Babel Bible

← all plans

Math lane — pickup plan

Written 2026-05-20 at session pause after wave-batch 1. Tip on Forgejo: 4d1d4ea.

Pick this up cold by reading the top of manifests/orchestrator_status.md and this file. Everything else is in the codebase.

§1 Where the math lane stands

Metric Value
Shipped units 714
Queued (audit-derived) 52
Audited Fast Track books 58 / 92 (34 explicitly out-of-scope)
Books at 100% in plan 36 / 50
Books partial 13
Books at 0% shipped 1 (Joe Harris Moduli of Curves)

Definition of done for this plan: every audited Fast Track book at 100% in manifests/production/plan.json. ~10 cycles @ 6 units, or ~15 wave-batches @ 4 units.

§2 Remaining queue by book

Ordered by "cheapest to close" (the book-completion framing introduced Cycle 12). Closing a book reduces the partial-count and moves it into the 100% column — that's the headline metric.

# Book Queued Cycles to close Closes at Notes
1 Manin IMNT 4 1 book #37 4 remaining: 21.06.02 Sato-Tate, 21.09.01 Arakelov, 21.09.02 Faltings/Mordell, 21.10.01 Langlands survey
2 Souriau Dynamical Systems 1 <1 book #38 spillover into another cycle
3 May-Ponto More Concise 1 <1 book #39 spillover
4 LL Stat Phys 1 4 1 book #40 thermodynamics/transport
5 Hirzebruch Manifolds & Modular Forms 2 <1 book #41 03.06.24 Bott-Taubes rigidity + 1
6 Hirzebruch Topological Methods 2 <1 book #42 04.05.11 worked HRR, 04.05.12 GRR pointer
7 Woit Quantum Theory Groups & Reps 2 <1 book #43 spillover
8 Fulton Toric Varieties 4 1 book #44 04.11.12 toric cohomology + 3
9 Weinberg QFT vol 1 4 1 book #45 QFT foundations cluster
10 Mumford GIT 5 1 book #46 deformation/DM moduli remainder
11 Chatterjee QFT 6 1.5 book #47 08.10.04 Wick + cluster
12 Parisi-Wu Stochastic Quantization 7 2 book #48 new chapter likely needed
13 Joe Harris Moduli of Curves 7 2 book #49 last zero-ship book
14 Serre Course in Arithmetic 8 2 book #50 Section 21 expansion (mostly NT)

Total: 57 queued, ~14 cycles or ~15 wave-batches.

The book-completion ordering above is the recommended dispatch order. Two reasons: (a) every cycle ends with a clean "book X closes at 100%" headline; (b) the ≤2-queue books at slots 2/3/5/6/7 fill spillover slots when a 4-queue book leaves room in a wave-batch.

§3 How to dispatch (pickup mechanics)

Quick path — autonomous wave-loop (recommended for sustained sessions)

Run the wave-loop skill. It refreshes plan, picks 4 next queued units, dispatches in parallel, integrates each, commits, pushes, loops. Hard-stops are well-defined.

# In Claude Code session:
/wave-loop

Skill driver loops until natural stop: 3+ stalls, 3+ consecutive validator-fix batches, continuity-metric regression, or out-of-scope topic.

Manual cycle path (when you want to curate dispatch)

cd "/Users/tyler/Documents/Code Projects/B.I.B.L.E"

# 1. Refresh plan
python3 scripts/build_production_plan.py

# 2. See the queue head
python3 scripts/produce_next.py --batch 6 | grep "^# Agent brief"

# 3. Read the full briefs (one per unit, ~80 lines each)
python3 scripts/produce_next.py --batch 6 > /tmp/briefs.txt

# 4. Dispatch 6 Agent calls in parallel using each brief
# 5. On return: re-validate each, run scripts/integrate_unit.py <id>
# 6. Pass-W cross-stitch (separate agent)
# 7. Commit + push
# 8. Append orchestrator_status entry

This was the pattern for Cycles 12-14. Use it when you want to enforce a specific book-by-book sequence or when an unusual topic needs custom briefing.

§4 Hard-won knowledge from this session arc (2026-05-20)

4.1 Term-of-art tension (NT lane)

The validator's \btrivial(?:ly)?\b regex prohibition collides with pinned analytic-number-theory phrases:

  • "trivial zero" → use "negative-even zero"
  • "trivial character" → use "principal character"
  • "trivial Galois representation" → use "identity / one-dimensional unit representation"
  • "trivial motive" → use "unit motive"
  • "inertia acts trivially" → use "inertia acts by the identity"

Affects every NT/Galois-rep unit. Future fix: add a per-domain validator exemption (whitelist) for chapter 21. Until then, agents work around with the substitutions above — established precedent in 21.03.01–21.07.02.

4.2 Catalog backfill pattern

If a unit ships with concept_catalog_id: namespace.foo but no matching entry exists in docs/catalogs/CONCEPT_CATALOG.md, validator fails on that one check. Backfill the missing entry as part of close-cycle (same shape as siblings). Two instances in this session: foundations.subgroup-coset-quotient (Cycle 13), diff-geom.padics-lie-group (Cycle 14).

4.3 Continuity-metric advisories ≠ regressions

scripts/integrate_unit.py may report forward_density: 2.21 < 3.0 or synthesis_master: 4.00 < 4 as warnings — these are not validator failures. Integrator still marks the unit shipped. Only stop on regression of a previously passing unit, not on advisories for newly-shipped ones.

4.4 Session B carry-over

The other Claude session (Session B, chem/bio/phys lane) keeps shipping cross-domain deepenings between math dispatches. Typical pattern: at commit time, 4–11 cross-domain files appear modified in working tree. They validate at 27/27 and should be committed separately (chem/bio/phys carry-over: prefix) before the math-lane commit. See 9d598f9, bc52204 for shape. Keeps Forgejo history clean and auditable.

4.5 Sibling-in-flight prereqs

When a unit's natural prereq is being produced in the same cycle, the agent must put it in ## Connections [Master] not prerequisites: — otherwise validator fails the prereq-resolve check. Codified in playbook §10.3. Briefs should call this out explicitly when relevant.

4.6 Pass-W reciprocity

After each cycle ships, run a Pass-W agent (dispatch + brief in this session's cycles) to add reciprocal Connections bullets in peer files. Typical: 20–40 reciprocal edits across 10–16 peers. Catches stale unit-ID references and reciprocity gaps. Pattern in manifests/weaving_report.md — append each cycle.

4.7 Tip-of-Forgejo discipline

Push after every cycle / wave-batch close. Forgejo remote is http://100.108.123.49:49152/Tyler/codex.git, branch main. Session B and math-lane sessions sync through here — push promptly to avoid divergence.

§5 Quality gates (don't compromise)

Per docs/plans/AGENTIC_EXECUTION_PLAN.md §8 and the cycles 12–14 precedent:

  1. 27/27 validator on every unit before commit. Re-validate independently after agent returns; agents have been known to over-claim.
  2. Re-dispatch on agent failure. Never accept partial — one transient API socket failure happened in Cycle 11; the re-dispatch was clean.
  3. Pass-W reciprocity after every cycle. Catches one-way links before they accumulate.
  4. Lean compile check every 6 cycles. lake build against current Mathlib. Pin lake-manifest.json if a Mathlib upgrade breaks anything.
  5. Continuity report sweep every 3 cycles. Drift triggers a polish cycle.

§6 What comes after "all books closed"

Phase 5 polish (3–4 cycles):

  • Continuity sweep across all 714+ shipped units
  • Lean compile pass against current Mathlib
  • Drift audit on Cycle-1 through Cycle-N units
  • Bibliography canonicalisation
  • Pass-W consolidation: every Master ≥ 3 lateral connections

After that, decide:

  • Backfill the ~200 missing entries in CONCEPT_CATALOG.md, OR
  • Reposition the catalog as a seed-list and update /concepts to reflect that

Site/UX polish (separate, ~2 hour focused session):

  • Add tier + Lean columns to /units index
  • Add search + section grouping to /concepts
  • Link prereqs on /concepts to in-page anchors
  • Fix "Babel Bible" → "Math STEM" branding in head titles
  • Add "Catalog entry →" reverse link on /u/[id]

These don't block math-lane completion and can run in parallel with Session B's chem/bio/phys lane.

§7 Where to look in the codebase (orientation)

Need Location
Current state manifests/orchestrator_status.md (tail)
Plan / coverage manifests/production/plan.json (regenerate via scripts/build_production_plan.py)
Next unit briefs python3 scripts/produce_next.py --batch N
Validator python3 scripts/validate_unit.py <path>
Integrator python3 scripts/integrate_unit.py <id>
Audit punch-lists plans/fasttrack/<book-slug>.md
Style contract docs/specs/UNIT_SPEC.md, docs/specs/QUALITY_RUBRIC.md, docs/plans/CYCLE_4_STYLE_PARITY_PLAN.md
Production playbook docs/plans/AGENT_PRODUCTION_PLAYBOOK.md (§5.1 contract, §10 originator-prose, §10.3 sibling-in-flight, §16 paths)
Cross-stitch log manifests/weaving_report.md
Cycle history manifests/orchestrator_status.md

§8 First action when picking this up

cd "/Users/tyler/Documents/Code Projects/B.I.B.L.E"
git pull origin main                                    # sync with Session B
python3 scripts/build_production_plan.py                # refresh
tail -100 manifests/orchestrator_status.md              # recent cycles
python3 -c "import json; p = json.load(open('manifests/production/plan.json')); print(f'shipped={p[\"counts\"][\"shipped\"]} queued={p[\"counts\"][\"queued\"]}')"

Then either invoke /wave-loop or pick the next book from §2 and dispatch manually.

End of pickup plan. Good luck — quality bar is holding; the orchestration loop works.