Skip to content

docs: Premium-A blueprint feature subplan (HU 1326-combo + realtime postflop)#163

Merged
amaster97 merged 1 commit into
mainfrom
feat/premium-a-blueprint-subplan
May 28, 2026
Merged

docs: Premium-A blueprint feature subplan (HU 1326-combo + realtime postflop)#163
amaster97 merged 1 commit into
mainfrom
feat/premium-a-blueprint-subplan

Conversation

@amaster97

Copy link
Copy Markdown
Owner

Summary

  • Adds docs/premium_a_blueprint_subplan.md — comprehensive 9-phase subplan for the Premium-A capability: 1326-combo HU preflop blueprint at 7-8 stack depths, shipped as a .dmg-bundled asset, plus realtime postflop subgame solving anchored on the blueprint
  • Covers goal + success criteria, per-phase ready-to-spawn agent prompts, dependency graph, UI/UX (with source-transparency badge), storage-format decision (sharded gzipped JSON), asset packaging paths (inline vs first-run download), 11-item risk register, and 5 open questions for the user
  • Phase 0 (PR docs(critical): preflop RvR degenerate Nash investigation #159 cs-bug fix) is a HARD prerequisite — without it, the 15-20 h overnight compute would bake the bug into every shard

Open questions (must resolve before Phase 1 dispatch)

  1. Ante configuration matrix — single config or multi-config?
  2. Final depth list — all 8 (20, 30, 40, 60, 80, 100, 150, 200 BB) or trim?
  3. Phase 7 diff-test tolerance — what's "close enough" vs Brown/Pio/GTO Wizard?
  4. Asset packaging path — defer to Phase 1.5 output size, or pre-commit?
  5. Release version — v1.9.0 (additive) or v2.0.0 (generational)?

Test plan

Auto-merge OK — docs-only.

Generated with Claude Code

…ostflop)

Adds docs/premium_a_blueprint_subplan.md — a comprehensive subplan for building
the Premium-A capability: 1326-combo HU preflop blueprint at 7-8 stack depths
shipped as a .dmg-bundled asset, plus realtime postflop subgame solving
anchored on the blueprint. Covers 9 phases (Phase 0 cs-bug prereq through
Phase 8 docs), success criteria, dependency graph, UI/UX with source-
transparency badge, storage-format decision (sharded gzipped JSON), asset
packaging paths, an 11-item risk register, and 5 open questions that need
user resolution before Phase 1 dispatch.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
@amaster97 amaster97 merged commit 799f55f into main May 28, 2026
5 of 8 checks passed
amaster97 added a commit that referenced this pull request May 28, 2026
Premium-A Phase 8 documentation pass for the preflop blueprint feature
(task #68). Adds:

- docs/blueprint_user_guide.md (~190 lines) — end-user explainer:
  what the blueprint is, what ships in the .dmg, when to use blueprint
  vs custom solve, stack-depth coverage + interpolation behavior, ante
  config selection, common questions (Nash multiplicity vs published
  charts, blueprint vs aggregator, custom-range slowness).
- docs/blueprint_developer_guide.md (~370 lines) — engineering reference:
  how to regenerate blueprints (CLI flag matrix), asset + manifest
  schema, 169-class vs 1326-combo paths, True Path B engine internals,
  how to add new depths or ante configs, blueprint -> postflop subgame
  wiring, validation pipeline.
- README.md — new "Preflop blueprint mode" section linking both docs.
- USAGE.md §5.9 — Python API example using BlueprintLoader (Phase 2)
  and interpolate_strategy (Phase 3), with the standard "what falls
  through to live solve" decision table.

All examples cite real shipped APIs (BlueprintLoader.from_dir / .lookup
/ .actions / .available_depths; interpolate_strategy from
blueprint_interp). Validation references point to
preflop_100bb_chart_validation_v2_2026-05-28.md for Nash-multiplicity
divergence framing. PRs cited: #163 (subplan), #167 (Phase 1 hybrid),
#171 (True Path B 169-class kernel), #173 (interpolation),
#174 (loader).

Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant