# Anakai Kai Coaching Mode — Terry Handoff Bundle v3 (2026-06-11)

Self-contained snapshot of the Behaviour Graph v0. Supersedes the 2026-06-10 (v2) bundle. Ch.1 content is unchanged from v2; **ch.0 is a different chapter now — discard the v2 copy entirely.**

## What's new since the v2 bundle

- **Ch.0 fully rebuilt from scratch.** The original ch.0 predated several pipeline amendments (kai_intent step enrichment, schema v1.1, dialogue layer, orchestrated adversarial review). It was torn down and re-extracted fresh on 2026-06-11. Same 9 node ids and slot semantics (your cross-references hold), all wording re-derived.
- **Ch.0 scope extended from pp.9–11 to pp.9–28.** The book's three framing chapters are now folded in. This adds the layer the chapter was missing: the *why-this-profession* case. Ch.0 is now Kai's **onboarding layer**, not just framing:
  - `P-00-05` — the better-way frame (Perfect-Career-List + five-careers teardown)
  - `P-00-06` — the New-Economy / performance-pay argument
  - `P-00-07` — poser / amateur / professional, as a framework lesson (ch.1 still owns the operational mappings)
  - `P-00-08` — earn-while-you-learn, paired with new `AP-00-04` (passion-without-skill)
  - `L-00-01` — Perfect-Career-List walk (career-frame grounding loop) → new state `S-evaluating-career-frame`
  - `L-00-02` — **the Go-Pro decision loop.** This closes the previously-open forward reference: `S-luck-seeker → S-skill-committed` is now a real extracted transition, and `L-00-02 --precedes--> L-01-01` hands off directly into ch.1's list-dump. Its final step produces the named first skill — the exact signal `S-skill-committed.entry_criteria` needs.
- **Dialogue layer (schema v1.1).** New node type `dialogue`: worked transcripts walking each loop end-to-end in Kai's voice, every Kai turn tagged to the loop step / principle / anti-pattern correction it performs. Ch.1 ships 5 (D-01-01..05), ch.0 ships 2 (D-00-01/02). Built for Roma's operational review — but each turn is persona-conformant and graph-tagged, so they double as few-shot voice exemplars for your generation layer.
- **Schema v1.1 published** — `schema/v1.1.json` (draft 2020-12), canonical. Additive over v1.0 (Dialogue type only, no breaking renames). v1.0 is still bundled because `tools/validate-chapter.py` currently loads it; the validator does not yet parse `# dialogues` (upgrade queued before ch.2) — dialogues were validated against v1.1 separately.
- **Adversarial rigor on ch.0:** 18 flags across 4 independent review passes (rebuild 7+2, extension 7+2), 17 resolved-fix + 1 resolved-deliberate, 0 unresolved — full audit trail in `protocols/adversarial-flags-ch.00.md`. Validator: 145/145 checks, 0 failures; paragraph coverage machine-verified (131 source paragraphs, zero unaccounted).

## Current graph totals (ch.0 + ch.1)

| | states | loops | principles | anti-patterns | scripts | dialogues |
|---|---|---|---|---|---|---|
| ch.0 (pp.9–28) | 3 | 2 | 8 | 4 | 0 | 2 |
| ch.1 (pp.29–32) | 4* | 5 | 4 | 5 | 1 | 5 |

\* ch.1 re-declares the global carry-through `S-skill-committed` with its own triggers; `S-luck-seeker` / `S-skill-committed` are global ids owned by ch.0.

KG: 10 predicates, ~70 chapter triples live (incl. cross-chapter `L-00-02 precedes L-01-01`, `P-01-01 depends_on P-00-01/02`, `L-01-03 backed_by P-00-02`). MemPalace `kai_coach_kb`: one drawer per node (35 + room README), bodies mirror the markdown.

## Read in this order

1. **`guides/terry-handoff.md`** — narrative walkthrough (updated: dialogue layer, ch.0's new role, schema-staircase resolved).
2. **`spec.md`** — frame, contract terms, ownership boundaries.
3. **`schema/v1.1.json`** — canonical schema. Consume directly.
4. **`chapters/00-intro.md`** — the new onboarding layer. Read the two loops and their dialogues first if you're short on time.
5. **`chapters/01-finding-prospects.md`** — unchanged from v2.
6. **`protocols/edge-cheatsheet.md`** + **`protocols/extraction.md`** — data model + extraction rules.
7. **`templates/_chapter.md`** — field schema per node type incl. dialogues.
8. **`voice/kai-voice-persona.md`** — voice register. Normative (note the §Newcomer register shift — it governs dialogue turns tagged `newcomer`).
9. **`tools/validate-chapter.py`** — conformance checker (v1.0-pinned; see caveat above).
10. **`canvases.md`** — FigJam URLs: ch.0 main canvas (v4) + NEW ch.0 dialogues board + ch.1 boards. Figma view permission via Dmitry.
11. **`protocols/chapter-workflow.md`**, **`protocols/adversarial-reviewer.md`**, **`protocols/adversarial-flags-ch.0X.md`**, **`lessons/*.md`** — how the sausage is made + audit trails.
12. **`protocols/kg-predicates.md`** + **`protocols/mempalace-convention.md`** — only if you want live MCP access; ignore for the PostgreSQL-runtime path.

## What this bundle does NOT contain

- The source book (paraphrased, not redistributed).
- MemPalace drawer bodies (duplicate of the chapter markdowns).
- Live MCP access.

## Standing decisions (carried from the 2026-06-10 sync — all still hold)

| # | Topic | Decision |
|---|---|---|
| 1 | Scenarios | In each chapter's `# validation_scenarios` (ch.0 now has 9). |
| 2 | Step enrichment | `kai_intent` required on every loop step; `script_ref` optional. |
| 3 | Multi-language | Native-authored variants, batched after all 7 EN chapters. |
| 4 | Loop-as-resumable-subgraph | Confirmed; loops may span sessions. |
| 5 | MemPalace runtime role | Editorial only. Runtime stack = your call. |
| 6 | JSON Schema | We generate, you consume — now `v1.1.json`. |

## Open items to know about

- **State entry without a loop.** Unchanged: some states are entered via `entry_criteria` detection, not `transitions_to` (ch.1's `S-running-low-on-names` is the canonical case). Both paths needed in your state-detection layer.
- **Forward references.** `precedes` may target placeholder ids (`L-02-XX`); resolved retroactively when the target chapter lands. Note the previously-open ch.0 inflection forward-reference is now CLOSED (`L-00-02`).
- **Deferred theme:** the social-esteem-cost material (book pp.21–22, "people will think less of you") is deliberately gap-listed as `forward-reference` — it belongs to a later resilience/objection-handling chapter, not ch.0.
- **One deliberate non-emission:** the "goodbye to upline" speech template (p.28) stays a gap-list restatement of P-00-03, not a script node — no loop-step anchor in ch.0, and coaching that exact call needs Roma's sign-off first. If your runtime wants it, say so and we'll route it through review.
- **Voice discipline.** Knowledge from the book, voice from us. `source_ref` fields are internal-only — never surface them.

## What's coming

Ch.2 (Inviting) next on the same pipeline — heaviest on scripts (expect 5–10 script nodes with register variants). Then ch.3–7. Native-language variant batch after ch.7.

Questions / blockers / gaps — surface to Dmitry early; we amend protocols between chapters, not mid-extraction.

— 2026-06-11