decisions/spine/SPINE_V3_PLAN.md
Status: Josh raised this 2026-07-12 (during the endgame re-cascade). This doc is the honest re-evaluation of "are the body prose + transitions usable by any downstream pipeline?" and the ratified division of labor: what Opus handles BEFORE the Fable-5 handoff, and what Fable-5 handles. The resume cron references this doc.
Josh's model is correct. The story spine is the machine-consumable GROUNDING for the entire ~90–110M-word game — origin → 3 endings, the integrity paths, personal + faction relationships, memory of events; a spiderweb. Per chapter it must specify the chain of events (what happens), where the player is, the path(s) they navigate with the forks, the choices and their consequences, and the anchors (bosses, NPCs, familiars, SQ seeds, sites, assets). Downstream, the Quest-Writing (QW) pipeline expands this into the ~3M-word main questline (6–9M with the branches), and the side-quest / weaving / exploration / L2 / L3 pipelines generate the other ~80–110M words — ALL grounded in the spine.
Precise diagnosis (verified by sampling CH_11 early / CH_39 modern / CH_74 realm-road against docs/SPINE_SCHEMA_v2.md and T1_Build_Pipeline_Contracts):
1. THE ANCHORS ARE STRONG and genuinely pipeline-consumable — the threads, civilizational marks, key-sites, boss structure, side-quest stages, ally NPCs, dialogue anchors, asset anchors, mechanical FKs, integrity-divergence. Dense, [SRC:]-cited, FK-clean (the fidelity/anchors gates prove it: 11,193 anchors, 0 unresolved). Josh's read — "Opus did great work in the anchors and what is happening in each chapter" — is TRUE. The v3 upgrade BUILDS ON these; it does not discard them.
2. THE BODY PROSE (Block 4) IS ABSTRACT BY DESIGN and is NOT a chain-of-events spec any pipeline can decompose. Per the v2.5 schema Block 4 is deliberately a 200–400w literary "personal-arc lens" (identity-hiding, attachment/home/loss, path-divergence framing) — explicitly "NOT where value is added by growth." It feeds the cinematic / voice / theme / tone pipelines (a real need). But Josh's core question — "how is any pipeline supposed to USE the body prose as written?" — is CORRECT: Block 4 alone cannot drive QW. It is mood, not a spec.
3. THE CHAIN-OF-EVENTS / NAVIGABLE-PATH / FORK-GRAPH that the game needs is nominally Block 10 (Main quest beats) + Block 11 (Side-questlines) — but it is authored TOO THIN and too prose-like: ~5 high-level bullets per chapter; forks named mostly at the resolution beat (not as split-points throughout); the integrity paths inconsistently enumerated (sometimes a two-band gesture, sometimes the full four); NO systematic per-beat STATE (location / companions / kit / integrity band), NO per-beat CONSEQUENCE / RELATIONSHIP / FACTION / MEMORY deltas, loose boss-phase integration. It is a beat SKELETON, not the traversable event-graph the pipelines need.
4. THE TRANSITIONS are structurally decent (Departure / Passage / Arrival / Continuity + through-line ledgers, 3D-build-targeted) but also prose-forward, not a machine-consumable seam spec.
NET: the spine under-delivers the explicit, structured, machine-consumable EVENT-GRAPH + FORK-GRAPH + STATE-DELTA + CONSEQUENCE/MEMORY/RELATIONSHIP layer that QW and the quest/3D pipelines need. This is the single most important gap because the spine grounds the entire game.
THE FIX IS NOT "make the literary body prose less abstract" (that fights Block 4's designed job). The fix is a STRUCTURAL UPGRADE (spine v3): keep Block 4 as the short emotional lens, and add a real, explicit event-and-fork graph (upgrade Blocks 10/11) + a machine-consumable transition seam spec.
Essentially: promote the spiderweb (origin → 3 endings, the integrity paths, relationships, memory) from prose into an explicit graph the pipelines can traverse — grounded in T1_Build_Pipeline_Contracts and the 5 named consumers (RP / QW / AG / 3D / XP), QW the load-bearing one.
1. FINISH THE AAA STRUCTURAL RE-CASCADE — the Ch 75-77 fold DONE (commit 045f567); then the modern-arc (Ch 39-65) WS3/WS4 CONTENT re-flesh at the ANCHOR level (the WS4 champion faces + their BOSS_PERSONAS rows [the Factor/Verderer/Blind-Spot + Corr construct-kit have NO personas yet — author first], Corr-as-armory, the materialized-argument net-reduce, the 59-63 five-win-shape restructure, the villain-type inversions Ch 41/53/56, the pishtaco 51-53 cluster, the care flags). Authored in the CURRENT v2.5 form (anchors are form-stable; the body prose there is explicitly PENDING v3). Plans already scoped to scratchpad/modern_scope_39_45.md .. 59_65.md. → the corpus becomes architecturally COMPLETE + consistent for a holistic review.
2. DERIVE THE SPINE-V3 STRUCTURE SPEC — rigorous, pipeline-grounded (deep-dive + adversarial against T1_Build_Pipeline_Contracts + the real QW/3D/asset/quest consumers): the event-graph + fork-graph + state-delta + consequence/memory/relationship layer + the transition seam spec (the section above, fully specified with field lists + examples). This is the design decision that grounds the whole game.
3. REBUILD 2-3 EXEMPLAR CHAPTERS to v3 (one early, one modern, one finale) — the gold-standard target, so Fable + Josh see the concrete form before committing 79. Prove the v3 form is decomposable by QW and preserves the anchors.
4. WRITE THE FABLE-5 HANDOFF framing all of the above + the routing.
JOSH RATIFIES the v3 spec (it is structural canon — it grounds the entire game; do not mass-rewrite 79 chapters before Josh signs off on the target form).
1. Adversarially REVIEW/VALIDATE the v3 spec + exemplars from the pipeline-consumer lens (does it actually serve QW/3D/asset-gen? is it complete? gaps?).
2. EXECUTE the 79-chapter restructure to v3: build every chapter's event/fork graph + state/consequence/memory/relationship layer (Blocks 10/11 → v3), tighten Block 4 to its lens role, and rewrite the transitions to the seam spec.
3. Route back to OPUS 4.8 (1M) for any canon-grounding / deep-research mid-rewrite; back to Fable for the full 79-ch pass. Rebuild docs/review/build_reel.py after any source change; reveal-discipline re-verify before commit-as-final.
The STRUCTURE must be right before 79 rewrites (getting it wrong wastes the most effort). Opus has the canon + pipeline grounding to derive it and owns the anchors; a concrete exemplar de-risks the whole engagement; Fable then validates + mass-produces to a ratified target (its strength). "Define the target precisely, then mass-produce." The anchors Josh praised are preserved and built on — v3 ADDS the event/fork/state graph and repurposes the body prose; nothing good is thrown away.
Order: (1) the v3 structure derivation + synthesis; (2) rebuild 2-3 exemplar chapters to v3; (3) THEN the modern-arc (Ch 39-65) anchor re-flesh; (4) final verification; (5) the Fable-5 handoff. Rationale: the v3 structure is the priority deliverable and the modern-arc body prose is pending v3 anyway, so completing the modern content first would not feed the exemplars; the modern exemplar can mini-reflesh its own single node to demonstrate v3 on real branching content. The structure-question ownership is settled: OPUS derives the spec (it is a data-model/architecture problem entangled with the canon web, not a prose problem), FABLE adversarially validates the spec EARLY (before exemplars) + owns the 79-chapter execution; optional 1-chapter Opus-vs-Fable bake-off if Josh wants a concrete head-to-head.