← all plans

Codex Completion Roadmap

Drafted 2026-05-20 after Cycle 4 ship + Layer A structural cleanup. The destination: every Fast Track book has a complete coverage shell of math-style units, every cross-domain chapter (phys/chem/bio/phil) meets the §5.1 contract, and every skipped_units entry is closed. This document quantifies the gap, sequences the cycles, and sets realistic expectations.


§1 Where we are (2026-05-20, verified)

Metric Value
Shipped units 376
Queued units 273
Audits done 58 / 58 books (books_unaudited: [])
Cycle pace (Opus 4.7, math T1 + bio/chem mix) ~6 units per cycle, ~40 min wall-clock
T1-only skipped backlog (from manifests/skipped_units.md) ~105 remaining (was 110, Cycle 4 closed 2: KR-theory 03.08.12, φ⁴ 08.10.03)
Structural collisions all 4 resolved (12.02.01 dedupe, 12.03.01 renumber, 11-thermo → 05-symplectic, 09-quantum-theory → 03.14)
deps.json prereq-registration debt resolved (24 IDs in pending)
CONCEPT_CATALOG missing entries resolved (6+ added 2026-05-20)

1.1 Queued by section

Section Queued Notes
00 precalc 16 Math lane (Tyler's session A)
01 foundations 13 Math
02 analysis 14 Math (includes ODE/PDE backbone)
03 modern geometry 61 Math — largest single bucket, T1-heavy
04 algebraic geometry 59 Math — large, T1-heavy (Hodge/GIT/toric/tropical)
05 symplectic 10 Math (incl. Souriau-Gibbs now here)
06 Riemann surfaces 16 Math
07 representation theory 34 Math (Lie groups + algebraic groups)
08 stat mech 14 Math/physics border
09 classical mech 8 Physics lane (session B)
11 stat mech physics 1 Physics lane
12 quantum 5 Physics lane
21 number theory (no dir yet) 22 Math T1 grad-research cluster — planned, not yet started
Math (00-08 + 21) 245 This lane (Opus session A)
Phys/Chem/Bio (09-19) 28 + ~30 unwritten chem/bio gaps Session B owns

1.2 Per-chapter shipped counts (rounded to nearest where verified)

Math (489 files total — close to "complete" in many chapters at the quantity level; quality and depth varies):

  • 03-modern-geometry: 122 (largest, deepest)
  • 06-riemann-surfaces: 65
  • 07-representation-theory: 64
  • 05-symplectic: 61
  • 04-algebraic-geometry: 51
  • 02-analysis: 43
  • 03-differential-geometry: 39
  • 00-precalc: 25
  • 01-foundations: 23

Cross-domain (the underbuilt half):

  • 08-stat-mech: 23
  • 17-mol-cell-bio: 18
  • 12-quantum: 14
  • 19-eco-evo-bio: 13
  • 09-classical-mech: 12
  • 15-orgchem: 12
  • 10-em-sr: 10
  • 14-genchem-pchem: 10
  • 11-stat-mech-physics: 9
  • 13-gr-cosmology: 9
  • 16-inorgchem: 8
  • 18-organismal-bio: 8
  • 20-philosophy: 2 (vestigial)

§2 Definition of "complete"

A coverage shell is "complete" for a book / chapter when:

  1. Every P1-level concept from that book's audit punch-list has a shipped unit at 27/27.
  2. Every chapter's prerequisite chain is internally consistent (no pending unresolved prereqs in deps.json).
  3. Every unit meets the §5.1 math-style contract for its lane (math: full Lean + multi-section Master; chem/bio/applied-phys: prose-first multi-section Master, Lean optional).
  4. The Pass-W cross-stitch density is ≥ 3 lateral connections per Master section (per CYCLE_4 §6.2 threshold).
  5. Continuity-report metrics: backward-reference density ≥ 8.0 / 1000w, lateral connections ≥ 3 per Master, anchor-phrase reuse ≥ 0.95.

"Complete corpus" = all 58 audited books in this state. Realistic estimate: ~600 more units beyond the current 376 (factoring queued 273 + ~325 from re-deepenings, audits' P2/P3/P4 punch-list items, and section-21 number-theory build-out).


§3 Phased plan

Phase 1 — Structural & meta (Cycle 5 cleanup) — DONE 2026-05-20

  • ✓ Flatten repo to project root
  • ✓ Cycle 4 ship + Pass-W
  • ✓ Layer A structural cleanup (12.02.01 dedupe, 12.03.01 renumber, 11-thermo move, 09-quantum-theory rename)
  • ✓ deps.json + catalog backfill
  • ✓ Parallel-lanes coordination doc

Phase 2 — Math T1 backlog drain (Cycles 5–15, ~10 cycles × 6 units = 60 units)

Focus: close the skipped_units.md T1-only backlog. These are units only an Opus-tier session can produce reliably.

Cycle 5 (next dispatch this session): Characteristic-classes cluster

  • 03.06.07 Chern-Simons / transgression form
  • 03.06.10 Stiefel-Whitney and Pontryagin numbers
  • 03.06.11 Hirzebruch signature theorem (Master rewrite)
  • 03.06.12 Unoriented bordism and Thom's theorem
  • 03.06.14 Steenrod squares and the Wu formula
  • 03.06.18 Chern character ring homomorphism

Cycle 6: Homotopy advanced cluster (Quillen / Bousfield-Kan)

  • 03.12.33 Kan-Quillen model structure on sSet
  • 03.12.35 Simplicial model category and function complex
  • 03.12.37 Homotopy colimit (Bousfield-Kan construction)
  • 03.12.38 Bousfield-Kan spectral sequence
  • 03.12.40 Postnikov tower of a Kan complex
  • 03.12.45 Arithmetic square and integral fracture

Cycle 7: K-theory backlog

  • 03.08.12 ✓ DONE Cycle 4
  • 03.08.13–20 remaining items

Cycle 8: Hodge / Voisin cluster

  • 04.09.05 d-bar-d-bar-lemma
  • 04.09.07 Hard Lefschetz
  • 04.09.08 Hodge-Riemann bilinear relations
  • 04.09.09 Lefschetz (1,1)-theorem
  • 04.09.10 Akizuki-Nakano vanishing
  • 04.09.11 Kodaira embedding theorem

Cycle 9: GIT / Moduli cluster (04.10.x) Cycle 10: Toric / tropical cluster (04.11.x + 04.12.x) Cycle 11: Surface / divisor advanced (04.05.10–12) Cycle 12: QFT-side backlog (08.10.01, .04–07, .09, .10) Cycle 13: Number-theory grad cluster (21.02.x–21.10.x) — opens new section Cycle 14: Spectra / stable homotopy (03.08.06, 03.13.x cluster) Cycle 15: Cleanup + Pass-W consolidation across Cycles 5–14

Phase 3 — Rep-theory + Lie depth (Cycles 16–21, ~6 cycles × 6 = 36 units)

Section 7 has 34 queued. Cycles dispatch in topological order from queue head: 07.01.09, 07.05.05 onward.

Phase 4 — Foundations / analysis tightening (Cycles 22–26, ~5 cycles × 6 = 30 units)

00–02 queued: 43 units. Foundations of Section 1 — measure theory, functional analysis advanced, group/ring/field deepening. Fills the back of the prerequisite cone.

Phase 5 — Quality sweep + continuity audits (Cycles 27–30)

  • Re-validate every Cycle-1 through Cycle-15 shipped unit for drift.
  • Pass-W consolidation: every Master section should hit ≥ 3 lateral connections after this sweep.
  • Continuity report: pull all metrics to passing.
  • Lean compile check: every lean_status: partial or full module must compile against current Mathlib.

Phase 6 — Chem/Bio/Phys parity (~30 cycles, parallel to math via session B)

Owned by the parallel session per CYCLE_5_PARALLEL_LANES.md. Independent stream. Targets:

  • Bio: 37 → ~150 units across §17–19; multi-section Master + 6000+ words throughout.
  • Chem: 30 → ~100 units across §14–16; same contract.
  • Physics 10-em-sr: Lean retrofit or accept prose-first; 10 → ~30 units.
  • Physics 12-quantum: 14 → ~40 units (most are already 8000+ words; depth check only).
  • Philosophy §20: 2 → ~20 units. New chapter build-out.

Phase 7 — Number theory full build (Cycles ~50–60)

Section 21 doesn't exist yet on disk. Build from scratch: 22 queued + ~30 more for completeness. T1-only — finite fields, Quadratic reciprocity, p-adics, Hensel, Hilbert symbol, modular forms, Hecke, Eichler-Shimura, Galois reps, BSD, Sato-Tate, Iwasawa, Langlands survey.

Phase 8 — Final polish + ship gate (Cycles ~70+)

  • Beginning-tier accessibility audit (some units may be too dense).
  • Image / diagram coverage check.
  • Bibliography canonicalization sweep.
  • Final Lean compile pass.
  • Continuity report green across all metrics.

§4 Wall-clock estimate

  • Per-cycle: ~40 min (6 units in parallel + re-validation + Pass-W).
  • Cycles needed: ~80 across math lane, ~30 across chem/bio/phys lane (parallel).
  • Single-session realistic: 2–4 cycles before context density forces stop.
  • Total wall-clock to completion: ~80 × 40 min = 53 hours of orchestrator time, spread across many sessions.

This session's plan: finish Cycle 5 math (characteristic classes), Pass-W, commit + push. Stop.


§5 Quality gates (don't compromise)

Per AGENTIC_EXECUTION_PLAN.md §8:

  • Every unit re-validated independently 27/27 before commit.
  • Re-dispatch on any agent failure; never accept partial.
  • Pass-W runs after each cycle. Cross-stitches must be reciprocal.
  • Continuity report runs at least every 3 cycles; drift triggers a polish cycle.
  • Lean compilations checked at least every 6 cycles (when lean_status: partial/full).

§6 Risks & mitigations

Risk Mitigation
Agent quality drift across cycles Re-validate every cycle; Pass-W catches prose drift; quarterly continuity sweep
Cross-lane git conflicts Both sessions pull+rebase before dispatch; shared files append-only; conflicts are usually benign and human-resolved
Lean infrastructure rot (Mathlib upgrades) Lean compile every 6 cycles; pin Mathlib version in lake-manifest.json
Sub-agent harness limits (context, parallel cap) Cap parallel dispatch at 6; orchestrator (top-level) handles dispatch, sub-agents don't recurse
Memory drift (this doc going stale) Update §1 metrics every 3 cycles; treat queued numbers as snapshots

§7 Next steps after this session

  1. Tyler boots session B with the prompt from CYCLE_5_PARALLEL_LANES.md §8 to start chem/bio/phys lane.
  2. Session A continues math from Cycle 6 (homotopy advanced) at next session start; reads orchestrator_status.md for current state.
  3. Both sessions commit/push at end of each cycle; sync via Forgejo origin/main.
  4. After Phase 2 (math T1 backlog drain) is done, Tyler can decide whether to push Phase 3 vs hand chem/bio/phys priority.

End of roadmap.