LawSynth — Expansion v2 (toward a ~10,000-file platform)

The original manifest (LawSynth_3161_Repository_Manifest.md) planned P0–P5 at 3,161 files; the repo is now ~3,460 with P6–P10 built. This document plans the next expansion — the competitor-informed capabilities from competitive-analysis.md built out across *every surface* (spec → crate → tests → conformance → benchmark → SDK → CLI → Studio → service → docs → cookbook → example). Building each capability across all surfaces is exactly what multiplies file count from ~3.5k toward ~10k — the same discipline the original manifest used, applied to a bigger scope.

Contract (unchanged): every capability ships as a *boundary spec + conformance suite*, is deterministic and offline, and never regresses the local core. Files are added only when the milestone is reached; this is a plan, not a scaffold.

The multiplier: one capability, ten surfaces

Each engine capability (e.g. "weak-form discovery") lands as, roughly: spec/ (3) + crate src (8) + crate tests (6) + conformance cases (5) + benchmark cases (6) + SDK module+tests (4) + CLI command+tests (3) + Studio screen (3) + service endpoint+tests (3) + docs+cookbook+example (5) ≈ ~46 files. That is why ~30 new capabilities → ~1,400 files just for the engine, and the ecosystem/data/platform tiers carry the rest.

Milestones

v2-A — Engine breadth (close the discovery gaps) · ~1,600 files

New Rust crates (each: src/ 6–12, tests/ 5–10, benches/ 1–2) + a spec dir + conformance cases + benchmark family + SDK/CLI/Studio/service surfaces:

| Capability | New crate(s) | Spec | ~files | |---|---|---|---:| | Weak / integral-form discovery | lawsynth-weakform | specs/weak-form/ | 55 | | Koopman / DMD / EDMD | lawsynth-koopman | specs/koopman/ | 60 | | Implicit / rational nullspace | lawsynth-implicit | specs/implicit-dynamics/ | 55 | | Units-in-discovery (dim. pruning + Buckingham-π) | extend lawsynth-units, lawsynth-discovery | specs/dimensional-search/ | 45 | | Symmetry / separability reductions | lawsynth-reduce | specs/structural-reductions/ | 55 | | Optimizer breadth (FROLS/SSR/trapping/L0) | extend lawsynth-sparse | specs/optimizers/ | 50 | | Deterministic ensembling | lawsynth-ensemble | specs/ensembling/ | 45 | | Control inputs (SINDYc discovery) | extend lawsynth-discovery, -world | specs/control-discovery/ | 45 | | MDL / description-length objective | extend lawsynth-score | specs/model-selection/ | 35 | | Analytic-Jacobian symbolic diff + codegen | lawsynth-jacobian | specs/jacobian/ | 50 | | Richer e-graph rewrite system | extend lawsynth-egraph | specs/egraph-rules/ | 45 | | Structured/template priors | lawsynth-template | specs/template-priors/ | 45 | | Performance: SIMD eval + rayon sweep + O(n log n) Pareto | extend lawsynth-features,-score,-runner | specs/parallel-determinism/ | 55 |

v2-B — Ecosystem & adoption · ~1,900 files

| Workstream | Where | ~files | |---|---|---:| | sklearn adapter (LawSynthRegressor/Transformer/Dynamics) + tests + docs | python/lawsynth-sklearn/ | 60 | | SymPy / torch / jax differentiable export (trainable constants) | python/lawsynth/ + crates/lawsynth-report | 50 | | SRBench harness (Docker + sklearn method + result reporting) | benchmarks/srbench/ | 80 | | Feynman equation family (public formulas, our own generators) | benchmarks/feynman/ (~120 cases) | 380 | | Strogatz ODE family (dynamics — we win) | benchmarks/strogatz/ (~15 cases) | 90 | | SRSD / black-box families | benchmarks/srsd/, benchmarks/blackbox/ | 260 | | Warehouse connectors (Snowflake/BigQuery/Databricks) | python/lawsynth-connectors/ + plugins/ | 180 | | MLflow / W&B interop (export + plugin) | plugins/mlflow-export/, plugins/wandb-export/ | 90 | | Benchmark whitepaper + reproducibility artifacts | docs/benchmarks/, papers/ | 60 | | Plugin marketplace content (10 → 40 example plugins) | plugins/ | 420 | | Connector expansion (kafka/iceberg/delta/arrow depth + 10 new sources) | python/lawsynth-connectors/ | 230 |

v2-C — Platform, hosting & distributed · ~1,700 files

| Workstream | Where | ~files | |---|---|---:| | Distributed-but-reproducible sweep (island model, seeded, shared Pareto) | crates/lawsynth-sweep + services/scheduler | 90 | | Real OIDC/SSO providers behind the P10 seam (Okta/Auth0/Azure AD) | services/gateway/ | 70 | | Multi-tenant quota/metering/billing hooks + usage dashboards | services/api/, apps/console/ (new admin app) | 220 | | Hosted self-serve trial (browser Studio + sample datasets + onboarding) | apps/trial/ (new) | 180 | | Admin console app (tenants, members, audit, plugins, usage) | apps/console/ (new) | 260 | | Distributed worker fleet ops (autoscale, placement, drain) | services/scheduler,worker, deploy/ | 160 | | Collaboration depth (P6): real-time presence, comments UI, review UX | apps/studio/, packages/collab/ (new) | 190 | | Streaming platform (P7): long-lived stream runs, dashboards | services/, apps/studio/ | 150 | | Governance depth (P9): approval workflows UI, audit explorer, compliance exports | apps/console/, services/api/ | 180 | | Deployment depth (helm/terraform/k8s for the fleet + observability) | deploy/ | 200 |

v2-D — Domain packs & verticals · ~1,100 files

Vertical wedges where an auditable deterministic equation has compliance value. Each pack: curated templates + example datasets + recipes + tutorials + a Studio "lab" preset + a benchmark family.

| Pack | Where | ~files | |---|---|---:| | Pharma / systems-biology (PK/PD, enzyme kinetics, gene regulation) | domains/biology/ | 240 | | Energy / industrial (power systems, thermal, control loops) | domains/energy/ | 220 | | Quant finance (mean-reversion, volatility, regime-switching) | domains/finance/ | 220 | | Climate / earth (compartment models, oscillators) | domains/climate/ | 200 | | Epidemiology (SIR/SEIR families, interventions) | domains/epidemiology/ | 180 |

v2-E — Docs, education & credibility · ~700 files

Cookbook expansion (10 → 40 recipes), tutorials (3 → 20), a full API reference site build, a benchmark leaderboard page, video/notebook galleries, migration guides ("coming from PySINDy/PySR/gplearn"), and academic artifacts.

Rollup

| Milestone | Focus | ~files | |---|---|---:| | v2-A | Engine breadth | 1,600 | | v2-B | Ecosystem & adoption | 1,900 | | v2-C | Platform & distributed | 1,700 | | v2-D | Domain packs | 1,100 | | v2-E | Docs & credibility | 700 | | Subtotal (new) | | ~7,000 | | Current repo | | ~3,460 | | Target | | ~10,460 |

Sequencing (what to build first, and why)

1. Units-in-discovery + weak-form (v2-A) — highest algorithmic ROI, both

deterministic, both directly neutralize a competitor's headline advantage.

1. sklearn adapter + SRBench harness + Strogatz/Feynman benchmarks (v2-B) —

the credibility + adoption unlock; without a published benchmark, engine work is invisible to the market.

1. Koopman/DMD + implicit (v2-A) — remove the two most common "it can't even

do X" objections from the SciML/PySINDy crowd.

1. Distributed-but-reproducible sweep (v2-C) — scales discovery while keeping

determinism, our flagship differentiator.

1. One vertical pack (v2-D) — convert the platform into paying value.

Each item ships behind a boundary spec with a conformance suite, deterministic and offline, on at least the CLI + SDK, with docs + a cookbook recipe — the same contract that carried P0–P10.

Build order note: the immediate next commits implement the sequencing #1 items (units-in-discovery dimensional pruning, then weak-form) as real, tested crates — starting now.