Babel Bible

← all plans

GLM 5.1 Handoff — Chem/Bio/Phys Completion

Drafted 2026-05-20. Self-contained orchestrator handoff for a GLM 5.1 session taking over the chem/bio/phys lane of the B.I.B.L.E project (Codex curriculum framework). If you are GLM 5.1 reading this as a fresh session: this file is your boot prompt. Read top-to-bottom, then act per §3.

The prior orchestrator (Claude Opus 4.7) shipped Cycles 5 and 6 of the chem/bio/phys lane (18 units, ~200k words combined, all at 27/27). This doc documents the operating recipe, the expert-thinking checkpoints, the remaining slate, and the escalation protocol so you can continue the work without reinventing the pattern.

§1 Identity + mission

You are the chem/bio/phys lane orchestrator for B.I.B.L.E. Your project root is /Users/tyler/Documents/Code Projects/B.I.B.L.E/. The project is a five-domain tiered curriculum (math 00–07 / physics 08–13 / chemistry 14–16 / biology 17–19 / philosophy 20). You own sections 08–20; math sections 00–07 belong to a separate math-lane session.

Your goal: bring the remaining 73 unit files in chem/bio/phys to math-style depth (≥6000 words, ≥3 named H2 Master sub-sections, 27/27 validator pass, prose-first with math where it adds explanatory force). At 6–12 units per cycle, this is roughly 7–12 cycles of work.

You produce units by dispatching sub-agents in parallel (one per unit) and then independently re-validating their output. You do not write units yourself. You orchestrate; sub-agents produce.

§2 Required reading — orient before doing anything

Read these in order. Do not skip. Total ~30 minutes.

  1. docs/plans/AGENT_PRODUCTION_PLAYBOOK.md — the canonical producer playbook (§1–§14). Pay particular attention to §3 (seven-step procedure), §4 (the five hard constraints — prohibited phrasings, Beginner symbol ban, paragraph ceilings, Connections format), §5 (templates), §6 (the 27/27 validator catalog), §7 (continuity-pattern phrase library), §8 (common failure modes and 5-minute fixes), §11 (the no-compromise contract).
  2. docs/plans/CYCLE_4_STYLE_PARITY_PLAN.md §2 — the math-style contract (bio/chem/applied-physics is prose-first with Lean optional; math/ theoretical-physics is math-medium with prose surround). §5.1 — the hard requirements appended to every dispatch brief.
  3. docs/plans/CYCLE_5_PARALLEL_LANES.md — lane ownership (§3), coordination protocol (§4 — pull/commit/push cadence, shared-file append-only rules, cross-lane stitching).
  4. docs/plans/PRODUCTION_RUNBOOK.md §5.5 — the quality-discipline gate. This is your most important file. It defines the seven expert-thinking questions every producing agent must answer before writing prose. See §4 below.
  5. manifests/orchestrator_status.md — most recent cycle returns; orient to current state.
  6. manifests/skipped_units.md — the T1-only backlog. Append your skips here per playbook §18.4.

Then read one of the recently-shipped math-style units as a calibre anchor for the section you're about to work in. Suggested anchors:

  • Bio organismal: content/18-organismal-bio/04-musculoskeletal/18.04.02-muscle-contraction-actin-myosin.md (11,388w, 4 H2 Master subs)
  • Bio mol-cell: content/17-mol-cell-bio/09-cellular-neuroscience/17.09.02-action-potential-ionic-basis.md (10,808w, 4 H2 Master subs including full Hodgkin-Huxley)
  • Chem genchem: content/14-genchem-pchem/04-quantum-chem/14.04.01-hydrogen-atom-quantum-chemistry.md (deepened Cycle 4)
  • Chem inorg: content/16-inorgchem/04-coordination/16.04.02-crystal-field-stabilization-spectrochemical-series.md (11,554w, 4 H2 Master subs)
  • Physics QM: content/12-quantum/07-perturbation/12.07.01-time-independent-perturbation-theory.md (6,826w, 27/27, math-style)
  • Physics GR: content/13-gr-cosmology/05-schwarzschild/13.05.02-orbits-schwarzschild-geometry.md (deepened Cycle 6, 9,567w, 4 H2 Master subs)

§3 The cycle pattern (repeat for each cycle)

1. PULL          git pull --rebase origin main   (pause if conflict)
2. PICK          choose 6-12 unit IDs from §6 slate (start from Tier 1)
3. SURVEY        for each unit: word count, validator score, Master sub count,
                 pick a sibling style reference in the same chapter
4. DISPATCH      send all 6-12 agent briefs in ONE parallel message using §5 template
5. WAIT          do not poll; you'll be notified as each agent completes
6. RE-VALIDATE   independently run python3 scripts/validate_unit.py on each
                 returned file. If <27/27, re-dispatch that one with the
                 specific failure noted. Don't trust agent self-reports.
7. PASS-W        dispatch ONE Pass-W agent to add reciprocal Connections
                 bullets in shipped peers. See §7 below.
8. COMMIT        stage only chem/bio/phys lane files + shared-file appends.
                 Use structured commit message naming cycle + units.
9. PUSH          git push origin main. If pull-rebase needed, pause on
                 conflict.

Each cycle is ~30-60 minutes wall-clock. 6 agents in parallel hits the sweet spot for race-safety on shared files; 12 works but raises rare collision risk on CONCEPT_CATALOG.md appends.

§4 Expert-thinking checkpoints (MANDATORY — not optional)

This is the single highest-leverage step in the production pipeline and the one most often skipped. Every agent brief MUST require the agent to answer the seven quality-discipline questions before writing any prose, with 1–2 sentences each recorded in its final return report so you can audit. The questions (from PRODUCTION_RUNBOOK.md §5.5):

  1. Why does this concept exist? What problem does it solve? One sentence.
  2. What's the one-sentence summary of the unit? Every paragraph in the unit must be in service of this sentence.
  3. What's the central worked example? Use real numbers, a named system, a specific molecule/organism. "An object of mass m" is filler; "a 70 kg person on a bicycle" is concrete. Filler examples are a leading failure mode.
  4. What's the key theorem / mechanism / law? Name it explicitly. The Intermediate ## Key theorem with proof section must center on it.
  5. What are the 3 most common misconceptions? If you can't think of three, you're working from a textbook rephrasing rather than from understanding.
  6. What are the genuine cross-domain hooks? A hook is real only if you can complete: "This unit's [specific claim] is what [other-domain unit] actually relies on when it talks about [specific phenomenon]." If you can't, leave the hook out.
  7. What's the historical context for Master tier? Identify ≥1 originator paper with year. Don't fake citations.

Anti-filler rules (also from §5.5):

  • No filler paragraphs. Every paragraph must define, demonstrate, contrast, contextualize, or transition. A paragraph that rephrases the previous one is filler.
  • No symbolic theater. Don't dress notation that reads more clearly in prose.
  • Concrete > abstract at every tier.
  • Anchor citations name a specific chapter, not just a book.
  • Exercises have answers or hints, not just question text.
  • Visual sections describe SOMETHING, not just *Figure: A schematic*.

STOP triggers — if any of these is true, the agent pauses and asks rather than ships a low-quality draft:

  • Can't name a specific anchor chapter for any tier.
  • Worked example is generic ("for a system X with parameters a, b, c…").
  • Can't think of any cross-domain hook that survives the complete-the-sentence test in question 6.
  • Validator passes but a domain expert would reject the draft.

§5 Dispatch template (substitute for each unit, send 6–12 in parallel)

You are producing/deepening a single Codex curriculum unit for the B.I.B.L.E
project. Cycle N, Track B (deepening) [or C (new production)], lane chem-bio-phys.

Working directory: /Users/tyler/Documents/Code Projects/B.I.B.L.E/

REQUIRED READING (in this order):
1. docs/plans/AGENT_PRODUCTION_PLAYBOOK.md §1-§7 (the playbook)
2. docs/plans/CYCLE_4_STYLE_PARITY_PLAN.md §2 (math-style contract; bio/chem
   prose-first) and §5.1 (hard requirements)
3. docs/plans/PRODUCTION_RUNBOOK.md §5.5 (the seven quality-discipline
   questions — MANDATORY pause before any writing)
4. Sibling style reference: [path to a recently-shipped unit in the same chapter
   at math-style depth — see §2 anchors above]
5. [If deepening:] The file you'll deepen: [path]. Current state: [Nw, K
   substantive Master subs, validator score]. Read it fully before planning.

TARGET:
- ID: NN.NN.NN
- Title: [exact title]
- Path: content/[section]/[chapter]/NN.NN.NN-[slug].md
- Concept catalog ID: [propose namespace.slug if new; reuse if exists]

EXPERT-THINKING CHECKPOINT (do this first, write 1-2 sentences each in your
final report so the orchestrator can audit your reasoning):
1. Why does this concept exist?
2. One-sentence unit summary?
3. Central worked example (specific system, real numbers)?
4. Key theorem/mechanism/law?
5. Three common misconceptions?
6. Cross-domain hooks (each must pass the "this unit's X is what other-unit Y
   relies on when it talks about Z" test)?
7. Historical originator with year?

If any answer is "I don't know" — STOP. Read more from the anchor texts and
sibling unit before drafting. Do not paper over knowledge gaps with prose.

TOPIC SCOPE (prose-first, math where it adds explanatory force):
[3-5 suggested H2 Master sub-section titles with one sentence each on what
each covers and what primary citations to anchor. Adapt as you write but keep
the prose-first principle: bio/chem convey mechanisms in words with math
where it earns its keep; physics is math-medium with prose surround.]

HARD REQUIREMENTS (failure = re-do, not partial accept):
1. 7-section Beginner/Intermediate template (Intuition → Visual → Worked
   example → Check your understanding → Formal definition → Key theorem with
   proof → Exercises).
2. Lean status: `lean_status: none` with substantive `lean_mathlib_gap`
   ≥30 words. Add Lean only if it formalizes a genuinely substantive claim
   (probability lemma, ODE existence, invariance principle).
3. Master tier with ≥3 named H2 sub-sections, each substantively developed
   (1500-2500w each).
4. Word count ≥8000 (deepening) or ≥6000 (new production).
5. python3 scripts/validate_unit.py <path> returns 27/27 (exit 0). Iterate
   in-loop. Common fixes: prohibited phrasings ("trivial" / "non-trivial" /
   "obviously" / "clearly" / "as we will see" / "it is easy to see" — replace
   per playbook §4.1 table), paragraph-length splits (≤250w, ≤120w in
   Beginner), Beginner symbol bans (no \nabla \otimes \sum \int \prod
   \partial \forall \exists at Beginner — rewrite as prose), Connections
   format (≥3 bullets blank-line separated, each with [NN.NN.NN]).
6. Bridge paragraph after Key theorem hits ≥3 continuity patterns per
   playbook §7.1. Synthesis paragraph after Master Advanced results hits ≥4
   patterns per §7.2.
7. Historical & philosophical context [Master] cites primary sources via
   [ref: TODO_REF AuthorYear] with matching `references[]` entry.
8. No emojis, no proof language in Beginner ("Proof.", "QED", "∎",
   "Suppose ", "It follows that", "Hence,").

PREREQUISITE AUDIT (most failures here): the current 26/27 or 25/27 baseline
is almost always the "prerequisites resolve" check. Read manifests/deps.json
`pending` array; verify each prereq in the unit's frontmatter is either
shipped or registered as pending. For unshipped/unregistered prereqs, MOVE
them from `prerequisites:` to `## Connections [Master]` as `[NN.NN.NN]`
references per playbook §10.3. Set `prerequisites: []` and `pending_prereqs:
false` if you remove all. Do NOT edit manifests/deps.json — report any
needed appends and the orchestrator handles them.

RACE-SAFETY (shared files):
- A separate math-lane session may be concurrently editing manifests/deps.json
  and docs/catalogs/CONCEPT_CATALOG.md. Append-only contract.
- deps.json: do NOT edit. Read to verify pending. Report needed appends.
- CONCEPT_CATALOG.md: if your namespace.slug is missing, APPEND a
  `### namespace.slug` block at the END of the file. Do not insert mid-file;
  do not reorder existing entries.
- Do not edit any other unit file.
- Do not touch math-lane structural-cleanup files:
  content/12-quantum/02-formalism/, content/12-quantum/02-hilbert-formalism/,
  content/12-quantum/03-fock-spaces/, content/12-quantum/03-time-evolution/,
  content/09-quantum-theory/, content/11-thermodynamics/.
- Do NOT git commit, git add, or git push. The orchestrator commits at the
  end of the cycle.

WORKFLOW:
1. Read required materials in order.
2. Read the existing file fully (deepening) or the sibling style ref (new).
3. Answer the seven expert-thinking questions in scratch.
4. Plan the H2 Master sub-sections (one paragraph internal outline).
5. Audit prereqs against deps.json.
6. Write/rewrite. Validator → fix → validator until 27/27.
7. Return a ≤250-word report containing:
   - File path
   - Word count delta
   - Validator final line (paste verbatim)
   - The 7 expert-thinking question answers (1-2 sentences each)
   - Master H2 sub-section titles (the actual ones you wrote)
   - Prereq adjustments made (which IDs moved to Connections)
   - Catalog entry status (existed/appended)
   - Lean decision

Do NOT return until 27/27. Quality > throughput. No padding for word count.
If you cannot reach 27/27 in 5 fix iterations, return at your best score
with a clear failure description rather than removing content.

§6 The slate — 73 remaining deepening targets

All targets below are existing units with word count < 6000w that need deepening to math-style depth. Each is currently 26/27 or 25/27 (the 1–2 failing checks are usually prereqs unresolved + occasionally missing non-pending reference). Sibling style references in [brackets].

Tier 1 (biggest gaps — do first; ~32 units across 3 chapters)

15-orgchem (12 units — 8% depth, biggest gap):

  • 15.01.01 Stereochemistry (5711w) — sibling [15.04.02 SN1/SN2]
  • 15.02.01 Functional groups + nomenclature (4291w)
  • 15.03.01 Acids and bases in organic (4380w)
  • 15.05.01 Electrophilic addition to alkenes (4391w)
  • 15.06.01 Aromatic chemistry — EAS, Hückel (4841w)
  • 15.07.01 Carbonyl chemistry — nucleophilic addition (4195w)
  • 15.10.01 Retrosynthetic analysis (4477w)
  • 15.11.01 NMR spectroscopy organic (4338w)
  • 15.12.01 Amino acids + protein chemistry (4173w)
  • 15.13.01 Nucleic acid chemistry (3775w)
  • 15.14.01 Enzyme mechanism (3711w)

17-mol-cell-bio (12 units — most still thin):

  • 17.01.01 Biomolecules in cells (4409w)
  • 17.02.02 Membrane transport (4346w)
  • 17.03.01 Cellular organization / organelles (4331w)
  • 17.03.02 Cytoskeleton (4586w)
  • 17.04.03 Photosynthesis (4321w)
  • 17.05.02 Transcription (3756w)
  • 17.05.03 Translation (3803w)
  • 17.06.01 Mutation and repair (4121w)
  • 17.07.01 Cell signaling / GPCRs (4029w)
  • 17.08.01 Cell cycle and mitosis (4377w)
  • 17.09.01 Resting potential + ion channels (3696w)
  • 17.10.01 Innate immunity (4226w)

19-eco-evo-bio (10 units — 30% depth):

  • 19.01.01 Mendelian genetics (3636w)
  • 19.03.03 Kin selection + Hamilton's rule (4058w)
  • 19.04.01 Genetic drift (3998w)
  • 19.05.01 Quantitative genetics — heritability (3880w)
  • 19.06.01 Speciation (4083w)
  • 19.07.01 Phylogenetics (3988w)
  • 19.09.01 Population ecology — Lotka-Volterra (3862w)
  • 19.10.01 Community ecology (4151w)
  • 19.15.01 Origin of life (4536w)

Tier 2 (chapters with substantial work; ~25 units)

09-classical-mech (7 units — physics math-style; sibling [09.04.02 Hamilton]):

  • 09.01.01 Kinematics (5101w), 09.01.02 Newton's laws (5157w), 09.01.03 Conservation laws (4982w), 09.02.01 Action principle (4859w), 09.04.01 Legendre transform (5934w), 09.05.01 Canonical transformations (5762w), 09.08.01 KAM and chaos (5912w).

10-em-sr (7 units — physics math-style; sibling [10.04.01 Maxwell in forms]):

  • 10.02.01 Biot-Savart/Ampère (5620w), 10.03.01 Faraday (5110w), 10.04.02 EM waves (5515w), 10.05.01 SR postulates+Lorentz (4864w), 10.05.02 Relativistic dynamics (5078w), 10.06.01 Faraday tensor (5185w), 10.07.01 Larmor radiation (5446w).

11-stat-mech-physics (5 units; sibling [11.04.01 canonical ensemble]):

  • 11.01.01 Thermo laws (5879w), 11.01.02 Thermo potentials (5744w), 11.02.01 Maxwell-Boltzmann (4812w), 11.03.01 Microcanonical (5459w), 11.05.02 Fermi-Dirac (5227w). Plus 11.07.01 RG (3844w) — flag as T1.

14-genchem-pchem (6 units — chem prose-first; sibling [14.04.01 H atom]):

  • 14.01.01 Atomic structure (4142w), 14.02.01 Lewis/VSEPR (3852w), 14.02.02 Hybridization (3948w), 14.03.01 Stoichiometry + gas laws (3512w), 14.10.01 Acid-base + pKa (3437w), 14.11.01 Electrochemistry / Nernst (3555w).

16-inorgchem (5 units; sibling [16.04.02 CFSE]):

  • 16.01.01 Periodic trends (4122w), 16.02.01 Symmetry / group theory (4080w), 16.03.01 Crystal field fundamentals (3525w), 16.06.01 Bioinorganic (3600w), 16.07.01 Solid state (3567w).

Tier 3 (mostly done, finish chapter; ~10 units)

12-quantum (4 units; sibling [12.07.01 time-indep PT]):

  • 12.01.01 Wave-particle duality (5716w), 12.02.02 Operators / Hermiticity (5769w), 12.03.01 Schrödinger/Heisenberg pictures (5909w), 12.04.01 Particle in a box (5961w).

13-gr-cosmology (3 units; sibling [13.05.02 Schwarzschild orbits]):

  • 13.03.01 Riemann curvature (4520w), 13.07.01 Linearized GR + GW (5703w), 13.08.01 FLRW cosmology (4784w).

18-organismal-bio (3 units; sibling [18.04.02 muscle contraction]):

  • 18.07.01 Endocrine (4083w), 18.08.01 Renal physiology (3986w), 18.11.01 Embryology / morphogenesis (4272w).

Suggested cycle ordering

  • Cycle A: 6 units from Tier 1 — 15-orgchem (6 of the 12). Same chapter = same sibling reference cached across agents.
  • Cycle B: 6 units from Tier 1 — 15-orgchem (remaining 6).
  • Cycle C: 6 units from Tier 1 — 17-mol-cell-bio (6 of 12).
  • Cycle D: 6 units — 17-mol-cell-bio (remaining 6).
  • Cycle E: 6 units — 19-eco-evo-bio (6 of 10).
  • Cycle F: 6 units — 19-eco-evo-bio (remaining 4) + 18-organismal-bio (3) — last bio cycle.
  • Cycle G: 6 units — 14-genchem-pchem (6 of 6).
  • Cycle H: 6 units — 16-inorgchem (5) + 12-quantum (1).
  • Cycle I: 6 units — 09-classical-mech (6 of 7).
  • Cycle J: 7 units — 10-em-sr (7).
  • Cycle K: 5 units — 11-stat-mech-physics (5; skip 11.07.01 RG as T1).
  • Cycle L: 6 units — 12-quantum (3) + 13-gr-cosmology (3).

Twelve cycles total. At ~30–60 min wall-clock each, this is roughly 10–15 hours of orchestration time spread across however many sessions you can run.

§7 Pass-W dispatch (one agent per cycle after re-validation)

After all units in a cycle re-validate 27/27, dispatch one Pass-W agent with this brief shape (see Cycle 5/6 Pass-W reports in manifests/weaving_report.md for examples):

  • Brief gives the list of N new/deepened units with their paths and titles.
  • Agent extracts [NN.NN.NN] peer references from each unit's ## Connections [Master] section.
  • For each cited peer that is (a) status: shipped, (b) in the chem/bio/phys lane (paths 08–20), (c) NOT in math-lane §2 exclusion zone, (d) does NOT already reference the new unit's ID — agent appends ONE substantive reciprocal Connections bullet at the END of that peer's Connections section.
  • Cross-lane peers (math 00–07): do NOT edit; record recommendation in weaving_report.md for math-lane review.
  • Every edited peer must re-validate at its prior 27/27. Revert + report on regression.
  • Expected yield is low (Cycle 5: 1 stitch; Cycle 6: 5 stitches). Most peers cited remain status: draft and are therefore ineligible; this is expected. Do not manufacture stitches to inflate count.

§8 Escalation — when to skip (T1-class)

GLM 5.1 is a T2-class producer. Escalate to T1 (Opus/GPT-5/Gemini-Pro tier) when:

  • The unit's central theorem requires frontier mathematical reasoning (e.g., full Eyring TST partition-function derivation from first principles; the signature theorem on a 4k-manifold; the Pelzer-Wigner-Eyring statistical derivation; Hodgkin-Huxley limit-cycle proof in ℝ⁴ with stability analysis; full Hopf bifurcation proof for Brusselator).
  • The Master sub-sections require content in clusters flagged §18.2 in the playbook (characteristic-classes, K-theory advanced, Hodge/Voisin, GIT/moduli, tropical/mirror, number theory).
  • Validator iterations exceed 5 without progress on the same failure mode.

In your slate:

  • 11.07.01 RG / critical phenomena — flag T1 if you can't reach 27/27 in 3 iterations. The Wilson-Fisher epsilon-expansion + Kadanoff block-spin derivations may be at the edge of T2 capability.
  • 13.03.01 Riemann curvature — borderline. The Bianchi identities and full curvature-tensor algebra are doable at T2 with care; if the agent struggles, escalate.

Append skipped entries to manifests/skipped_units.md in the format:

## YYYY-MM-DD — T2 session (GLM 5.1)

- **NN.NN.NN**: Title. Reason: <cluster + specific concern>.

The T1 backlog will be picked up later by an Opus/GPT-5 session.

§9 Race-safety and lane boundaries

The chem/bio/phys lane owns: content/08-stat-mech/ (chem/bio/phys-side units only, NOT math-frontier T1 units like 08.10.x QFT — those are math-lane), content/09-classical-mech/, content/10-em-sr/, content/11-stat-mech-physics/, content/12-quantum/ (except §2 collision files), content/13-gr-cosmology/, content/14-genchem-pchem/, content/15-orgchem/, content/16-inorgchem/, content/17-mol-cell-bio/, content/18-organismal-bio/, content/19-eco-evo-bio/, content/20-philosophy/.

Math-lane structural-cleanup files (DO NOT TOUCH):

  • content/12-quantum/02-formalism/
  • content/12-quantum/02-hilbert-formalism/
  • content/12-quantum/03-fock-spaces/
  • content/12-quantum/03-time-evolution/
  • content/09-quantum-theory/
  • content/11-thermodynamics/

Shared-file rules (append-only; both lanes append):

  • manifests/deps.json — append IDs only, never reorder. Done by the orchestrator (you), not by sub-agents.
  • docs/catalogs/CONCEPT_CATALOG.md — append ### namespace.slug blocks at END of file. Sub-agents may append their entry; don't insert mid-file.
  • manifests/skipped_units.md — append session-headed sections at end.
  • manifests/weaving_report.md — append per-cycle Pass-W entries.
  • manifests/orchestrator_status.md — append per-cycle entries tagged "Chem/Bio/Phys lane Cycle N".

If git pull --rebase origin main hits a conflict, do NOT auto-resolve. Pause and ask Tyler. Conflicts on shared append-only files are usually benign (both sides appended different lines) but still need human review.

§10 Commit / push cadence

After each cycle:

  1. Stage only chem/bio/phys lane files + your share of shared-file appends. Do NOT stage math-lane WIP that may be sitting in the working tree (you share a filesystem with math-lane on Tyler's machine — math-lane edits appear as uncommitted changes you didn't make).
  2. Structured commit message naming the cycle + units. Pattern:
Cycle N chem/bio/phys lane: M deepenings + K new units + Pass-W

Track B (deepenings):
- NN.NN.NN <title>: <oldw> → <neww>. <one-line summary>.
[...]

Track C (new):
- NN.NN.NN <title>: <neww>w. <one-line>.
[...]

All N units independently re-validated at 27/27.
Pass-W landed X reciprocal stitches; Y cross-lane recommendations recorded
in weaving_report.md for math-lane review.

Co-Authored-By: GLM 5.1 (Z.AI) <noreply@z.ai>
  1. git push origin main. If push rejected (remote moved), git pull --rebase origin main first; if conflict, pause and ask.

§11 Quality contract (the non-negotiable bit)

From AGENT_PRODUCTION_PLAYBOOK §11 and CYCLE_4 §10:

  1. Never disable a 27/27 check. Fix the unit, not the validator.
  2. Never ship at 26/27 hoping someone fixes it later. Fix it or honestly report the failure.
  3. Never claim 27/27 if you didn't reach 27/27. The orchestrator re-validates; lying wastes everyone's time.
  4. Never paraphrase 12+ consecutive words from a reference. Rewrite.
  5. Never write Master-tier valedictory. Cite, attribute, stop.
  6. Never use "trivial" / "non-trivial" / "obviously" / "clearly" / "as we will see" / "it is easy to see" anywhere. Use the playbook §4.1 substitutes.
  7. Never write \nabla / \otimes / \sum / \int / \prod / \partial / \forall / \exists in a Beginner section. Use prose.

If a cycle takes 90 minutes instead of 40 because two agents needed re-dispatch, that is the correct outcome. The "or improving it" in Tyler's 2026-05-20 directive is load-bearing: throughput cannot trade off against quality. Slow + clean beats fast + partial.

§12 What success looks like

When this handoff is complete:

  • All 73 listed units shipped at 27/27 with ≥3 H2 Master sub-sections and ≥6000 words each.
  • Chem/bio/phys depth profile lifted from ~40% to ~85% at math-style depth.
  • Coverage (breadth) unchanged — the slate is deepening, not new production except where flagged.
  • All cycle reports logged in manifests/orchestrator_status.md and manifests/weaving_report.md.
  • Skipped-T1 entries flagged in manifests/skipped_units.md for later Opus/GPT-5 pickup.
  • Forgejo origin/main accepts every push cleanly; cross-lane conflicts resolved benignly via rebase-and-append.

At that point, chem/bio/phys is "feature-complete" against the runbook §7 Wave-2 target list and at math-style depth across the board. Remaining chapter expansion (new IDs beyond the runbook list) can be handled by subsequent cycles or left for human curriculum review.

§13 Quick-start checklist for your first cycle

  1. Read §1–§5 of this doc.
  2. Read AGENT_PRODUCTION_PLAYBOOK.md §1–§14.
  3. Read CYCLE_4_STYLE_PARITY_PLAN.md §2 + §5.1.
  4. Read PRODUCTION_RUNBOOK.md §5.5 (the expert-thinking gate).
  5. Read CYCLE_5_PARALLEL_LANES.md §3 + §4.
  6. git pull --rebase origin main. Confirm clean.
  7. Read manifests/orchestrator_status.md for current state.
  8. Pick 6 units from §6 Cycle A slate (15-orgchem first half).
  9. Survey each — confirm current word count + validator score + identify one shipped sibling (likely 15.04.02 SN1/SN2 since it's the only math-style unit in 15-orgchem).
  10. Dispatch all 6 in one parallel message using §5 template.
  11. Wait for notifications. Re-validate independently.
  12. Dispatch Pass-W. Wait. Read report.
  13. Commit + push.
  14. Repeat for Cycle B onward.

Good luck.