Domain Inventory — Per-Arrangement Quick Reference

Status: Live operational doc. Cross-arrangement quick-reference for the methodology computation framework. Use when picking up after a break or switching arrangements. Pairs with project-tracking.md (phase/decision log) and probability-model-architecture.md (registry design).

How to use: When starting a session, scan the "Arrangements at a glance" table to find the arrangement you care about, then jump to its section for files + scripts + figures + how-to-rerun.

Arrangements at a glance

Pilots status as of

1. Technogenesis (entity) — complete. Reuses git-evolution.v1.json + github-evolution.v1.json. 4 pop-contexts authored.
2. Phylogenesis (biology) — complete. 5-snapshot stem trajectory + plant-branch trajectory + 4 major-transition pop-contexts.
3. Cell-evolution-post-LUCA (biology) — complete. Bridge trajectory connecting abiogenesis (R2-LUCA) to phylogenesis (ecoli). The three trajectories now chain continuously across two arrangements.
4. Proto-eukaryote bundle (biology) — complete. Transitional/crystallizing-bundle case study mirroring proto-replicator+proto-compartment from abiogenesis. Validates the bundle pattern on a non-abiogenesis instance.
5. Backwards-inference demo (entity + abiogenesis) — complete. Wires up existing lib/rate_corridor.py machinery to a chart via compute/scripts/plot_backwards_inference.py. Demo on Git substrate + abiogenesis R0→R2.
6. Ontogenesis (biology) — pending. Heaviest data lift; need 4-6 new stage Mns of one organism + pop-contexts.

Standalone domains (not engine arrangements)

Domains analyzed on their own terms, not as part of the three realization-chain engine arrangements above. They are first-class (see methodology.md §1 — the methodology is not the SSA; standalone domain analysis is normal). Listed here so a cold return doesn't miss them by scanning only the arrangement table.

Parked research (post-release, do not re-open before then): a candidate multi-edge / return-direction primitive domain on the SSA's Selection role — captured with resume instructions in communication_domain_analysis/PARKED-selection-as-return-direction-domain.md. The Fx/cost-primitive gap in the agentic ecosystem domains (cultural + digital) is a real, separable observation noted there too.

Common infrastructure (arrangement-neutral)

These apply to every arrangement — no per-arrangement equivalents needed.

Compute lib (compute/lib/)

• loader.py — arrangement / domain / bridge / walk loading
• lattice.py — Position, tangent_set, canonical positions
• corridor.py — BFS, gates analysis
• rate_corridor.py — forward/backward rate-weighted mass propagation
• probability_models.py — registry + 3 baselines + 3 implemented + 4 placeholders (see Models)
• cache.py — mtime-based skip caching
• header.py — L4 header validation
• util.py — shared helpers (safe_load_json, _quantity_value, _bernoulli_at_least_one, _entropy_2, _bounds_from_walk, _gather_primitives, _floor_grid)

Arrangement-neutral plot scripts (compute/scripts/)

• plot_corridor.py — width bars + entropy
• plot_corridor_gates.py — width bars + active deps + floor heatmap
• plot_rate_corridor.py — structural vs effective width + mass concentration + floor heatmap
• plot_hexagon.py, plot_hasse.py, plot_landscape.py, plot_paths.py, plot_ridge.py — lattice geometry
• plot_mn_profile.py, plot_mn_sweep.py — manifestation projections
• plot_chain_summary.py — per-arrangement status board
• plot_phase_cascade.py — phase-transition gating
• plot_trajectory.py — line plot through epochs
• topology_dot.py — DOT/SVG/PNG topology renderer

Compute scripts (analysis-neutral)

• compute_corridor.py — corridor BFS for any arrangement+walk
• validate_data.py, verify_corridors.py — schema + canonical filter checks
• coherence_distance.py, joint_coherence.py, check_coherence.py — Mn coherence checks
• coupled_trajectory.py — Mn + scaffold trajectory rendering
• aggregate_population.py, population_match.py, neighborhood.py, summary.py, report.py

Bundle-aware compute (compute/lib/bundle.py)

Arrangement-neutral helpers for multi-constituent (bundle) analysis:

• position_at_time(trajectory, timestamp) — last-known-position lookup with lex-sortable date strings
• joint_position(positions, strategy='max') — max-merge constituent positions into bundle position
• collect_timestamps(trajectories) — unified time axis across constituents
• position_summed_rank, position_per_chain_level, constituent_chain_levels — aggregation helpers

Plot scripts (after consolidation):

• compute/scripts/plot_bundle.py — generic config-driven bundle plotter. Takes --config <path> pointing to a JSON config in compute/configs/bundles/*.json. Currently configured for Git+GitHub (git-github.json) and eukaryogenesis (eukaryogenesis-proto-eukaryote.json) cases. Supports phase shading, bundle-formation + crystallization annotations, broader-walk-context boxes, optional constituent-launch annotations.
• compute/scripts/plot_bundle_trajectory_abiogenesis.py — specialized variant retained for the abiogenesis case because it includes a domain-specific compartment-substrate-shift middle panel (mineral pore → micelle → vesicle → cell membrane succession with active/ambient hatching). Not generalized into plot_bundle.py because that panel uses arrangement-specific primitives (env.St, substrate.Cmp, bridge.Cmp+Gr) and there's only one instance.

python3 compute/scripts/plot_bundle.py --config compute/configs/bundles/git-github.json
python3 compute/scripts/plot_bundle.py --config compute/configs/bundles/eukaryogenesis-proto-eukaryote.json
python3 compute/scripts/plot_bundle_trajectory_abiogenesis.py

Adding a new bundle case: author 2+ constituent trajectories under data/trajectories/, write a config under compute/configs/bundles/<name>.json, run plot_bundle.py --config compute/configs/bundles/<name>.json. No code changes.

Composition overview (compute/scripts/plot_composition_overview.py)

Cross-arrangement pattern-spotting chart for any trajectory file. Three panels: (A) total summed-rank line, (B) per-chain-level stacked area, (C) per-primitive composition heatmap with phase-transition cell highlighting. Run as:

python3 compute/scripts/plot_composition_overview.py data/trajectories/<name>.v1.json

Outputs output/figures/composition-overview-<name>.png. Used to surface patterns: stable rows (constant primitives), diagonal climbs (driving primitives), discrete jumps (phase transitions), substrate-freeze (cells stop transitioning). Works on git-evolution, abiogenesis-trajectory, github-evolution, and any future trajectory files.

Pair-trajectory matrix (compute/scripts/plot_pair_matrix_trajectory.py)

NxN matrix combining pair-relationship structure with trajectory dynamics — rendered as upper-triangular + diagonal (symmetric so the lower triangle is redundant). Each cell shows the pair's evolution over a trajectory's snapshots: off-diagonal cells are 2D phase-space plots (col primitive on x, row primitive on y) tracing the joint trajectory through level-space; diagonal cells are 1D timelines of the primitive's level. Snapshot direction shown via viridis color gradient (light=early, dark=late) + end-arrow. Cell background tinted by pair intensity.

python3 compute/scripts/plot_pair_matrix_trajectory.py \
    data/domains/entity-system.v1.json data/trajectories/git-evolution.v1.json

Outputs output/figures/pair-matrix-trajectory-<domain>-<trajectory>.png. Cell shapes reveal pair dynamics:

• diagonal climb = co-evolution
• L-shape = sequential advance (one primitive rises before the other)
• path stuck at corner = persistent gap
• quick climb then plateau = substrate-freeze
• wiggle = back-and-forth dynamics

Existing renders surface contrasting trajectory signatures:

• entity-system × git-evolution: top-left {E,I,T} block shows substrate-freeze (early climb then plateau); bottom-right {M,X,P} block shows persistent gaps (paths stuck at origin); cross-quadrant cells show one-active-one-dormant L-shapes
• entity-system × github-evolution: nearly all cells show diagonal climbs into upper-right; gaps fill, M and X primitives genuinely advance
• abiogenesis-substrate × abiogenesis-trajectory: all cells show diagonal climbs (continuous co-evolution; no gaps, no freezes)

Pair-relationship matrix (compute/scripts/plot_pair_matrix.py)

NxN cross-correlation matrix view of a domain's pair-relationships — rendered as upper-triangular + diagonal (symmetric). Reads domain.v1.pair_relationships and domain.v1.compositions. Diagonal cells show primitive identity (with level if --mn provided); off-diagonal cells show pair relationships colored by intensity (heavy/medium/light/negligible). Top-right of each cell shows composition-membership glyphs (△triad, ◇quad, ⬡higher) so you can see which pairs anchor the most structures. With --mn overlay, cells display level-product (e.g., "4×0") with bold = both active, red = one-active+one-zero (incomplete pair), gray = both dormant. Multiple --mn flags produce small-multiples for shape comparison.

python3 compute/scripts/plot_pair_matrix.py data/domains/entity-system.v1.json
python3 compute/scripts/plot_pair_matrix.py data/domains/entity-system.v1.json --mn github --mn nostr --mn postgres

Outputs output/figures/pair-matrix-<domain>[-mns-<list>].png. Key analytical use: surfaces "completeness signatures" of manifestations — github fires nearly all pairs (saturated), nostr shows T-row-and-column-all-red (T=0 hole, breaks all triads touching T), postgres shows I-row-and-column-all-red (I=0 hole). Different primitive omissions produce visually distinct matrix signatures.

Combinatorial structure diagram (compute/scripts/plot_combinatorial_structure.py)

Domain-level visualization of the FULL combinatorial relational structure. Reads domain.v1.pair_relationships and domain.v1.compositions directly — no hand-specified subsets. Renders all primitives as nodes (circular layout), all pair-relationships as edges colored by intensity (heavy/medium/light/negligible), all documented compositions as filled colored polygons (triads, quads, higher-arity). Optional --mn flag overlays a manifestation's primitive levels (node sizes + colors). Handles range-positioned population Mns (vertebrates et al.) via min-max averaging.

python3 compute/scripts/plot_combinatorial_structure.py data/domains/entity-system.v1.json
python3 compute/scripts/plot_combinatorial_structure.py data/domains/entity-system.v1.json --mn nostr
python3 compute/scripts/plot_combinatorial_structure.py data/domains/application-architecture.v1.json --skip-negligible

Outputs output/figures/combinatorial-structure-<domain>[-mn-<name>].png. Existing renders cover 6 domains (entity-system, chemistry, biology-substrate, application-architecture, organism-architecture, digital-ecosystem) plus 9 manifestation overlays (entity: github / postgres / instagram / nostr / git; biology: ecoli / yeast / human / vertebrates). Reveals at-a-glance: which compositions are "active" in a Mn (members all > 0), which are gaps, which pairs span heavy edges.

Primitive subset analysis (compute/scripts/plot_primitive_subset.py)

Generic chart for analyzing how a SPECIFIC SUBSET of primitives (a heavy pair, core triad, quartiad, or any analyst-defined subset) evolves through a trajectory. Three panels: (A) per-primitive level lines, (B) subset rank vs total rank with share %, (C) subset composition heatmap with transition highlighting. CLI: trajectory file + repeated --primitive chain_level.primitive_id + --label + --kind. Run as:

python3 compute/scripts/plot_primitive_subset.py data/trajectories/git-evolution.v1.json \
    -p entity-system.E -p entity-system.I -p entity-system.T \
    --label "Entity Core Triad EIT" --kind triad

Outputs output/figures/primitive-subset-<trajectory>-<label-slug>.png. Surfaces three "expansive dynamics" patterns: substrate-then-freeze, persistent gap, emergent dominance. Operates on any documented combinatorial structure from domain.v1.compositions or domain.v1.pair_relationships.

Existing demonstration outputs:

• primitive-subset-git-evolution-entity-core-triad-eit.png — substrate-then-freeze pattern
• primitive-subset-git-evolution-reactive-dispatch-triad-tmx.png — persistent gap pattern (reactive dispatch absent in Git solo)
• primitive-subset-abiogenesis-trajectory-abiogenesis-substrate-quartet-rgcmpp.png — emergent dominance pattern

Schemas (compute/schemas/)

All v1, all schema-valid:

• _l4_header.v1.schema.json — common artifact header (Fw/Sc/Mn/Cx/Ls/Tj/Cpl + provenance)
• arrangement.v1.schema.json, domain.v1.schema.json, bridge.v1.schema.json, topology.v1.schema.json
• walk.v1.schema.json, corridor.v1.schema.json
• manifestation.v1.schema.json, trajectory.v1.schema.json
• rate.v1.schema.json, population_context.v1.schema.json

Make targets (compute/Makefile)

Most are walk-parameterized — pass WALK=<name> and ARRANGEMENT=<name> via env. See make help. Notable:

• make validate — schema-validate all data/**/*.json
• make verify — canonical filter % verification
• make all-outputs — corridor + report + topology DOT for the ARRANGEMENT
• make plot-corridor, plot-corridor-gates, plot-rate-corridor, plot-hexagon, plot-hasse, plot-paths, plot-ridge, plot-landscape, plot-mn-profile, plot-trajectory
• make report, make topology-dot

Abiogenesis arrangement

Single-domain-extending-to-multi-chain arrangement. Sub-resolution (R/G/Cmp/P) split out from biology.

Chain levels

chemistry → chemistry-to-biology-bridge → abiogenesis-substrate → environment-context

Data files

• Arrangement: data/arrangements/abiogenesis.v1.json
• Topology: (none — uses chemistry's topology implicitly via the chain)
• Domains: data/domains/{chemistry, abiogenesis-substrate, environment-context}.v1.json (chemistry-to-biology-bridge in data/bridges/)
• Bridges: data/bridges/chemistry-to-biology-bridge.v1.json
• Manifestations (5): abiogenesis-r0, abiogenesis-r0p2, abiogenesis-r1, abiogenesis-r1p7, abiogenesis-r2-luca (all in data/manifestations/)
• Walks: data/walks/abiogenesis-r0-to-r2.v1.json
• Trajectories (3): proto-replicator-evolution.v1.json, proto-compartment-evolution.v1.json (constituent isolate trajectories that max-merge to the unified abiogenesis Mns); the 5 Mns + walk also IS the unified trajectory implicitly.
• Rates: data/rates/abiogenesis-placeholder.v1.json
• Population contexts (4): abiogenesis-r0-hadean, abiogenesis-r0p2-hadean, abiogenesis-r1-hadean, abiogenesis-r1p7-hadean (all in data/population_contexts/)

Arrangement-specific scripts

• plot_abiogenesis_comprehensive.py — corridor + Mn placement + rate-weighted mass + SSA + per-Mn pop ctx + typical-path overlay
• plot_abiogenesis_extended.py — composite scroll-down (3007×4865 px) of comprehensive + SSA + sensitivity
• plot_abiogenesis_sensitivity.py — N-sweep across 3 scenarios × 4 transitions via probability registry
• plot_ssa_hardening.py — En/Vr/Mc/Sf/Cx/Cm/Se hardening trajectory
• compute/lib/ssa_roles.py — SSA-role overlay (abiogenesis-arrangement-specific)

Key output figures (output/figures/)

• abiogenesis-comprehensive.png, abiogenesis-extended.png, abiogenesis-sensitivity.png
• ssa-hardening-trajectory.png
• abiogenesis-r0-to-r2-corridor-gates.png, abiogenesis-r0-to-r2-rate-corridor-abiogenesis-placeholder.png

How to rerun

cd compute
python3 scripts/plot_abiogenesis_comprehensive.py
python3 scripts/plot_abiogenesis_sensitivity.py
python3 scripts/plot_ssa_hardening.py
python3 scripts/plot_abiogenesis_extended.py  # composes the above

Status notes

• Phase 3+4 closed. Comprehensive analysis complete + validated against canonical (4/4 high-confidence claims).
• Future: implement remaining placeholder models (vesicle_group_selection, bootstrap_expansion, darwinian); ground per-trial probabilities in literature.

Biology arrangement

Full biology chain. Closed Phase 2 phylogenesis pilot complete.

Chain levels

dirac-substrate → chemistry → chemistry-to-biology-bridge → biology-substrate → biology-to-organism-bridge → organism-architecture → organism-to-ecosystem-bridge → ecosystem → environment-context

Data files

• Arrangement: data/arrangements/biology.v1.json
• Topology: data/topologies/biology.v1.json
• Domains (8): dirac-substrate, chemistry, biology-substrate, organism-architecture, ecosystem, environment-context (+ shared abiogenesis-substrate, + nothing extra)
• Bridges: dirac-to-chemistry, chemistry-to-biology, biology-to-organism, organism-to-ecosystem
• Manifestations (6): ecoli, yeast, arabidopsis, drosophila, vertebrates, human (all adult-stage Mns in data/manifestations/)
• Walks (7): dirac-substrate-min-to-max, chemistry-min-to-max, chemistry-to-biology-crossing, biology-substrate-min-to-max, biology-to-organism-crossing, organism-architecture-min-to-max, ecosystem-min-to-max, environment-context-min-to-max
• Trajectories (1): phylogenesis-stem.v1.json — 5 snapshots ecoli (~3.5 Gya) → yeast/LECA (~2.0 Gya) → drosophila (~0.6 Gya) → early-vertebrate (~0.52 Gya) → human (~0 Gya)
• Rates: data/rates/biology-substrate-placeholder.v1.json
• Population contexts (4): phylogenesis-prokaryote-to-eukaryote, phylogenesis-eukaryote-to-bilaterian, phylogenesis-bilaterian-to-vertebrate, phylogenesis-vertebrate-to-human. Models exercised: frozen_accident (eukaryogenesis, with reference_dependency_count=12 cellular-machinery-scale override), darwinian placeholder (3 transitions, falls back through population_sampled).

Key output figures

• biology-topology.png/.svg/.dot
• Per-domain hexagon/hasse/landscape/paths/ridge: biology-substrate-*, chemistry-*, organism-architecture-*, ecosystem-*, dirac-substrate-*
• Bridge crossings: chemistry-to-biology-bridge-*, biology-to-organism-crossing-*
• biology-mn-sweep.png, biology-chain-summary.png, biology-phase-cascade.png
• Phylogenesis pilot: composition-overview-phylogenesis-stem.png, primitive-subset-phylogenesis-stem-central-dogma-triad-gtr.png, pair-matrix-trajectory-biology-substrate-phylogenesis-stem.png, pair-matrix-trajectory-biology-to-organism-bridge-phylogenesis-stem.png

How to rerun

Generic plot scripts work against the biology arrangement directly:

cd compute
make all-outputs ARRANGEMENT=biology
make plot-mn-profile ARRANGEMENT=biology DOMAIN=biology-substrate

Status notes

• Phase 2 closed. All Mns coherent (after 4 bridge dep mis-spec fixes).
• Empirical finding worth following: ecosystem and digital-ecosystem both compute to 7.23%/512 — possible ecosystem-class convergent filter.
• Phylogenesis pilot complete: stem trajectory + 4 major-transition pop-contexts authored. Tests frozen_accident (eukaryogenesis with cellular-machinery reference scale 12) + darwinian placeholder (3 transitions, falls back through population_sampled). Cross-arrangement plot scripts (composition-overview, primitive-subset, pair-matrix-trajectory) ran on biology data unchanged. Hypothesis-grade observation: phylogenesis exhibits within-substrate co-evolution in early phase (snap 0→2, eukaryogenesis + multicellularity) and substrate-frozen + bridge-elaboration in late phase (snap 2→4, vertebrate origin + mammalian/human). One trajectory shows BOTH the abiogenesis-shape and technogenesis-shape at different times. See discipline doc for full caveats; needs ontogenesis + multi-VCS pilots to validate.
• Open follow-up trajectories on biology: plant-branch (yeast → arabidopsis), proto-eukaryote bundle (endosymbiont + host crystallizing into LECA, mirroring proto-replicator+proto-compartment), cell-evolution-past-R2-LUCA (connects abiogenesis to phylogenesis).
• Ontogenesis pilot also lives here: pick one organism, author 4-6 stage Mns (zygote/blastula/gastrula/organogenesis/adult/senescent), author per-stage population_contexts.

Entity arrangement

Full entity-system chain. Closed Phase 1.

Chain levels

physical-hardware → digital-computing → hardware-to-computing-bridge → computing-to-entity-bridge → entity-system → entity-to-app-bridge → application-architecture → app-to-ecosystem-bridge → digital-ecosystem → digital-context

Data files

• Arrangement: data/arrangements/entity.v1.json
• Topology: data/topologies/entity.v1.json
• Domains (6): physical-hardware, digital-computing, entity-system, application-architecture, digital-ecosystem, digital-context
• Bridges (4): hardware-to-computing, computing-to-entity, entity-to-app, app-to-ecosystem
• Manifestations (7): git, github, postgres, nostr, instagram (point Mns) + content-addressed-data-systems, interactive-web-platforms (population Mns)
• Walks (11): physical-hardware-min-to-max, digital-computing-min-to-max, hardware-to-computing-crossing, entity-system-min-to-max, entity-system-min-to-full, entity-to-app-crossing, application-architecture-min-to-max, app-to-ecosystem-bridge-..., digital-ecosystem-min-to-max, environment-context-min-to-max
• Trajectories (2): git-evolution.v1.json (5 epochs 2005-04 → 2024-12), github-evolution.v1.json (4 epochs 2008-04 → 2024-01) — together form the Git+GitHub bundle constituent trajectories
• Rates: data/rates/entity-system-placeholder.v1.json
• Population contexts (4): git-2005-initial-to-vcs, git-2005jul-to-2006-distributed, git-2006-to-2010-github-tipping, git-2010-to-2024-standardization

Key output figures

• Per-domain hexagon/hasse/landscape/paths/ridge: entity-system-*, application-architecture-*, digital-ecosystem-*, physical-hardware-*, digital-computing-*
• Bridge crossings: hardware-to-computing-crossing-*, entity-to-app-crossing-*, app-to-ecosystem-bridge-*
• entity-topology.png

Status notes

• Phase 1 closed. All 5 Mn corridors verify against canonical filter percentages.
• Technogenesis pilot complete: 4 population_contexts + 1 focused walk (git-ecosystem-2006-to-2024) + 2 plot scripts (plot_technogenesis_sensitivity, plot_technogenesis_comprehensive). Models exercised: population_sampled, autocatalytic_threshold, frozen_accident. Two new placeholders registered for future use: network_effects, competitive_displacement. Pilot validated cross-domain registry dispatch without code changes (only frozen_accident.reference_dependency_count parameterized).
• Question discipline: per analytical-discipline-and-question-audit.md, the technogenesis pilot is FOCUSED on the meaningful probabilistic segment (2006 → 2024 ecosystem dominance + standardization). Substrate-implementation epochs (2005 → 2006) are deterministic-given-conditions and deliberately excluded from the comprehensive chart's walk.

Probability models

Registry at compute/lib/probability_models.py. Per-stage selection via population_context.v1.transitions[].probability_model; fallback chain ensures the analysis always produces a value.

Cross-domain caveats baked into implemented models:

• chemistry_kinetic defaults are Hadean-vent-tuned (T=350K, FeS/NiS/Pt catalysts). Override temperature_K and provide custom catalyst factors via parameters when reusing for ontogenesis (e.g. fertilization at T=310K).
• frozen_accident reference scale is now parameterizable: model_parameters.reference_dependency_count (default 20). Use 20 for genetic-code analogs, 8 for HTTP-style standards, etc. — value compares to code_dependency_count for the factor.
• autocatalytic_threshold uses fidelity_threshold + baseline_fidelity parameter names but the math is general. For technogenesis tipping points, set them to adoption_threshold semantics (e.g. 0.16 Rogers tipping, baseline current adoption fraction).

Cross-arrangement workflow patterns

Authoring a new manifestation

1. Identify which chain levels are relevant (cross-domain unified position — see framework-synthesis-).
2. Write data/manifestations/<name>.v1.json against manifestation.v1.schema.json.
3. Validate: python3 compute/scripts/validate_data.py.
4. Coherence check: python3 compute/scripts/check_coherence.py --mn <name>.

Authoring a new population_context

1. Pick the manifestation (or walk, or arrangement) the context describes.
2. Identify the dominant probability mechanism per transition (see per-stage-mechanisms.md tables).
3. Write data/population_contexts/<name>.v1.json declaring transitions[].probability_model + transitions[].model_parameters. Even if model isn't implemented, declare it — fallback chain handles gaps.
4. Validate: python3 compute/scripts/validate_data.py.

Authoring a new arrangement (rare)

1. Author topology + arrangement files first.
2. Then domains, bridges (extracted from canonical analysis docs).
3. Validate, then run make all-outputs ARRANGEMENT=<name>.

Switching between arrangements (operational)

1. Read this doc's "Arrangements at a glance" section.
2. Read the arrangement-specific section to find data files + scripts + how-to-rerun.
3. Read project-tracking.md for the latest phase status + decisions.
4. Read the most recent session-*-closing-summary.md for in-flight context.

Open organization questions

1. Subdivide data/manifestations/ and data/walks/ by arrangement? Currently flat with naming convention. Subdirs would clean browsing but break all manifestation_ref paths in JSON + script glob patterns. Decision: keep flat; this doc is the index instead.
2. Subdivide output/figures/ by arrangement? Same trade-off. Decision: keep flat; rely on naming prefix.
3. Per-arrangement plot script directories? Currently scripts are flat under compute/scripts/. Decision: keep flat; arrangement-specific scripts are explicitly named (plot_abiogenesis_*).

If these become painful as more arrangements come online, revisit.