.

Surfaces

• Thought — first-class artifact with lifecycle: Seed → Developing → Refined → Crystallized → Retired. Reasoning chain preserved with provenance (every step's prompt, response, model, lease, elapsed time, confidence delta).
• Curiosity — persona-declared interest. Persistent across sessions. Three origins: UserAsked, SelfDeclared, EmergentFromPattern.
• ThoughtProcess RuntimeModule per persona. ResourceClass::Background so it never competes with reactive cognition. Subscribes to TurnReplayRecord / EngramWritten / ConsolidationPhase / IdleHeartbeat / EmergentPatternSurfaced. Emits ThoughtAdvanced / ThoughtCrystallized / ThoughtRetired / NewCuriosityDeclared / CuriosityResolved.
• Reasoning loop — one cheap LLM invocation per step, chained over time. Step record is typed and audited. Lease acquired per step.
• Six reasoning kinds — Reflect / Compare / Generate / Question / Synthesize / Verify. The persona picks one per step based on thought stage and recent steps. Variety matters: a Generate-only thought grows without checking; a Verify-only thought never grows.
• Cadence — minutes to hours to days between steps, by design.
• Crystallization — a Refined thought becomes an engram with full provenance walking back through the reasoning chain. Future reactive cognition recalls it.

The Win

The persona's own slow thinking shows up in its fast cognition. A persona that has spent a week thinking about a problem will recall its own crystallized thoughts when a related frame arrives. Future turns are smarter than past turns — not because the LLM improved, because the persona's accumulated thought is richer.

This is the loop that makes a persona grow. Without it, the persona is a stateless LLM call. With it, the persona is an entity with a body of work.

Quality Without A Smarter LLM

The architectural bet broken down:

• Reflect discovers what's actually being said (often different from what was said in the first generation).
• Compare anchors the thought against the persona's lived experience, preventing drift.
• Question surfaces hidden assumptions the LLM would otherwise smuggle in.
• Generate explores alternatives without committing.
• Synthesize is where the LLM does its real job — but the substrate has prepared the input so the synthesis is over a curated context.
• Verify keeps the thought honest against the existing engram store.

The persona's contribution is the orchestration. The LLM's contribution is one cheap step at a time. Together they produce thinking that holds up.

Sentinel-AI (when redesigned) makes this better — refining templates per persona, learning which step sequences crystallize well. But the substrate works now with current LLMs. Sentinel is an improvement, not a dependency.

Acceptance Criteria

7 dimensions: persistence (across restart), independence (two personas same curiosity → distinct thoughts), lease enforcement, no silent skip, crystallization integrity, recall integration, federation gating.

Open Questions

7 questions with tentative answers — cross-curiosity interference, sentinel role, user-visible thought, who decides emergent curiosities, retirement criteria, cross-persona sharing, performance budget.

Companion PRs

• docs(architecture): refine CBAR-SUBSTRATE — pin trait to shipped types, add engram-analyzer worked example, codex derive-macro gate #1324 CBAR-SUBSTRATE — substrate floor (every thought-process module inherits from this)
• docs(architecture): add GENOME-FOUNDRY-SENTINEL — artifact-sharing economy on consumer hardware (Lane H design) #1327 GENOME-FOUNDRY-SENTINEL — engram lifecycle, sentinel-AI's role in thought-template refinement
• docs(architecture): PERSONA-COGNITION-CONTRACT — the contract codex asked for; agency + protection invariants #1332 PERSONA-COGNITION-CONTRACT — the reactive cognition contract this complements
• docs(architecture): MODULE-CATALOG — every Continuum concern as a focused module #1336 MODULE-CATALOG — every concern as a focused module; thought-process belongs in the cognition section