An interactive computational laboratory for structured instability, adaptation, state transitions, information dynamics, and resilience.
The source framework defines PCC as Pressure–Chaos–Control: pressure drives or amplifies change, chaos introduces disruption or unpredictability, and control stabilizes or constrains. EBID is Entropy-Based Instability Dynamics, an observational layer relating entropy deficit and instability. Those definitions are preserved; the cognitive modules are exploratory extensions, not established consequences of PCC/EBID.
Educational and research software only. It is not a clinical assessment, diagnostic device, treatment tool, or validated model of universal psychological mechanisms.
- Cognitive Flexibility Sandbox — deterministic task-switching comparison
- Thought-Loop & State-Transition Model — weighted named-state trajectories
- Cellular-Automata Resilience Lab — Wolfram rules, XOR maps, normalized Hamming distance
- Information-Dynamics Explorer — entropy, signal/noise, and bottlenecks
- Adaptive-Agent Playground — transparent policy baselines
- Personal Data Import — browser-local CSV preview with synthetic data
# web
pnpm install
pnpm dev
# API (second terminal)
python -m venv .venv
.venv/bin/pip install -e '.[dev]'
.venv/bin/uvicorn apps.api.main:app --reload --port 8000Open http://localhost:3000; API documentation is at http://localhost:8000/docs.
pytest
pnpm build
node --test tests/rendered-html.test.mjsOr run both services with docker compose up --build.
app/ React/TypeScript studio UI
apps/api/ FastAPI routes and validation
packages/simulations/ deterministic scientific engines
data/synthetic/ non-personal demonstration data
data/schemas/ import contracts
docs/ audit, architecture, limitations, roadmap
tests/python/ deterministic metric and engine tests
The original /Volumes/External/pcc repository was audited read-only and was not modified. See docs/AUDIT.md for provenance and findings.
The amber cells show exact divergence after a single bit flip at t=8. Regenerate it with PYTHONPATH=. python scripts/generate_sample.py.
Every stochastic demo uses a visible fixed seed. API inputs are validated. The UI runs locally without personal-data upload. Results are simulation outputs, not empirical findings.
No license was present in the supplied source repository. Choose an open-source license before public distribution; Apache-2.0 or MIT are reasonable candidates, but this project does not presume ownership rights over archived manuscripts or third-party PDFs.
Entropy Studio now separates experiment definitions from immutable run records. The Simulation Bench can execute the registered E-007 cyclic-replicator toy protocol, capture measurement series, compute registered observables, preserve a reproducibility manifest, and derive result-backed Evidence Graph relations. See EXPERIMENT-RUNNER.md.
The 08 · Experiment Orchestrator workspace executes registered research campaigns from a frozen seed/parameter grid through statistics, figures, evidence summaries, manuscript refresh, and reproducibility packaging. Validated local engines run directly; external engines remain explicitly import-gated. See EXPERIMENT-ORCHESTRATOR.md.
The default landing page is now a front-facing Mission Control workspace. It derives presentation data from the active research registry and currently loaded immutable run records rather than calling a separate dashboard API.
The dashboard includes:
- time-aware workspace greeting and next-action focus
- research-health metrics with direct navigation
- campaign progress based on expected seed/parameter combinations
- actionable attention items for blocked campaigns, missing figures, challenged runs, and incomplete Results sections
- recent session and registry activity
- evidence balance from loaded run conclusions
- publication-pipeline readiness
The composition layer lives in app/lib/missionControl.ts; responsive presentation is implemented in app/studio.tsx and app/globals.css.
pnpm install
pnpm typecheck
pnpm test
pnpm build
pnpm devOpen http://localhost:3000. The repository intentionally uses pnpm as its single package manager.
Entropy Studio now treats rule spaces and observers as first-class research objects. The initial program is documented in RULIOLOGY.md and begins with the complete 256-rule elementary cellular-automaton family before extending the same contracts to Boids and other engines.
Key additions:
app/models/ruliology.ts- rule-space, observer, profile, equivalence, and transition types.app/data/ruleSpaces.ts- registered ECA, Boids, and replicator rule spaces plus frozen observer definitions.app/lib/elementaryCA.ts- deterministic ECA calibration engine with matched single-cell perturbations.app/lib/ruleSpaceExplorer.ts- finite rule enumeration and normalized rule distance.app/lib/rulialAnalysis.ts- observable-profile distance, equivalence components, and sensitivity detection.app/lib/crossSystemAnalysis.ts- shared-observable summaries across rule spaces.schemas/rule-space.schema.json,schemas/rulial-profile.schema.json, andschemas/rulial-campaign.schema.json- versioned rulial artifact contracts.examples/ruliology/eca-atlas.json- non-frozen RUL-001 campaign template.04 · Rulial Atlas- Studio workspace for browsing rule spaces, observers, and a deterministic ECA plumbing preview.
The scientific boundary is explicit: the ECA preview verifies infrastructure only. Claims about classes, phase boundaries, observer dependence, PCC regimes, or cross-system universality require frozen campaigns and held-out validation.
The first complete Ruliology benchmark can now be regenerated locally with npm run ruliology:eca. It enumerates all 256 elementary cellular automata over the frozen four-seed ensemble and writes the campaign report, rule profiles, one-bit rule-neighborhood transitions, and compact Rulial Atlas payload under data/ruliology/eca-atlas/.
The browser Rulial Atlas consumes the committed compact atlas payload, so the 256-rule population and highest one-bit sensitivity edges can be inspected without rerunning the campaign on every page load. npm run ruliology:eca:quick preserves a seven-rule smoke-test path.
Run npm run ruliology:eca:validate after generating or loading the frozen RUL-001 atlas. The validator generates a disjoint four-seed holdout ensemble, deterministic 95% bootstrap intervals for every rule-feature, rule-distance and one-bit-edge rank-stability statistics, replicated top-tail sensitivity edges, and conservative complete-link candidate EBID equivalence classes. Results are committed under data/ruliology/eca-validation/ and surfaced in 04 · Rulial Atlas.
The external-classification comparison is intentionally left unrun until a provenance-bearing label table is frozen. This prevents Wolfram or other CA class labels from influencing the unsupervised EBID metrics and candidate classes.
Entropy Studio can now test observer dependence without re-simulating the underlying system. npm run ruliology:eca:observers simulates a 16-seed ECA ensemble once, reprojects the same stored run summaries through four frozen observer definitions, calibrates each observer's equivalence tolerance from disjoint split halves, and writes candidate quotient classes plus cross-observer comparisons under data/ruliology/eca-observer-dependence/.
After generating the fixed-trajectory RUL-004 observer dataset, enumerate all 31 non-empty subsets of the five-feature ECA observer basis and compare observer distance to induced rulial geometry/quotient distance:
npm run ruliology:eca:observer-geometryThe analysis reuses the existing 4,096 source runs and writes a 31-node / 465-pair observer lattice analysis to data/ruliology/eca-observer-geometry/.
Run the frozen multidimensional Boids benchmark with:
npm run ruliology:boidsIt writes data/ruliology/boids-rulial/ with the discovery profiles, local candidate boundaries, held-out endpoint tests, adaptive boundary probes, a compact Studio summary, and a provenance-oriented full report. RUL-006 is intentionally a stress test rather than a claimed universal flocking phase diagram.
Run the first frozen ECA-Boids structural comparison with:
npm run ruliology:cross-substrateThe command reuses the committed ECA discovery/holdout artifacts and the frozen RUL-006 Boids discovery design, then adds complete held-out Boids coverage at all 32 rule coordinates (64 new simulations). It writes data/ruliology/cross-substrate/ with a versioned report, compact summary, substrate metrics, held-out Boids profiles, and checksum.
RUL-007 compares dimensionless structural summaries rather than raw cross-system observables. Its five criteria are frozen in the script and failures are retained. The current benchmark passes 2/5 criteria across both substrates, so it is a mixed cross-substrate result rather than evidence of universality.
Entropy Studio now includes a topology-blocked stochastic network substrate. A deterministic 24-point Latin hypercube samples threshold, neighbor coupling, node memory, and stochastic temperature. Each local rule is evaluated on fixed ring, small-world, and matched-degree Erdos-Renyi graphs, with three discovery and two disjoint validation seeds.
Run:
npm run ruliology:networkThe benchmark writes entropy-rulial-network/1.0.0 artifacts to data/ruliology/network-rulial/. Topology is retained as an explicit experimental block rather than hidden inside the local rule metric. The committed result shows strong held-out stability, but no cross-substrate universality claim is made until the pre-existing RUL-007 contract is applied unchanged to all three substrates.
The Ruliology program now includes a versioned three-way comparison across elementary cellular automata, stochastic Boids, and topology-blocked stochastic networks. RUL-009 reuses the five structural criteria frozen in RUL-007 without changing any threshold after seeing RUL-008. It performs no new simulations.
Current result: ECA passes 5/5 criteria, Network passes 5/5, and Boids passes 2/5. Two criteria are retained across all three substrates: positive global rule/observable association and a heterogeneous local-sensitivity tail. The stronger geometry and local-boundary replication criteria remain challenged by Boids.
npm run ruliology:three-substrateSee data/ruliology/three-substrate/ and schemas/rulial-three-substrate.schema.json.
RUL-010 keeps the RUL-009 benchmark frozen and asks why Boids was the replication outlier. It reuses the same 32 RUL-006 rule coordinates with two new disjoint seed pools, a nested 1/2/4-seed averaging ladder, a diagnostic per-step-noise-suppressed arm, and four observer projections computed from identical runs.
npm run ruliology:boids:resolutionThe committed diagnostic indicates that averaging improves the full Boids geometry only modestly, while observer choice matters strongly: state/structure and order/entropy projections are much more stable than the sparse transition/dwell projection. Suppressing per-step Gaussian forcing does not repair the one-seed geometry, so the current evidence does not support blaming the RUL-009 gap primarily on that forcing term. See data/ruliology/boids-resolution/ and schemas/rulial-boids-resolution.schema.json.
npm run ruliology:boids:observer-validation runs a new 40-point Boids rule-space sample under two new four-seed pools and projects the identical 320 trajectories through three preregistered observers. The primary test asks whether the RUL-010 state/structure observer beats the frozen six-feature observer by at least 0.05 in both full-geometry and local-edge split-half Spearman stability. The committed result passes both margins on unseen rule coordinates; earlier RUL-009/RUL-010 benchmarks remain unchanged. Outputs are frozen under data/ruliology/boids-observer-validation/.
RUL-012 asks whether a simple coordinate-level reliability rule generalizes across ECA, Boids, and Network: do observable coordinates that move more for the same rule across independent pools induce less reproducible rule-space geometry? The analysis adds zero new simulations, uses 17 registered coordinates from frozen prior profile pairs, normalizes same-rule displacement by discovery-pool robust range, and applies a 5,000-draw substrate-stratified permutation test.
The frozen primary hypothesis is challenged. The pooled association is approximately rho = 0.004 with stratified permutation p = 0.792, rather than the preregistered rho <= -0.50. This preserves the narrower prospective RUL-011 result while rejecting the stronger idea that one scalar same-rule-shift score is a general cross-substrate predictor of observer geometry conditioning. The next observer theory should therefore model feature semantics, degeneracy, and interactions among coordinates rather than relying on a single noise magnitude.
Run with npm run ruliology:observer-conditioning. Outputs live under data/ruliology/observer-conditioning/.
RUL-013 follows the challenged RUL-012 scalar-shift hypothesis without adding simulations. It reuses the same 17 ECA, Boids, and Network coordinate profile pairs, but separates between-rule discrimination from independent-pool uncertainty. For two independent aggregate pools A and B, Var(A-B)/2 estimates per-pool error variance; midpoint variance is then corrected to estimate between-rule variance. Their ratio defines an explicitly labeled ICC-like aggregate reliability rather than a classical per-trial ICC.
The frozen primary test is supported: ICC-like reliability is strongly positively associated with single-feature geometry stability (rho ≈ 0.882) with a 5,000-draw within-substrate permutation p ≈ 0.0010, exceeding the preregistered rho >= 0.70 threshold. A separately frozen robust signal-to-uncertainty ratio also tracks geometry stability (rho ≈ 0.833, p ≈ 0.0134). The result is not simply a rebranding of RUL-012: raw same-rule shift alone had essentially no pooled association, whereas signal relative to uncertainty is strongly predictive.
Degeneracy is retained as a secondary diagnostic. The pooled degeneracy association is weaker (rho ≈ -0.198), although the Boids transition-rate and metastable-dwell coordinates remain conspicuous low-reliability, low-support cases. This motivates a prospective observer-selection experiment rather than retroactively changing any frozen RUL-006 through RUL-012 observer.
Run with npm run ruliology:observer-information. Outputs live under data/ruliology/observer-information/.
RUL-014 tests whether the information-conditioning result from RUL-013 can be turned into a prospective measurement-design rule. Before generating any RUL-014 outcomes, the selector is frozen to include Boids coordinates whose RUL-013 ICC-like reliability is at least 0.80. That rule selects spatial entropy, speed variance, polarization, and heading entropy, while rejecting transition rate and metastable dwell.
The experiment then generates a completely new 48-point Boids Latin-hypercube design and two disjoint four-seed pools, for 384 new simulations. Three observers are projected from those identical trajectories: the full six-feature core, the RUL-013-selected four-feature observer, and the two rejected coordinates as a negative-control observer. The primary success margin is fixed at +0.05 for both complete-geometry and local-edge split-half Spearman stability.
The selected observer improves in the predicted direction, but the frozen primary criterion is challenged: geometry improves by about +0.045 and local-edge stability by about +0.016, both below the preregistered +0.05 margins. The secondary top-10% boundary Jaccard improves strongly (+0.283), and the rejected-coordinate control is near-zero/negative in geometry stability. The correct interpretation is therefore not that RUL-013 failed, nor that the selector is validated, but that the simple hard-threshold selection rule has directional predictive value without meeting the stronger prospective effect-size criterion on this sample.
Run with npm run ruliology:observer-selection. Outputs live under data/ruliology/prospective-observer-selection/.
RUL-015 tests the natural follow-up to RUL-014: replace binary keep/drop selection with a continuous metric fixed entirely from prior RUL-013 information. Before any new outcome is generated, each Boids coordinate receives weight proportional to ICC-like reliability × log(1 + signal-to-uncertainty) × (1 - degeneracy). The experiment then generates a new 56-point Latin-hypercube rule design and two disjoint four-seed pools, for 448 new simulations, and compares equal full-core, RUL-013 hard selection, reliability-only weighting, and the information-conditioned continuous metric on identical trajectories.
The preregistered primary criterion is challenged. The information-weighted metric improves complete geometry by only about +0.001 and local-edge stability by about +0.006 relative to equal weighting, below the frozen +0.03 margins. Its top-10% local-boundary overlap is also lower. In contrast, the hard four-feature RUL-013 selection again gives the strongest complete-geometry stability (rho ≈ 0.939) and top-boundary Jaccard (≈ 0.684) on this unseen design. This does not invalidate RUL-013; it challenges this particular smooth weighting equation and suggests that severely degenerate coordinates may need exclusion rather than merely small positive weights.
Run with npm run ruliology:observer-weighting. Outputs live under data/ruliology/information-weighted-observer/.
RUL-016 keeps the RUL-015 population frozen and asks a different question: are observer coordinates acting independently, or does the geometry depend on feature combinations? It introduces zero new unique simulations. The deterministic RUL-015 56-point design and two four-seed pools are replayed, then every one of the 2^6 - 1 = 63 non-empty subsets of the six-coordinate Boids observer is evaluated on identical trajectories.
For each subset, the analysis records complete-geometry stability, local-edge stability, and top-10% local-boundary overlap. It then computes leave-one-out effects, exact Shapley contributions with an explicit v(empty)=0 bookkeeping baseline, and exact pairwise Shapley interaction indices. These are finite observer-set decompositions, not causal claims about the physical observables.
The diagnostic finds clear non-additivity. The strongest complete-geometry interaction is polarization × speed variance (I ≈ -0.298), far above the frozen |I| >= 0.05 diagnostic magnitude. Both regime coordinates (transition rate and metastable dwell) have negative complete-geometry Shapley contributions (about -0.041 each), and removing either from the full six-feature observer improves both complete and local stability slightly. However, the largest interaction does not involve those regime coordinates, so the result is richer than “two bad features”: observer conditioning depends materially on combinations among otherwise useful structural coordinates.
Run with npm run ruliology:observer-ablation. Outputs live under data/ruliology/observer-ablation/.
RUL-017 turns the diagnostic RUL-016 subset result into a prospective test. Before reading new outcomes, it freezes the three-feature Boids observer polarization + spatial entropy + speed variance, compares it against the established four-feature RUL-013 hard-selection observer and the six-feature full-core baseline, and generates a new deterministic 40-point Latin-hypercube rule-space design with two disjoint four-seed pools (320 new simulations). Pool-A feature scaling is held fixed for pool B.
The primary margins are frozen at +0.01 for both complete-geometry and local-edge split-half Spearman stability relative to the four-feature observer. The three-feature candidate passes both: complete geometry improves by about +0.082 and local geometry by about +0.012. The secondary top-10% boundary-overlap margin is challenged: Jaccard is lower than the four-feature comparator. This supports predictive value for the RUL-016 interaction-informed geometry choice without claiming universal observer optimality.
Run with npm run ruliology:observer-interaction-validation. Outputs live under data/ruliology/interaction-informed-observer-validation/.
RUL-018 reuses the frozen RUL-017 Boids population and introduces zero new unique simulations. It evaluates all 63 non-empty subsets of the six registered Boids observer coordinates under three fixed objectives: complete rule-space geometry stability, local-edge geometry stability, and top-10% local-boundary recovery. The objectives are intentionally kept separate rather than collapsed into a post-hoc scalar score.
The optimum depends on the scientific target. Global geometry is maximized by spatial entropy + speed variance (rho = 0.914), while local geometry is maximized by speed variance + transition rate (rho = 0.845). Boundary recovery is discrete and has ten co-optimal subsets at Jaccard = 0.571; the smallest representative is polarization alone, and the local-geometry optimum is also boundary co-optimal. Global- and local-objective rankings are strongly related (rho = 0.901), but boundary-recovery rankings are nearly orthogonal to both (rho = 0.079 and 0.017).
This is a diagnostic result, not a prospective claim that these subsets are universally optimal. It supports treating observer design as task-dependent: preserving the global rulial map, preserving local neighborhoods, and recovering extreme boundaries are distinct measurement goals. Run with npm run ruliology:observer-objectives.
RUL-019 extends the RUL-018 task-dependence analysis beyond Boids without adding any new simulations. It reuses the frozen ECA calibration/holdout atlas, the RUL-018 Boids population, and the RUL-008 Network discovery/holdout population. For each substrate, every non-empty subset of the substrate's native core observer basis is evaluated under the same three objectives: complete rule-space geometry stability, local-edge geometry stability, and top-10% local-boundary recovery. ECA contributes 31 observer subsets from five coordinates; Boids and Network contribute 63 subsets each from six coordinates.
The cross-substrate pattern is consistent: all three substrates show objective dependence, none has a single observer subset co-optimal for all three objectives, and boundary-recovery rankings are more decoupled from geometry rankings than global and local geometry are from each other. Global-vs-local observer ranking Spearman correlations are approximately 0.962 for ECA, 0.901 for Boids, and 0.806 for Network. By contrast, global-vs-boundary correlations are approximately 0.623, 0.079, and -0.040 respectively.
The representative optima remain substrate-specific: ECA global geometry favors Shannon entropy + compression while boundary recovery favors compression alone; Boids global geometry favors spatial entropy + speed variance while local geometry favors speed variance + transition rate; Network global geometry favors order + switch rate + transition rate while boundary recovery has many co-optimal subsets. These identities are not treated as cross-substrate invariants. The recurring result is the higher-level structure: observer choice is task-dependent, geometry-preservation objectives are more closely related to each other, and extreme-boundary recovery behaves as a distinct measurement target.
Run with:
npm run ruliology:cross-substrate-objectivesOutputs are written to data/ruliology/cross-substrate-objectives/.
Entropy Studio now includes a first artificial-life ecology in which agents inherit and mutate a four-dimensional behavioral rule vector. RUL-020 compares stable mutable ecology, resource-scarcity ecology with mutation, and a matched scarcity control with mutation disabled across 12 frozen seeds (36 runs total). The committed pilot tracks population-level rule-centroid displacement, cumulative rule-space path length, rule diversity, founder-lineage diversity, and demographic recovery. Run it with:
npm run ruliology:alifeThe committed pilot passes all four frozen criteria, but the interpretation is intentionally narrow: it demonstrates reproducible motion of a rule distribution inside an engineered ALife model, not biological evolution, universal adaptation, or a PCC derivation.
RUL-021 extends the mutable-rule ALife substrate with a 180-step burn-in and a depth-matched demographic control. For each of 12 new frozen seeds, resource scarcity is compared with a stable-resource condition that retains mutation but receives a one-time rule-blind random cull matched to the scarcity run's realized bottleneck depth. The primary outcome is post-shock rule-centroid displacement from the pre-shock centroid, together with directional reproducibility and bottleneck-match diagnostics.
npm run ruliology:alife:selection-controlThe committed result is intentionally preserved as challenged (3/5 frozen criteria): scarcity does not exceed the neutral bottleneck in median post-shock displacement and wins in only 5/12 paired seeds, although its motion is more directionally reproducible. This narrows the interpretation of RUL-020 by showing that demographic bottlenecking alone can explain much of the observed centroid displacement.
RUL-022 reuses the frozen RUL-021 post-shock rule-motion vectors and adds 96 separate finite-difference ecological probes. Around each seed's pre-shock rule centroid, the four normalized rule coordinates are perturbed by ±0.06 and evaluated with homogeneous, no-mutation populations under immediate scarcity. The frozen performance target is 120-step time-averaged population persistence. Scarcity motion is positively aligned with this local gradient (median cosine 0.431; 9/12 positive seeds), but it does not exceed the matched neutral bottleneck alignment: median paired scarcity-minus-neutral cosine is -0.059. Three of four frozen criteria pass, so the stronger selection-specific directionality hypothesis remains challenged. Run with npm run ruliology:alife:fitness-gradient.
RUL-023 reuses the frozen RUL-021 scarcity/neutral post-shock motion vectors and the 96 committed RUL-022 scarcity finite-difference probes, then adds 96 matched stable-resource probes at the same seed-specific pre-shock rule centroids. The preregistered contrastive direction is grad(F_scarcity) - grad(F_stable) using the same ±0.06 normalized-rule step and 120-step homogeneous no-mutation population-persistence target. The contrastive gradient is identifiable in all 12 seeds, but scarcity alignment is weak (median cosine 0.038), positive in only 6/12 seeds, and does not exceed neutral bottleneck alignment (median paired scarcity-minus-neutral cosine -0.079). Only 1/4 frozen criteria passes, so the scarcity-specific contrastive-gradient hypothesis is challenged. Run with npm run ruliology:alife:contrastive-gradient.
RUL-024 reuses the frozen RUL-021 scarcity and neutral post-shock rule-motion vectors, but replaces homogeneous fitness probes with context-dependent tagged-mutant invasion probes. For each of the 12 matched seeds, the exact mixed pre-shock resident population is regenerated by the frozen burn-in. A rule-blind 10% sample is assigned a mutant rule at ±0.06 normalized units along each of four rule dimensions, post-introduction mutation is disabled, and mutant frequency change estimates the local invasion score. This adds 96 new probe simulations and zero new motion simulations. The result is challenged: median scarcity alignment is 0.000, only 4/12 scarcity trajectories have positive alignment, only 7/12 gradients are identifiable, and the paired scarcity-minus-neutral median alignment is 0.000. Run with npm run ruliology:alife:invasion-gradient.
RUL-025 stops treating the ALife population as a single point in rule space. It introduces no new seed/parameter conditions; instead it deterministically replays the 12 frozen RUL-021 scarcity runs and 12 matched neutral-bottleneck runs with non-invasive founder-lineage logging. Replay checks require exact post-shock motion vectors and matching final populations before the lineage analysis is accepted.
For founder lineage (\ell), let (p_{\ell,0},p_{\ell,1}) be pre-shock/final abundance fractions and (\mu_{\ell,0},\mu_{\ell,1}) the corresponding normalized rule centroids. The population centroid motion is decomposed exactly as
[ \Delta\mu = W + B, ]
with
[ W=\sum_\ell \tfrac12(p_{\ell,0}+p_{\ell,1})(\mu_{\ell,1}-\mu_{\ell,0}), ]
and
[ B=\sum_\ell (p_{\ell,1}-p_{\ell,0})\tfrac12(\mu_{\ell,0}+\mu_{\ell,1}). ]
The committed result is mixed/challenged (4/5 frozen criteria pass). The decomposition reconstructs the population motion to numerical precision, and scarcity motion is dominated by lineage reweighting: median reweighting norm share is 0.788, versus 0.212 for within-lineage change. Surviving scarcity lineages also move heterogeneously (median pairwise directional cosine 0.190) with at least three moving survivor lineages in all 12 seeds. However, the matched neutral bottleneck is itself strongly reweighting-dominated (0.764), and the paired scarcity-minus-neutral difference is only +0.038, below the frozen +0.10 selection-specific margin.
The result therefore supports a structural claim—population-centroid motion can hide lineage-level heterogeneity and is largely generated by changing lineage weights—but does not support the stronger claim that scarcity uniquely amplifies this mechanism relative to the depth-matched neutral bottleneck. Founder labels are bookkeeping identities, and the decomposition is descriptive rather than a biological Price-equation claim.
Run with:
npm run ruliology:alife:lineage-motionOutputs are written to data/ruliology/alife-lineage-motion/.
RUL-026 extends the RUL-025 endpoint decomposition to every adjacent recorded post-shock lineage snapshot while preserving the exact frozen RUL-021 scarcity and neutral-bottleneck seed/parameter conditions. No new unique simulation conditions are introduced; 24 deterministic replays add only observational lineage logging at the existing five-step cadence.
Each interval is decomposed exactly into within-lineage rule-centroid motion and between-lineage abundance reweighting. The primary time-resolved observables are cumulative lineage total-variation turnover, cumulative reweighting path share, reweighting path tortuosity, and dominant-lineage switches. All six frozen criteria pass: scarcity has median cumulative reweighting share 0.839, median reweighting tortuosity 4.12, 11/12 runs switch dominant lineage, and paired scarcity-minus-neutral cumulative turnover is +0.201. These are transport descriptors, not proof of adaptive selection.
Run with npm run ruliology:alife:lineage-transport. Outputs are written to data/ruliology/alife-lineage-transport/.
RUL-027 takes the time-resolved founder-lineage transport from RUL-026 and projects it onto a fixed four-dimensional grid with four equal-width bins per normalized rule coordinate (4^4 = 256 possible cells). A lineage segment contributes J_l,t = 0.5(p0+p1)(mu1-mu0) to its midpoint cell and a directed source-to-target cell edge; absolute abundance change is tracked separately as local turnover mass. The grid, five-step cadence, seed halves, and seven pilot criteria are frozen before the RUL-027 outcomes. No new seed/parameter conditions are added: 24 deterministic RUL-021 replays are used only to recover lineage-resolved segment geometry.
Scarcity occupies 29 flux-bearing cells and 78 directed coarse edges. The top 20% of occupied cells carry 0.751 of total advective flux, flux-weighted local directional persistence is 0.282, and the top-flux cells recur across the two frozen six-seed halves with Jaccard 0.667. Scarcity and matched neutral bottlenecks differ in their cell-flux profile (JSD = 0.144 bits), abundance-turnover profile (JSD = 0.055 bits), and directed edge-flux profile (JSD = 0.230 bits). All seven frozen pilot criteria pass, with exact interval reconstruction against RUL-026 and exact RUL-021 replay.
The result supports a coarse transport-network description in this engineered ALife model: lineage rule motion is not uniformly distributed through rule space, and scarcity changes where transport and abundance turnover are allocated. These cells and edges are observer-dependent summaries, not literal physical currents, and the result does not by itself establish adaptive selection.
RUL-028 tests whether the RUL-027 coarse-flux result is an artifact of choosing exactly four bins per normalized rule coordinate. It performs zero new simulations and reuses the exact committed RUL-027 lineage segment coordinates. Before inspecting the new multiscale outcomes, the resolution family was frozen to b in {3,4,5,6} and each original qualitative RUL-027 threshold was declared robust only if it passed at least three of the four resolutions. The committed b=4 metrics must also reproduce RUL-027 exactly.
The result is supported under that contract. Channel concentration passes at 4/4 scales, seed-half top-channel recurrence at 4/4, flux-weighted directional persistence at 4/4, and scarcity-versus-neutral cell-flux Jensen-Shannon divergence at 4/4. Abundance-turnover profile divergence passes at 3/4: the coarsest b=3 grid gives 0.028 bits, while b=4,5,6 give 0.055, 0.088, and 0.117 bits. Channel concentration ranges from 0.578 to 0.751; recurrence from 0.286 to 1.000; directional persistence from 0.253 to 0.423; and cell-flux divergence from 0.070 to 0.233 bits.
All six frozen RUL-028 criteria pass. The result reduces concern that RUL-027's qualitative transport-channel picture is unique to a single 4^4 grid, but it does not establish a grid-free continuum current, scale invariance over arbitrary resolutions, adaptive selection, or observer independence. The analysis is intentionally a small-family multiscale robustness test.
RUL-029 tests whether the RUL-027 coarse-flux result depends on the original five-step lineage sampling cadence. It performs zero new simulations. Per-lineage snapshot states are reconstructed from the exact committed RUL-027 segments, the four-bin-per-coordinate spatial observer is held fixed, and transport segments are rebuilt at the frozen cadence family dt in {5,10,20,40} while always retaining the 179->180 intervention boundary and the final endpoint. The original RUL-027 thresholds are unchanged; a qualitative finding is declared temporally robust only when it clears its threshold at at least three of four cadences.
All six frozen criteria pass. Channel concentration, seed-half recurrence, directional persistence, and scarcity-vs-neutral cell-flux divergence pass at all four cadences; abundance-turnover divergence passes at three of four, missing only at dt=20 (0.0492 bits versus the frozen 0.05 threshold). The five-step projection reproduces RUL-027 exactly. Total measured scarcity advective flux decreases from about 0.963 at dt=5 to 0.753 at dt=40, which is consistent with coarse temporal sampling compressing reversals and path length rather than preserving every transient. The result supports robustness over this declared cadence family, not continuum-time invariance, arbitrary temporal renormalization, adaptive selection, or observer independence.