Babel Bible — Five-Domain Expansion Plan

Drafted 2026-05-17. Successor to the implicit "math+physics only" scope (OVERVIEW.md §3 invariant 6, BRIEF.md §Scope v1) which is hereby revised. Sibling of FASTTRACK_COVERAGE_ROADMAP.md — that plan governs depth of the math/physics axis; this plan governs the lateral expansion across all five domains the site already advertises in its nav: Math · Physics · Chemistry · Biology · Philosophy.

Status: Draft. Locks the high-level structure; each domain gets its own per-domain plan (PHYSICS_PLAN.md, CHEMISTRY_PLAN.md, BIOLOGY_PLAN.md, PHILOSOPHY_PLAN.md) authored sequentially in the order defined in §6 below.

Read OVERVIEW.md (the canonical reference) first. This plan refines §11 ("what's fixed vs what's open") by moving "scope v1 = math + physics only" from Fixed to Decided-otherwise.

§1 What this plan does

It does three things and only three things:

Revises the v1 scope from "math + physics only" to "math + physics + chemistry + biology + philosophy". Decision logged in §9; OVERVIEW Decision Log to be amended on the same date.
Defines the cross-domain link contract — the rules by which a unit in one domain references a unit (or pending-unit slot) in another domain. This is the load-bearing new piece. Without it the five-domain expansion is five parallel curricula that share a header bar; with it, the cross-coherence becomes a structural property of the build, not a wish.
Orders the per-domain plans and names what each one must answer before its first unit ships.

What this plan does not do:

Pick specific apex units in chem/bio. That lives in CHEMISTRY_PLAN.md / BIOLOGY_PLAN.md.
Replace the math/physics production roadmap. FASTTRACK_COVERAGE_ROADMAP.md continues to govern the math/physics axis; this plan is orthogonal to it.
Alter the OVERVIEW invariants 1–5, 7, or 8 (tier structure, progressive disclosure, Lean-where-Mathlib-covers, reference resolution, mastery operationalization, cross-link graph, tool-agnostic content). All carry forward unchanged.

§2 Five-domain map

Each domain answers seven questions. Filled here at the high level; per-domain plans flesh out the details.

2.1 Math

Aspect	Setting
Authoring mode	Tiered units (Beginner / Intermediate / Master) per existing `UNIT_SPEC.md`.
Prereq topology	Strict DAG. Logical entailment is rigid.
Anchor literature per tier	B: 3Blue1Brown / Strogatz · I: Axler / Apostol · M: Lang / Hartshorne / Bott-Tu. (Already locked in `BRIEF.md`.)
Mastery endpoint per tier	B: read pop-math with comprehension · I: work anchor-textbook exercises cold · M: read research papers.
Lean stance	Ground truth at Intermediate+ where Mathlib covers. (Invariant §3.3.)
Cross-domain role	Spine. Other domains cite into math; math rarely cites out.
Current state	313 shipped units across sections 00–08. Active push: 79 of 93 Fast Track books unaudited (see `FASTTRACK_COVERAGE_ROADMAP.md`).

2.2 Physics

Aspect	Setting
Authoring mode	Tiered units, same scaffold as math.
Prereq topology	Multi-entry DAG, not a linear spine. Five branches: classical mech / EM+SR / stat mech / QM+QFT / GR+cosmology. Each branch has internal order; branches can be entered independently once math prereqs are met.
Anchor literature per tier	B: Feynman Six Easy Pieces / Susskind Theoretical Minimum · I: Griffiths (EM, QM) / Taylor (CM) / Schroeder (thermo) · M: Weinberg (QFT) / Wald (GR) / Peskin-Schroeder / Polchinski / Sakurai (modern).
Mastery endpoint per tier	B: explain mechanism in plain language with correct dependence on conserved quantities · I: solve anchor problem sets · M: open a recent paper in the subarea and identify novel vs background.
Lean stance	`lean_status: none` dominates (Mathlib gap); named human reviewer required. Decision 2026-04-23 carries forward.
Cross-domain role	Cites heavily into math; emits hooks out to chem (atomic structure, spectroscopy, thermo), bio (electrochemistry, photo-everything, biomechanics), phil (interpretation, time, determinism).
Current state	Section 08 stat-mech populated (9 subdirs); spin-geometry / gauge-theory units from the pilot live in §03 (those are mathematical-physics border units, not "physics proper"). No standalone classical-mech / EM / QM units yet.

2.3 Chemistry

Aspect	Setting
Authoring mode	Hybrid: tiered units + synthesis essays. Units carry the load (mechanisms, structures, exercises). Essays cover synthetic ideas that don't fit cleanly under one unit (e.g., "what reactivity is", "the bond as a many-body phenomenon").
Prereq topology	Soft DAG. Even softer than physics — gen chem can enter from atomic structure, bonding, or stoichiometry; orgo is partly its own DAG; pchem leans heavily on physics.
Anchor literature per tier	B: Crash Course / Khan · I: Atkins Physical Chemistry / Clayden-Greeves-Warren Organic Chemistry / Shriver-Atkins Inorganic · M: McQuarrie Quantum Chemistry / Levine Quantum Chemistry / Carey-Sundberg Advanced Organic.
Mastery endpoint per tier	B: predict whether a reaction will proceed in qualitative terms · I: work textbook problems including mechanism arrows and rate analysis · M: design a synthesis or interpret a research-paper spectrum.
Lean stance	`lean_status: none` everywhere; no Mathlib coverage. Human reviewer or computational-chem-package-verified (e.g., DFT result) where the claim is computational.
Cross-domain role	Cites into physics (QM, thermo, EM) and math (linear algebra, group theory for crystallography). Emits hooks out to bio (biochem, enzyme kinetics, metabolism) and phil (emergence at the molecular scale).
Excludes (lives elsewhere)	Materials science → physics (condensed-matter subarea of §09–13). Geochemistry / atmospheric chem / cosmochemistry → not in scope v1. Nuclear chem / particle-physics overlap → physics, not chem. Biochemistry mechanism (e.g., enzyme kinetics, redox chemistry, polymer reactions) belongs to chem; biochemistry as it operates in specific organisms or pathways (e.g., the citric acid cycle in S. cerevisiae) belongs to bio. Chemical engineering / process scale-up → not in scope v1.
Current state	Route stub only. No units, no essays.

2.4 Biology

Aspect	Setting
Authoring mode	Hybrid: tiered units + synthesis essays, like chemistry.
Prereq topology	Multi-tree, weakly connected. Three large independent entry trees: molecular/cellular, organismal/physiological, ecological/evolutionary. They share fewer prereqs with each other than physics branches do.
Anchor literature per tier	B: Why Evolution Is True / Crash Course Biology · I: Alberts MBoC (mol/cell) / Campbell Biology (organismal) / Futuyma Evolution · M: Alberts MBoC (deep dive) / Nelson Biological Physics / primary lit by subarea.
Mastery endpoint per tier	B: explain a biological mechanism (e.g., action potential) in plain language with correct causal direction · I: work anchor-textbook problems, interpret experimental figures · M: read a primary paper in the subarea and assess methods.
Lean stance	`lean_status: none` everywhere; reviewer-attested. (Note: certain pop-dyn / epidemiology models are mathematically tractable and could carry `partial` Lean status if Mathlib covers the relevant ODE / probability tooling.)
Cross-domain role	Cites into chem (biochem, enzyme mechanism), physics (electrochem, optics, thermo, stat mech of biopolymers), math (probability, dynamical systems, graph theory). Emits hooks out to phil (consciousness, agency, life).
Excludes (lives elsewhere)	Biochemistry mechanism → chem. Biochemistry as it operates in specific organisms / systems → bio. Medicine / pharmacology / clinical applications → not in scope v1. Psychology and cognitive science → not in scope v1 (philosophy of mind covers the conceptual overlap; empirical psych deferred to v2+). Neuroscience: cellular/molecular → bio; cognitive/computational/consciousness → phil and v2+. Bioinformatics methods → math; bioinformatics-specific tooling → not in scope v1. Public health / epidemiology models → math (probability, dyn-sys); domain-specific epidemiology → v2+.
Current state	Route stub only. No units, no essays.

2.5 Philosophy

Aspect	Setting
Authoring mode	Currently essays only (6 essays in `site/src/content/philosophy/`). May change. The reflexivity / disclosure / practice essays are written as long-form synthesis; this works for foundational orienting material but does not operationalize mastery. `PHILOSOPHY_PLAN.md` decides whether philosophy gets additional tiered units (e.g., for analytic-philosophy topics where exercises and definitions matter: logic, philosophy-of-mathematics arguments, philosophy-of-science methodology) or stays essay-only.
Prereq topology	Largely flat. Essays can be read in any order, though some sequencing is editorially recommended. Tiered units (if introduced) would have their own micro-DAGs.
Anchor literature per tier	Deferred to `PHILOSOPHY_PLAN.md`.
Mastery endpoint per tier	Deferred. If essay-only: no formal endpoint. If unit-augmented: tier-anchored as elsewhere.
Lean stance	`lean_status: none` everywhere except formal logic / set theory / Gödel-class material, which could carry Lean status via Mathlib's logic libraries.
Cross-domain role	Receives hooks from all four other domains (interpretation of QM, philosophy of biology, philosophy of mathematics, philosophy of mind). Emits hooks back into the other four as "see also: what does this mechanism imply for X" footers.
Current state	6 essays exist (`01-reflexivity` through `06-realization`). Site route live. No tiered units.

2.6 Unit vs essay decision criteria (for hybrid domains)

Chem and bio are hybrid (units + essays). Without explicit criteria authors guess, and essays accumulate faster than units because no validator gates their proliferation. Decision rule:

Author as a tiered unit if all three hold:

A testable mastery endpoint exists at at least one tier — something the unit's exercises can evidence.
A clear prereq tree can be declared in frontmatter and validated by validate_unit.py.
Master-tier reading fits in roughly 20–50 minutes, matching the unit-size invariant in OVERVIEW.md §4.

Author as a synthesis essay if any of these hold:

It spans subareas in ways that make it un-anchorable to one prereq tree (e.g., "what reactivity is, structurally" cuts across atomic structure, orbital theory, kinetics, and thermodynamics).
It is interpretive or synthetic — its job is to draw out a stance or connection, not to teach a mechanism.
It has no testable endpoint that survives operationalization ("the role of metaphor in chemistry teaching" is essay material).

Tiebreaker: prefer the unit. Default toward the gated, exercisable, machine-validatable form. Essays earn their place; they don't get it by default.

§3 The cross-domain link contract

The single most important section. The cross-coherence goal — "everything cross-correlates between math, phil, chem, bio, phys" — is achieved at build time by enforcing a contract, not at author-time by hoping people remember to link.

3.1 Two kinds of cross-references

UNIT_SPEC.md already supports [ref: source-locator] (citation into the external reference archive) and [unit-id] (cross-link within Codex). The latter is the load-bearing primitive. Cross-domain linking uses the same syntax — [03.02.01] (math), [09.02.04] (physics), [10.03.07] (chem), etc. — with the section-number prefix conveying domain.

Section-number prefixes (locked as part of this plan):

Prefix range	Domain
`00.–08.`	Math (existing; do not renumber)
`09.–13.`	Physics (5 branches × ~1 section each)
`14.–16.`	Chemistry (gen / orgo / pchem-inorg)
`17.–19.`	Biology (mol-cell / organismal / eco-evo)
`20.–20.`	Philosophy (if it gets units)

Routing rule (locked). Physics units are addressable in three places, and this is not a bug to be fixed by renumbering:

03.* — math-physics border units already shipped under modern-geometry (Clifford, Spin, Dirac, Yang-Mills, Atiyah-Singer, Chern-Weil, Bott periodicity, etc.). These stay where they are; renumbering would break ~50+ shipped cross-refs across manifests/deps.json and existing units. Their primary discipline is mathematics; the physics interest is downstream.
08-stat-mech.* — mathematical-physics flavor (Onsager, RG, lattice gauge, mathematical path integrals, etc.). Stays put for the same reason.
09–13.* — physics proper going forward: classical mech, EM+SR, GR+cosmology, QM+QFT, plus a phys-flavor stat-mech section that may eventually co-exist with §08 or absorb its physically-flavored siblings (decision deferred to PHYSICS_PLAN.md).

The deciding question for new units is "what is the primary discipline?" Math → 00–08. Physics → 09–13. Border units default to math; cross-domain hooks_out from a §03.* unit to a §09–13.* physics unit are the mechanism by which math-physics border units expose themselves to the physics nav. Physics landing page (/physics) and /dag filtering query both the 09–13 namespace and the union of hooks_out from math-side border units that target physics.

Philosophy IDs. Existing 6 essays at site/src/content/philosophy/01-reflexivity … 06-realization will be retrofitted with IDs 20.essays.01 … 20.essays.06 so other domains can hooks_out to them and the validator can resolve those targets. Whether philosophy also grows a tiered-units ID space (e.g. 20.01.NN for tiered chapters) is a PHILOSOPHY_PLAN.md decision; the prefix table reserves 20 for philosophy as a whole.

3.2 Inbound vs outbound

A unit declares both:

prerequisites:          # inbound: required reading
  - 03.02.01            # math, linear algebra
  - 09.04.02            # physics, classical mech: Hamiltonian
hooks_out:              # outbound: this unit is referenced BY
  - target: 14.05.03    # chem, molecular orbitals — proposed
    kind: proposed
    why: "MO theory cites this unit's spin-statistics result"
  - target: 17.04.02    # bio, photosynthesis — proposed
    kind: proposed
    why: "Light-harvesting cites this unit's photon-absorption section"
  - target: 14.03.04    # chem, hybridization — confirmed
    kind: confirmed
    confirmed_by: "reviewer-id"
    why: "Hybridization picks up the orbital-angular-momentum decomposition stated here"

prerequisites already exists. hooks_out is new, with two states:

proposed — the emitting author declared the link. The target may not exist yet, and even if it does, no reviewer from the receiving domain has audited the semantic correctness of the link. This is the default state.
confirmed — the target unit is shipped AND a reviewer from the receiving domain has signed off that the linked claim is correct on their side. Carries a confirmed_by: reviewer-id field as lightweight attestation.

The validator never auto-flips proposed → confirmed. Promotion is an explicit step recorded when the receiving domain produces or revisits the target unit. This is the failure mode the binary pending/shipped design missed: a target unit existing is not the same as a target unit being a true counterpart on the receiving side.

The hooks_out field is non-blocking — neither proposed nor confirmed hooks fail the build. prerequisites is partially blocking — see §3.4.

3.3 The connections manifest

manifests/connections.json (already exists per OVERVIEW.md §9) gets a new role: it becomes the bidirectional cross-domain edge index built from all prerequisites + hooks_out declarations across every unit. Today it logs "semantic connections"; under this plan it's the authoritative cross-domain graph the site queries to render the "this unit is referenced by …" sidebar and the cross-domain DAG view at /dag.

Migration of existing rows. Existing connections.json content is preserved and tagged with a source: legacy-semantic discriminator. Auto-generated rows from prerequisites carry source: prereq; rows from hooks_out carry source: hook plus the kind (proposed/confirmed). The validator patch in §6 step 3 emits the auto-generated rows in a one-time sweep; legacy rows are reviewed during the cross-domain audit (§6 step 9) and either preserved as-is, upgraded to a declared hook on a specific unit, or retired.

3.4 Build-time validation

scripts/validate_unit.py extends with:

hooks_out.target must be a syntactically valid unit ID — section.chapter.unit form, or 20.essays.NN for philosophy essays.
hooks_out.kind is proposed or confirmed. confirmed requires (a) the target unit exists with status: shipped, and (b) a confirmed_by: reviewer-id field is present on the hook entry.
prerequisites entries that are cross-domain follow the same pending_prereqs mechanism as intra-math prereqs (per existing UNIT_SPEC.md rules): they may be pending, with the pending edges registered in manifests/deps.json and the unit's frontmatter carrying pending_prereqs: true. Aligning with the existing rule rather than inventing a stricter parallel system avoids author confusion and lets the existing pending-deps tooling extend naturally.
hooks_out.why is required and ≥ 30 chars (forces the author to articulate the link reason).

3.5 Why this contract is enough (and what it doesn't catch)

It catches:

Dangling cross-refs (build fails).
"I forgot to think about whether bio cares about this physics result" (the unit can ship with hooks_out: [], but reviewers can flag empty hooks lists on units whose subject obviously projects into other domains).

It does not catch:

A genuine missing connection that no author thought of. Mitigated by the connections.json review pass after each domain's first wave ships — a separate audit deliverable in each per-domain plan.
Semantic incorrectness of a stated hook ("this chem unit cites that physics unit, but the physics result doesn't actually imply what the chem unit claims"). That's a reviewer judgment, not a build check — addressed by the confirmed state's reviewer attestation, but only on hooks that get promoted.

3.6 Pending-hook sweep — preventing rot

Non-blocking proposed hooks can accumulate forever if no one harvests them. Three mechanisms keep them honest:

Per-domain plan absorption. At drafting time, each per-domain plan queries manifests/connections.json for kind: proposed edges where target is in its domain. These become suggested first units in the plan's backlog. Not mandatory ("you must build everything other domains proposed"), but the per-domain plan must explicitly list each proposed inbound hook and decide per hook: accept (queue for production), reject (the proposing unit's why doesn't survive scrutiny — proposing author edits or removes the hook), or defer (mark priority: low and leave for later waves).
Quarterly staleness sweep. A periodic scan of long-lived proposed hooks: if a hook has been proposed for > 6 months with no movement, surface it for promote/reject/retire review. Mechanical reminder, not enforcement.
The cross-domain audit step (§6 step 9). First formal place this happens across all five domains simultaneously.

§4 Coverage philosophy: how each domain decides "what to include"

Math has Fast Track as its coverage spec (93 books → audit → unit production). Physics has partial coverage from Fast Track (§2 Quantum, §3 advanced topics). Chem, bio, and phil have no analogous canonical list.

Per-domain plans must answer:

Domain	Coverage source candidate
Math	Fast Track booklist + pre-calc ramp. Already locked.
Physics	Fast Track §2 + 't Hooft / Rigetti syllabi for the parts FT thins out (classical mech ramp, intro QM).
Chemistry	No single canonical roadmap exists. Candidates: ACS undergrad chemistry curriculum + graduate-level subarea standards (Atkins / Clayden / McQuarrie).
Biology	Even less canonical. Candidates: Alberts MBoC as the molecular-cell spine; Campbell as the organismal/integrative spine; Futuyma as the evolutionary spine. Three spines, not one.
Philosophy	Stays essay-driven for foundational material. Tiered-unit coverage (if introduced) would be subject-narrow (logic; philosophy of mathematics; philosophy of physics; philosophy of biology; philosophy of mind).

The point: don't pretend every domain admits a Fast Track equivalent. Chemistry and biology need their own coverage scaffolds, designed in their per-domain plans. Forcing a Fast-Track-shaped scaffold onto a domain that doesn't have one is a recipe for a curriculum with arbitrary holes.

§5 Mastery-endpoint operationalization across domains

The OVERVIEW invariant is "mastery is operationalized, not vibes." That carries forward, but the form of the test varies by domain:

Math: prove a theorem, solve a proof exercise, fill in a Lean tactic.
Physics: derive a result, predict a measurement, identify a regime.
Chemistry: propose a synthesis, predict a spectrum, push curly arrows, justify a rate law.
Biology: read an experimental figure and identify confound + conclusion; predict a mutation phenotype; trace a metabolic pathway.
Philosophy: reconstruct an argument; identify hidden premises; produce a counterexample; map a position onto its commitments.

Honest machine-verifiability ranking. Not all of these are auto-gradable, and the project should not pretend otherwise:

Domain	Auto-gradable share at I/M tier	Reason
Math	High where Mathlib covers; medium where it doesn't	Lean compile is the gate; falls back to human review for the gaps
Physics	Medium-low	Symbolic derivations partially checkable; predictions and regime-identification are not
Chemistry	Low	No widely-used auto-grader for curly-arrow notation; spectra prediction requires DFT-level tooling; synthesis design is essentially open-ended
Biology	Very low	Figure-reading, phenotype prediction, pathway tracing are all judgment-heavy
Philosophy	Very low	Argument-reconstruction is the canonical "needs a human" task

This means chem, bio, and phil rubrics at Intermediate / Master tier will be human-review-heavy. The validator can check structure (frontmatter, citation resolution, hooks_out well-formedness) but not the semantic correctness of mastery exercises. REVIEWER_PLAN.md per-domain extensions must plan for this — it's the bottleneck OVERVIEW.md already flagged at ~750hr for math alone, and the four less-mechanizable domains multiply it rather than divide it.

Per-domain plans expand these into per-tier rubrics analogous to QUALITY_RUBRIC.md's pedagogical-reviewer checks. Each rubric extension is part of that domain's first deliverable.

§6 Order of operations

Sequence chosen to validate the cross-domain link contract on the highest-leverage pair first (physics ↔ math, since math is already populated and physics has the richest hook density into math).

Order	Deliverable	Owner	Blocks
1	This plan (`BIBLE_EXPANSION_PLAN.md`)	drafted 2026-05-17	everything below
2	`UNIT_SPEC.md` patch: add `hooks_out` field + section-number-prefix table	next session	all per-domain plans
3	`scripts/validate_unit.py` patch: hooks_out validation	next session	shipping any cross-domain hook
4	`PHYSICS_PLAN.md`	next	first physics unit
5	First 5 physics units (one per branch — classical, EM, stat mech, QM, GR)	after #4	validates the link contract; calibrates physics rubric
6	Retrospective: did the link contract work? amend §3 if needed	after #5	chemistry/biology plans
7	`CHEMISTRY_PLAN.md` + `BIOLOGY_PLAN.md` (parallel, both informed by physics retro)	after #6	first chem/bio units
8	First 5 chem + 5 bio units	after #7	cross-domain audit
9	Cross-domain audit pass: review `manifests/connections.json` for missing edges	after #8	`PHILOSOPHY_PLAN.md`
10	`PHILOSOPHY_PLAN.md` (decides essay-only vs unit-augmented, informed by what cross-refs the other four want to send into phil)	after #9	philosophy rebuild if needed

Rationale for putting philosophy last:

Philosophy's role is to receive hooks more than emit them. We learn what shape phil needs to be by seeing what the other four point at.
The existing 6 essays aren't blocking anything; they continue to serve readers while the plan is finalized.

§7 What's locked by this plan vs what's still open

Locked

Five-domain scope (math + physics + chem + bio + phil).
All OVERVIEW.md §11 fixed items except "scope v1 = math + physics" carry forward unchanged (tier structure, progressive disclosure, Lean stance, reference resolution, mastery operationalization, cross-link graph, tool-agnostic content).
Section-number-prefix allocation + routing rule for the fragmented physics namespace (§3.1).
hooks_out field shape with proposed / confirmed states (§3.2).
Cross-domain prereqs follow the existing pending_prereqs mechanism — they may be pending, registered in manifests/deps.json (§3.4). Aligned with the intra-math rule rather than inventing a stricter parallel.
Unit-vs-essay decision criteria for hybrid domains (§2.6).
Pending-hook sweep convention (§3.6).
Order of operations (§6).

Open (deferred to per-domain plans)

Specific apex units for chem, bio, phil.
Whether phil gets tiered units in addition to essays.
Anchor-text picks at each tier for chem, bio, phil.
Specific coverage roadmaps for chem and bio.
Per-domain reviewer roster (mirrors REVIEWER_PLAN.md per-domain).
Whether stat-mech (currently §08, mathematical-physics-flavored) gets re-homed under §11 once §09+ physics sections exist.

Open (deferred to a later epoch — explicitly not v1)

Computer science as a sixth domain. (Existing global CLAUDE.md mentions CS as a v2 candidate; this plan does not relitigate that.)
Translation / localization (still English-only).
K–8 arithmetic / pre-algebra ramp (still out of scope).

§8 Risks and how each is mitigated

Risk	Mitigation
Five domains × per-tier rubrics = 15 rubrics to maintain.	Per-tier mastery endpoints (§5) replace per-domain rubric duplication for the human-judgment portion. The automated validator rubric stays one file.
Chem and bio have no Fast-Track-equivalent → coverage holes go undetected.	Per-domain plans must explicitly define their coverage source (§4 table). If no canonical list exists, the plan invents one and names what was excluded.
`hooks_out` becomes lip service ("everybody declares an empty list, nobody links").	Reviewer checklists for chem/bio/phys include "scan this unit's claims — name three plausible hooks into other domains; if none, justify." Codified in per-domain plans.
Five-domain expansion stalls math production.	Math production is on a separate track (`FASTTRACK_COVERAGE_ROADMAP.md`); the per-domain plans here are sequenced so that physics-1st only consumes ≤1 reviewer-week before pausing for retrospective. Math axis keeps moving in parallel.
Cross-domain reviewer expertise is rare (few people are credible in both physics and bio).	Hooks-out reviewer is the emitting domain's reviewer (they justify why the link makes sense from their side). The receiving domain's reviewer audits incoming hooks during their own production passes. No reviewer is asked to credential two domains simultaneously.
Philosophy rewrite churn — the 6 existing essays may need restructuring if `PHILOSOPHY_PLAN.md` introduces units.	Essays are kept intact; tiered units (if introduced) live alongside them. Site nav exposes both. No deletions; only additions.

§9 Decision log entries (to be added to `OVERVIEW.md` §12 on the same date)

Date	Decision	Rationale
2026-05-17	v1 scope expanded from "math + physics" to "math + physics + chemistry + biology + philosophy"	Cross-domain coherence is the project's distinct thesis; math+physics alone cannot demonstrate it. Site nav already advertised the five domains; this aligns scope with the surface.
2026-05-17	Cross-domain link contract introduced: `hooks_out` field with two states (`proposed` / `confirmed`); cross-domain `prerequisites` follow existing `pending_prereqs` mechanism rather than a stricter parallel rule; section-number-prefix table locks domain identity with a routing rule for the fragmented physics namespace (§03 math-physics border units + §08 math-flavor stat-mech + §09–13 physics proper)	Mechanical enforcement prevents the cross-coherence goal from degrading to lip service. Two-state hooks separate "author declared it" from "receiving-domain reviewer signed off"; the original `pending/shipped` binary conflated those. Aligning cross-domain prereqs with the existing pending mechanism avoids author confusion.
2026-05-17	Philosophy essay IDs retrofitted to `20.essays.NN` so cross-refs can target them	Without IDs, other domains have no validated way to `hooks_out` to existing essays.
2026-05-17	Chem and bio adopt hybrid authoring (tiered units + synthesis essays); phil mode decided in `PHILOSOPHY_PLAN.md`	Essays alone can't operationalize mastery; tiered-units alone can't carry trans-disciplinary syntheses. Hybrid covers both.
2026-05-17	Unit-vs-essay decision criteria locked (§2.6): unit if testable endpoint + clear prereq tree + 20–50min scope; essay if multi-subarea, interpretive/synthetic, or has no testable endpoint; tiebreaker defaults to unit	Without explicit criteria authors guess; essays accumulate worse than units because they have no validator gating proliferation.
2026-05-17	Explicit exclusions added to chem and bio scope tables (§2.3, §2.4)	Without scope fences (materials → physics; pharmacology / psych / clinical → not v1; biochem-mechanism vs biochem-in-organism boundary), chem and bio balloon during per-domain planning.
2026-05-17	Pending-hook sweep convention (§3.6): each per-domain plan harvests `hooks_out → its domain` as suggested backlog; quarterly staleness sweep on long-lived `proposed` hooks	Prevents `proposed` hooks from accumulating without ever being audited, built, or retired.
2026-05-17	Honest machine-verifiability ranking added (§5): chem, bio, phil rubrics at I/M are explicitly human-review-heavy with no Lean-equivalent	Reviewer bandwidth was already flagged as the math bottleneck (~750hr); the four less-mechanizable domains multiply it, and pretending otherwise would set unrealistic scaling expectations.
2026-05-17	`manifests/connections.json` migration: existing "semantic connections" rows tagged `source: legacy-semantic`; new auto-generated rows tagged `source: prereq` or `source: hook`	Avoids destroying existing content while clarifying provenance.
2026-05-17	Per-domain plans authored sequentially: PHYSICS → (retro) → CHEM ∥ BIO → (audit) → PHIL	Physics validates the link contract with the densest math-side hook surface; phil last because it's mostly a receiver.

§10 Next immediate actions

Patch docs/specs/UNIT_SPEC.md: add hooks_out field schema (with proposed/confirmed states, confirmed_by: for confirmed entries, required why ≥ 30 chars); add §3.1 section-number-prefix table + routing rule; document 20.essays.NN philosophy-essay IDs.
Patch scripts/validate_unit.py: enforce hooks_out per §3.4; extend pending_prereqs validator to accept cross-domain pending edges registered in manifests/deps.json; emit source: prereq / source: hook rows into manifests/connections.json; tag existing rows as source: legacy-semantic.
Patch OVERVIEW.md: append §9 decision-log rows (10 new rows per §9 of this plan); revise §11 fixed-vs-open table (move "scope v1 = math + physics" to the open/superseded column with a pointer to this plan); confirm the other 7 §11 fixed items carry forward unchanged.
Retrofit IDs on the 6 philosophy essays (20.essays.01 … 20.essays.06) and update the philosophy route to surface them so the validator can resolve cross-refs.
Draft docs/plans/PHYSICS_PLAN.md — the first per-domain plan. Must include: the harvested backlog of hooks_out → physics proposed edges from manifests/connections.json (empty at first; populated as math units start declaring them).

This plan is the canonical reference for the five-domain expansion. When uncertain about cross-domain mechanics, check here. When updating, propagate to OVERVIEW §12 the same day.