Campaign Codex

Status · as of v0.7 alpha

What works, what doesn't, what's next.

The canonical, closure-updated statement of real-vs-stubbed state lives in docs/architecture/status.md in the repo — and that file supersedes anything written here. This page is a stable public summary suitable for the website; for the per-bundle receipts, read the release notes linked below.

Posture

Codex today is a developer proof harness and a buildable desktop workbench, not a finished character-management product. The corpus-ingest pipeline, the deterministic compute chassis, the boundary contract, and every persistence store are real, tested, and exercised end to end.

Character coverage is narrow: single-class Fighter at levels 1-3, for any race, is the only path that reaches a fully Computed receipt today. Every other class/level combination returns real claim-blocking diagnostics from the engine rather than fabricated output. That honesty is a load-bearing commitment of the project — we say so up front.

Real today

Corpus ingest

  • PCGen .pcc/.lst parsing into canonical SourcePackageContent
  • Repo-resident JSON corpus cache for seven books (CRB, APG, ACG, Bestiary 1, ARG, Pathfinder Unchained, Bonus Bestiary)
  • Per-domain engines: spellbook (9/9 schools), skill allocation, feat prerequisites (4/4), equipment effects (4/4), damage total, level-up (11/11 classes)

Compute + boundary

  • Pilot compute + boundary contract (compute_pilot_base_chassis → to_pilot_receipt) is deterministic and fail-honest throughout
  • Fighter + Wizard multiclass base-chassis dispatch (SD-24): BAB/save stacking + per-class named-feature explanations, levels 1-10
  • Encounter difficulty and party challenge rating compute

Desktop workbench (Tauri)

  • Character Hub: create, load, clone, portrait upload/load/delete, JSON export, recompute — real engine compute + real persistence
  • Campaign manager (local) with on-disk persistence and nonce-based conflict detection
  • Self-update chain + browser-handoff feedback to GitHub issues
  • Rule-system adapter seam — typed object-safe trait dispatching to Pf1Adapter (real) or a governed StubAdapter (honest "not yet implemented")

Homebrew + oracle

  • Homebrew authoring workbench — Guard Stance proof package's validate/persist/preview round trip
  • PCGen runner scaffolding — scripts/pcgen-run-character.sh drives real headless PCGen Gradle batch-export; oracle_validation::pcgen_runner wraps it in one Rust call
  • Support-state matrix (34-row typed truth ledger, bridged to the desktop tester workbench)

Honest structural ceilings

The two structural ceilings this section names are surfaced by recent work and neither fixed by it:

  • Race chassis. The engine models exactly seven races (CRB hardcoded). Most race_trait units across the corpus cannot ground until a real race chassis lands.
  • Companion kind is unstarted. All 1,696 companion units are not-ingested/not-started; 0 are grounded. The companion lane never began — its pilot cycle refused at the preflight-disk gate and was left unclaimed rather than falsely marked attempted.

Out of 38,540 units across 38 book directories, the corpus-wide status (as of SD-29 closure, 2026-08-11) is: grounded 491, text-complete 2,402, ingested-magnitude 6,548, not-ingested 14,582, not-started 11,190, unknown 3,291, deferred-with-reason 36. The full per-kind breakdown lives in status.md.

In flight

  • SD-30 — Class Feature & Archetype Bundle. 16 books' per-record rollup across archetypes and class features; targeted for the next release cut.
  • SD-31 — PCGen Character Import. Building the path for a user to open an existing PCGen character file in Codex. Rule: an import either brings the character across intact, or tells the user exactly what it could not bring and refuses — no silent losses.

Follow the in-flight bundles via the releases index or the docs/release/ tree.