Ophamin plugin catalog — open-source tools mapped to Kimera-SWM aspects¶
Status: deep-research synthesis, 2026-05-15. Generated from web searches across the major Kimera-SWM aspects. Each entry: what it does, license, which Ophamin wheel + Kimera aspect it serves, install command, and an honest note on when to pick it vs alternatives.
Reading order: §3 is the priority shortlist (the ~12 tools with the highest leverage given Ophamin's current state). §4–§16 are the per-aspect catalogs. §17 is the integration roadmap with sequenced PRs.
1. The matrix — Kimera aspect × Ophamin wheel¶
Each row is a Kimera aspect that needs measurement; each column is an Ophamin wheel that should host it.
| Kimera aspect ↓ \ Ophamin wheel → | seeing | measuring | comparing | instrumenting | auditing | reporting | interop |
|---|---|---|---|---|---|---|---|
| Substrate physics (thermo / quantum / SPDE) | qiskit, qutip, fenics, firedrake | scipy.stats, mpmath | drift on partition functions | jax+numba acceleration | — | matplotlib | — |
| Cognitive (Walker / Φ / IIT) | pyphi, geomip | pyphi, infomeasure, dit | drift on Φ across commits | scalene per-cycle | — | matplotlib | mlflow |
| Memory / SCAR topology | gudhi, ripser, giotto-tda | persistence diagrams | drift on β₀/β₁/β₂ | — | — | persim | — |
| Distributed / Archipel | py-crdt, automerge | conformal prediction (puncc) | reconciliation drift | — | — | — | — |
| Encoder / Rosetta primes | sentence-transformers, BGE | UMAP/PaCMAP, optimal transport (POT) | encoder snapshot drift | — | — | UMAP plots | — |
| Time / Cronos / KCCL | kuramoto, multiSyncPy | sktime, darts, statsforecast | Allan deviation drift | viztracer | — | — | — |
| Statistical pillars | — | mapie, puncc, river, infomeasure, dit | drift detection (frouros, evidently) | — | — | arviz | — |
| Wire format / determinism | hypothesis | property-based equivalence | semver/CRDT drift | — | semgrep custom rules | — | spdx-tools |
| Code substrate health | tree-sitter | radon, lizard | LOC/complexity drift | austin, viztracer | semgrep, pyright, prospector, fawltydeps, deptry | — | sarif (have) |
| Telemetry / observability | jaeger, tempo, loki | prometheus_client (have) | metric cardinality drift | otel (have) | promtool | grafana | otel-exporter |
| Causal / counterfactual | — | dowhy, econml, causalml, causalpy | causal-graph drift | — | — | dowhy plots | — |
| Supply chain | syft (SBOM gen) | osv-scanner | dependency drift | — | grype, scancode, ossf scorecard | — | cyclonedx (have), spdx-tools |
2. Ophamin's current plugin baseline (recap)¶
| Layer | Tools currently integrated |
|---|---|
| Statistical core | numpy, scipy, statsmodels, pandas, scikit-learn, mapie, river |
| Audit pillars | ruff, bandit, mypy, vulture, radon, pip-audit |
| Telemetry | prometheus_client, opentelemetry-api/sdk |
| Profile | psutil, py-spy, memray |
| Provenance | prov, lxml, rdflib, mlflow, dvc |
| Interop | SARIF, JUnit XML, MLflow, CycloneDX |
| Reporting | matplotlib, hydra-core |
| Datasets | h5py (TheWell) |
These cover ~30% of what Kimera needs measurement-wise. The catalog below is the other 70%.
3. Priority shortlist — top 12 to integrate next¶
Ranked by signal-per-LOC-of-integration against Kimera's current state.
| # | Tool | License | What it gives Ophamin | Ophamin wheel | Install |
|---|---|---|---|---|---|
| 1 | PyPhi | GPL-3 | Native Φ computation — Kimera's Φ is computed inline; PyPhi is the canonical reference implementation to cross-check against | measuring | pip install pyphi |
| 2 | giotto-tda | AGPL-3 / Comm. | TDA pipeline (Vietoris-Rips, persistence diagrams, kernels) for the manifold — Kimera has a TopologicalComplexityAnalyzer but no time-series-of-diagrams comparator |
measuring + comparing | pip install giotto-tda |
| 3 | POT (Python Optimal Transport) | MIT | Wasserstein / Earth Mover for repertoires + barycentre for averaging Φ-trajectories. Kimera replaced _emd_hamming with closed-form recently (commit 9c055d303); POT is the reference oracle for cross-checking |
measuring | pip install POT |
| 4 | Frouros | BSD-3 | Pure-purpose drift detection (no anomaly / outlier / etc. wraps). Plug into comparing/drift/ to ratchet thresholds per commit |
comparing | pip install frouros |
| 5 | DoWhy + EconML | MIT | Causal inference. Kimera has 222 feedback-loop pathways → ideal substrate for causal-graph identification (does Φ drift CAUSE walker M2 amplitude_death, or just correlate?) | measuring + comparing | pip install dowhy econml |
| 6 | scalene | Apache-2 | Line-level CPU + memory profile (per-cycle). py-spy + memray we have, but scalene fuses both with line precision | instrumenting | pip install scalene |
| 7 | deptry + fawltydeps | MIT | Find unused / undeclared deps in Kimera's pyproject vs actual imports. Direct extension of seeing/wiring/'s scope to dependency-level |
auditing | pip install deptry fawltydeps |
| 8 | Hypothesis | MPL-2 | Property-based testing — for Kimera's CRDT primitives (G-Set, SCAR-DAG, Echoform-chain) where invariants like idempotence / commutativity / associativity should hold by construction | auditing | pip install hypothesis |
| 9 | Semgrep | LGPL / Comm. | Custom-rule SAST that catches Kimera's no-fallback rule violations (try: ... except: pass) at the AST level — beyond what bandit catches |
auditing | pip install semgrep |
| 10 | infomeasure | MIT | Comprehensive entropy + mutual info + transfer entropy estimators (newer than dit/pyitlib, 2025 paper). Foundation for the Σ correlation pillar measuring Kimera's 222 feedback paths | measuring | pip install infomeasure |
| 11 | Syft + Grype | Apache-2 | SBOM generation + vuln scanning. Anchore-backed, complementary to pip-audit's narrower Python focus | auditing + interop | brew install syft grype |
| 12 | OSSF Scorecard | Apache-2 | Upstream-dependency health scoring — surfaces which of Kimera's pinned deps are abandoned / unmaintained | auditing | go install github.com/ossf/scorecard@latest |
These 12 are the "fix next" surface. Sections §4-§16 are the broader catalog.
4. IIT / Φ computation¶
Kimera-SWM computes Φ inline via its IIT30 implementation (CLAUDE.md §Family E reports mean Φ 0.621 ± 0.065 across 200 cycles). External tools provide cross-check + alternative parameterizations.
| Tool | Notes | License |
|---|---|---|
| PyPhi | Canonical IIT 4.0 reference — Albantakis L. et al. PLoS Computational Biology 2023. Slow but exhaustive. | GPL-3 |
| GeoMIP | 165-326× speedup over PyPhi, 98-100% partition-identification agreement; recasts MIP search as graph optimization on hypercube | (academic, status unclear) |
Decision: ship PyPhi as a cross-check oracle for Kimera's iit30_emd
output. Pre-registered claim shape: "Kimera's per-cycle Φ falls within
[PyPhi's Φ ± δ] on identical TPM input".
5. Topological data analysis (TDA)¶
For the Vietoris-Rips filtration, β₀/β₁/β₂ that Kimera's
TopologicalComplexityAnalyzer already computes (Q13 measured: β₀=1,
β₁≈322, β₂≈192) and the SCAR-network homology that VaultHomology
tracks.
| Tool | Strengths | License |
|---|---|---|
| GUDHI | C++ with Python interface; broadest simplicial-complex coverage (Rips, Witness, Alpha, Čech). Bottleneck distance, persistence | MIT (Python) |
| Ripser | Fastest persistent homology computation at scale; doesn't construct complex explicitly | MIT |
| giotto-tda | Scikit-learn-API wrapper around GUDHI + Ripser. Persistence-diagram vectorization + kernels for ML downstream | AGPL-3 / Commercial |
| scikit-tda | Umbrella package — ripser, persim, kepler-mapper | MIT |
Decision: giotto-tda → seeing/topology/, persistence-diagram-drift
scenario in comparing/. Track β₀/β₁/β₂ time-series across Kimera commits.
6. Information theory (entropy, mutual info, transfer entropy)¶
Kimera's seven entropy measures (CLAUDE.md §Semantic Entropy) need an external reference for cross-checking. Also: Σ pillar (cross-stratum correlation between cognitive Φ and telemetry latency) needs proper mutual-information estimators, not Pearson r.
| Tool | Strengths | License |
|---|---|---|
| infomeasure | Newest (2025 paper, Scientific Reports); discrete + continuous; entropy / MI / transfer entropy / divergences | MIT |
| NPEET | Greg Ver Steeg's non-parametric Kraskov-Stögbauer-Grassberger MI estimator — the academic reference for continuous variables | MIT |
| ennemi | Easy-API nearest-neighbour MI; continuous focus | MIT |
| dit | Discrete information theory — partition-mutual-info, intrinsic-mutual-info, multivariate generalizations | BSD-3 |
| pyitlib | Older reference impl; standard MI / entropy / KL | GPL-3 |
Decision: infomeasure for the Σ pillar (covers both continuous + discrete with state-of-the-art estimators). NPEET as a second-opinion oracle for the non-parametric KSG estimator.
7. Optimal transport / Earth Mover's Distance¶
Kimera replaced the IIT30 _emd_hamming HiGHS LP with an O(N·2^N)
closed form (CLAUDE.md §2026-05-15 IIT30 fix; commit 9c055d303,
~10% throughput gain). External validation:
| Tool | Strengths | License |
|---|---|---|
| POT | Network simplex (exact EMD), Sinkhorn (entropic regularization, GPU), Gromov-Wasserstein, barycentres | MIT |
Decision: POT as Ophamin's reference oracle for _emd_hamming —
write a regression scenario asserting Kimera's closed form ≈ POT's
network-simplex within 1e-12. Ratchet on every commit touching
_effect_repertoire / _cause_repertoire.
8. Drift detection¶
Ophamin already has a comparing/drift/ directory but the scope is per-PR
proof-record drift. For per-commit metric drift across many strata, a
purpose-built library helps.
| Tool | Strengths | License |
|---|---|---|
| Frouros | Pure-purpose drift detection (no overlap with anomaly / adversarial / imbalance — single responsibility). Both concept + data drift. | BSD-3 |
| Evidently | 100+ pre-built checks, interactive HTML reports, well-suited for ratchet-style monitoring | Apache-2 |
| NannyML | Performance estimation without ground truth (univariate + multivariate) | Apache-2 |
| TorchDrift | PyTorch-native, kernel MMD, classifier two-sample tests | Apache-2 |
| River (installed) | Already in Ophamin. Online algorithms include drift detectors (ADWIN, DDM, EDDM, KSWIN) | BSD-3 |
Decision: Frouros for the new dedicated comparing/drift/ library
(one purpose, signed proof-records). Evidently for HTML-report ratchets.
9. Causal inference¶
Kimera has 222 feedback-loop pathways. Today Ophamin's Σ pillar (planned) would only measure CORRELATION. Causal inference distinguishes X causes Y from X correlates with Y but Z drives both — exactly the question needed when Φ drift correlates with walker M2 amplitude_death.
| Tool | Strengths | License |
|---|---|---|
| DoWhy | PyWhy ecosystem orchestrator. Four-step API (model → identify → estimate → refute). Integrates EconML + CausalML | MIT |
| EconML | Microsoft's HTE estimators (DML, DR-Learner, X-Learner, Causal Forests) | MIT |
| CausalML | Uber's uplift modeling — counterfactual + doubly-robust | Apache-2 |
| CausalPy | Bayesian causal inference (synthetic control, RDD, ITS) on top of PyMC | Apache-2 |
| Tigramite | Time-series causal discovery (PCMCI). Directly applicable to Kimera's per-cycle Φ trajectory | GPL-3 |
Decision: DoWhy as the orchestrator + Tigramite for time-series-causal discovery on Kimera's 638-field per-cycle OrchestratorResult traces.
10. Code-substrate health (static analysis + dependency wiring)¶
Kimera is a 3,363-module substrate (per Ophamin's wiring scan). Today's audit pillars (ruff, bandit, mypy, vulture, radon, pip-audit) cover linting + security + types + dead code + complexity + dep vulns. Gaps:
| Tool | Fills | License |
|---|---|---|
| Semgrep | Custom-rule SAST. Can encode Kimera's no-fallback rule (try: ... except: pass), Pattern-P naming-violation patterns (Enhanced, Unified, *Manager) |
LGPL-2.1 / Commercial |
| Pyright / Pyrefly | Microsoft's fast type checker. Pyrefly (the next-gen) provides IDE features. Faster than mypy on large codebases | MIT |
| Pylint | Deep type-inference + custom plugins beyond what ruff replicates | GPL-2 |
| Prospector | Multi-tool aggregator (Pylint + Mypy + McCabe + dodgy + frosted) | GPL-2 |
| FawltyDeps | Find undeclared + unused 3rd-party deps; reads pyproject / requirements / setup.py / setup.cfg / environment.yml | MIT |
| deptry | Same shape, faster, supports Poetry / pip / PDM / uv / PEP 621 | MIT |
| lizard | Cyclomatic complexity, multi-language; complements radon | MIT |
| refurb | Modernization suggestions for Python ≥ 3.10+ idioms | GPL-3 |
| autoflake | Remove unused imports + variables | MIT |
Decision: ship semgrep + pyright + fawltydeps + deptry as the "audit pillar v2" tier. Each as an opt-in pillar.
11. Profiling + tracing¶
Kimera's takwin.py is 32,743 lines; the 6 audit pillars surfaced
616 findings on it. For per-cycle profile, where is the 8s wall-time
actually going?
| Tool | Strengths | License |
|---|---|---|
| Scalene | Line-level CPU + GPU + memory in one tool. Best for deep substrate-cycle timing | Apache-2 |
| Pyinstrument | Beautiful call-stack visualization. Best for narrative reports | BSD-3 |
| VizTracer | Interactive timeline (Chrome trace viewer). Best when timing IS the bug | Apache-2 |
| Austin | Frame-stack sampler (low overhead). Pairs well with austinp graphics | GPL-3 |
| py-spy (installed) | Already in Ophamin. Production-safe sampling, zero modification of running process | MIT |
| memray (installed) | Already in Ophamin. Per-allocation tracker | Apache-2 |
| line_profiler | Line-level CPU only. Surgical for hot spots after py-spy locates them | BSD-3 |
Decision: add scalene as the "deep cycle inspection" tier. Wire
into instrumenting/ as an opt-in.
12. Time-series + anomaly detection (Cronos / KCCL telemetry)¶
Kimera's Cronos atomic clock has Allan-deviation tracking; KCCL oscillators emit per-phase wall-time. Ophamin should track these as proper time-series.
| Tool | Strengths | License |
|---|---|---|
| sktime | Unified scikit-learn-style API for forecasting / classification / anomaly | BSD-3 |
| Darts | Forecasting + anomaly + filtering. Anomaly module wraps any model into detector | Apache-2 |
| Kats | Meta's one-stop time-series. Detection + forecasting + features + multivariate | MIT |
| TSFresh | Feature-extraction (1000+ features) + selection. Pairs with River for online drift | MIT |
| PyOD | Outlier detection (50+ algorithms). PyODScorer trivially wraps for Darts | BSD-2 |
| STUMPY | Matrix profile (Eamonn Keogh's framework) — fast motif + discord discovery | BSD-3 |
| NeuralForecast | 30+ neural models (NHITS, NBEATS, TFT, Informer, PatchTST). Good for multi-step Φ-trajectory forecasting | Apache-2 |
| StatsForecast | Classical statistical forecasters (AutoARIMA, ETS, MSTL); fast | Apache-2 |
Decision: STUMPY for matrix-profile drift on per-cycle Φ trajectory. Darts for forecasting + anomaly bands on telemetry streams.
13. Distributed tracing + observability stack¶
Ophamin currently consumes Prometheus. Production deployment also wants distributed tracing.
| Tool | Strengths | License |
|---|---|---|
| Jaeger | CNCF-graduated. Adaptive sampling, scalable, rich UI | Apache-2 |
| Grafana Tempo | Cost-efficient (object storage only). Integrates with Prom+Loki+Grafana | AGPL-3 |
| Loki | Log aggregation (queries via LogQL). Part of LGTM stack | AGPL-3 |
| OpenTelemetry (installed) | Industry-standard collector. Already in Ophamin | Apache-2 |
| Mimir | Long-term Prometheus storage (not in Ophamin scope) | AGPL-3 |
| Pyroscope | Continuous profiling (Grafana family). Complements py-spy / scalene | AGPL-3 |
Decision: Tempo as the natural extension of Ophamin's telemetry consumer; both speak OTel. Pyroscope as a continuous-profile addition.
14. CRDT / reconciliation (for Archipel cross-checks)¶
Kimera has its own G-Set, SCAR-DAG, Echoform-chain CRDTs. External references for cross-check + interop:
| Tool | Strengths | License |
|---|---|---|
| Yjs | Most-deployed CRDT library globally as of 2026. Modular (text, array, map, xml) | MIT |
| Automerge | JSON CRDT (Rust core). Automerge 3.x has columnar storage + Peritext for rich text | MIT |
| pycrdt | Python bindings for Yrs (the Rust port of Yjs) | MIT |
| y-py | Python bindings for Y-CRDT | MIT |
Decision: pycrdt as a property-based-test oracle for Kimera's G-Set + SCAR-DAG (Hypothesis strategies that compare Kimera's CRDT against Yrs's CRDT for the same operation sequence).
15. Statistical + uncertainty quantification¶
Kimera's empirical record (per Ophamin scenarios) uses Wilson 95% CI + Cohen's d + Mann-Whitney U today. To raise the rigor bar:
| Tool | Strengths | License |
|---|---|---|
| MAPIE (installed) | Already in Ophamin. v1+ has 20+ conformal methods | BSD-3 |
| PUNCC | Conformal prediction + classification + object detection + anomaly. Integrates with PyTorch / TensorFlow | MIT |
| crepes | Conformal regressors + classifiers + predictive systems. Semi-online | BSD-3 |
| TorchCP | PyTorch-native, fastest of the four (90% faster than PUNCC). LLM / GNN support | Apache-2 |
| ArviZ | Bayesian-posterior visualization + diagnostics. Standard across PyMC / Stan / NumPyro | Apache-2 |
| PyMC | Bayesian PPL. Compiles to C / JAX / Numba. For posterior-tempering scenarios (CLAUDE.md §Family L L9 already does this manually) | Apache-2 |
| NumPyro | JAX-backed PPL. Faster than PyMC for HMC / NUTS at scale | Apache-2 |
| pingouin | Effect-size + correlation + multiple comparison. Cleaner API than scipy.stats for many tests | GPL-3 |
Decision: pingouin for proper effect-size + multiple-comparison correction in cross-stratum scenarios. PyMC + ArviZ for any Bayesian scenario (Family L's posterior-tempering would benefit from a proper PPL implementation).
16. SBOM / supply chain / dependency security¶
Beyond pip-audit (which we have):
| Tool | Strengths | License |
|---|---|---|
| Syft | SBOM generation for containers + filesystems + archives. Outputs CycloneDX + SPDX. Anchore-backed | Apache-2 |
| Grype | Vulnerability scanning of SBOMs. Pairs with Syft | Apache-2 |
| Trivy | All-in-one (vuln + IaC + SBOM). WARNING: compromised twice in March 2026 supply chain attack — re-evaluate before use | Apache-2 |
| ScanCode-toolkit | License + copyright + dependency detection. Outputs SPDX + CycloneDX | Apache-2 |
| Dependency-Track | Long-term SBOM monitoring (server) | Apache-2 |
| OSV-Scanner | Google's CLI scanner against OSV.dev DB | Apache-2 |
| Socket | Runtime install-time analysis (network requests, dynamic code execution) — closest to real-time supply chain protection | Commercial |
| OpenSSF Scorecard | Upstream-dep health scoring (signed releases, branch protection, fuzz coverage) | Apache-2 |
| Safety | Python-specific vuln scanner | MIT |
| Snyk Open Source | Reachability analysis, auto-fix PRs, license compliance | Commercial |
| SPDX-tools | SPDX format generation/validation | Apache-2 |
Decision: Syft + Grype as the bulk SBOM layer (covering Kimera's container deployments). OSV-Scanner as a CI fast-fail gate. SPDX-tools for the interop wheel (currently CycloneDX-only).
17. Quantum / thermodynamic / SPDE substrate libraries¶
For cross-checking Kimera's physics-side claims:
| Tool | Aspect | License |
|---|---|---|
| Qiskit | Quantum circuits + pulse + dynamic. Most feature-rich. Cross-check PrimeWaveQuantumEngine outputs |
Apache-2 |
| Cirq | Google's NISQ-focused. Lightweight | Apache-2 |
| PennyLane | Quantum + ML differentiable programming. For Kimera's quantum-RL primitives | Apache-2 |
| QuTiP | Quantum dynamics simulation. Best for the thermofield-double + open-system dynamics | BSD-3 |
| Strawberry Fields | Photonic quantum circuits | Apache-2 |
| FEniCS | FEM-based PDE solver. Reference oracle for SPDE engine | LGPL |
| Firedrake | Newer, FEniCS-compatible UFL, automated code generation | LGPL |
| DeepXDE | Physics-informed neural networks (PINNs) for Allen-Cahn, Gray-Scott, Eikonal | Apache-2 |
Decision: Qiskit + QuTiP as quantum cross-check oracles. FEniCS or Firedrake as the SPDE oracle. Out-of-scope for v0.2; flagged for v0.3.
18. Graph + network analysis (manifold + SCAR-network)¶
Kimera builds graphs everywhere (geoid manifold, SCAR network, Arachne web, Piovra arms). Today's NetworkX is fine for small graphs but slow at scale.
| Tool | Strengths | Speed vs NetworkX |
|---|---|---|
| NetworkX (implicit) | Pure Python, easiest install | 1× (baseline) |
| graph-tool | C++ core. Statistical inference (SBM, hierarchical SBM) | 40-250× faster |
| igraph | C core. Mature graph algorithms | ~30× faster |
| NetworKit | Parallel C++. Page rank in 0.2s on Pokec dataset | Up to 250× faster |
| SNAP-py | Stanford's network analysis platform | ~10× faster |
Decision: NetworKit for any graph-analysis scenario hitting Kimera's full SCAR network (1M+ scars). graph-tool if SBM-style community detection is needed for substrate clustering.
19. Encoder / embedding (for Rosetta primes cross-check)¶
Kimera has its own EnhancedTransformer + FastText fallback. External encoders for cross-check:
| Tool | Strengths | License |
|---|---|---|
| sentence-transformers | 15,000+ pretrained models on Hugging Face. Standard | Apache-2 |
| BGE | BAAI's BGE-M3 (already in Kimera as a fallback) | MIT |
| GTE | Alibaba's multi-lingual encoders | Apache-2 |
| mxbai-embed | Mixedbread AI's embeddings | Apache-2 |
| Nomic Embed | Nomic's open-source encoders, including ModernBERT-based | Apache-2 |
| gensim | Word2Vec, FastText, Doc2Vec — for cross-checking Kimera's FastText | LGPL |
Decision: sentence-transformers as the canonical reference oracle
for assign_via_lyriform's output. Per-encoder cross-check scenario:
"two semantically equivalent sentences produce primes whose Wilson CI
on Jaccard ≥ 0.5".
20. Probabilistic / fuzz testing¶
For Kimera's CRDT primitives + the no-fallback rule:
| Tool | Strengths | License |
|---|---|---|
| Hypothesis | Property-based testing with strategies. CRDT laws (idempotent / commutative / associative) are textbook strategies | MPL-2 |
| Atheris | Coverage-guided fuzzing (libFuzzer-based). Catches crashes in PrimePacket parsers, RIBLT decoders | Apache-2 |
| Mutmut | Mutation testing — 1,200 mutants/min, 88.5% kill rate. Faster than Cosmic Ray | MIT |
| Cosmic Ray | Mutation testing — most operators, parallel execution. 82.7% kill rate | MIT |
| Schemathesis | API contract testing from OpenAPI / GraphQL specs. For Kimera's 47 REST routers | MIT |
| Coverage.py | Standard line + branch coverage | Apache-2 |
| Slipcover | Faster (10×) coverage for Python | Apache-2 |
Decision: Hypothesis for CRDT laws. Schemathesis as the API-contract audit pillar. Mutmut as the mutation-testing audit pillar.
21. Acceleration (when scenarios get slow)¶
Kimera's scenarios run cycles in batches; some take 8s+ each. If Ophamin's analytic pillars become bottlenecks:
| Tool | Strengths | License |
|---|---|---|
| JAX | Autograd + JIT for scientific computing. NumPy-API. Hundreds of × faster than pure Python | Apache-2 |
| Numba | JIT for NumPy + scalar code. 6× speedup typical, no source rewriting | BSD-3 |
| Cython | Compiles annotated Python to C. 50× speedups, but rewrite needed | Apache-2 |
| Polars | Faster pandas-replacement; Apache-Arrow-backed | MIT |
| DuckDB | In-process analytical SQL; great for proof-record corpus queries | MIT |
Decision: Polars for comparing/drift/ corpus queries. JAX for
any matrix-heavy pillar. Numba for hot inner loops in the wiring probe.
22. Recommended integration sequence (PRs)¶
| PR | Adds | Wheel | LOC budget | Tier-2 / Tier-3 |
|---|---|---|---|---|
| 1 | PyPhi cross-check oracle for IIT30 | measuring | 200 + 100 tests | Tier-2 |
| 2 | POT cross-check oracle for _emd_hamming |
measuring | 150 + 80 tests | Tier-2 |
| 3 | giotto-tda persistence-diagram drift | measuring + comparing | 350 + 200 tests | Tier-2 |
| 4 | infomeasure + Σ correlation pillar | measuring | 400 + 250 tests | Tier-2 |
| 5 | Frouros drift detector for comparing/drift/ |
comparing | 250 + 150 tests | Tier-2 |
| 6 | DoWhy + Tigramite causal-graph identification | measuring + comparing | 500 + 300 tests | Tier-2 |
| 7 | Semgrep custom-rule audit pillar (no-fallback + Pattern-P) | auditing | 200 + 150 tests | Tier-2 |
| 8 | Pyright audit pillar (faster type-check than mypy) | auditing | 100 + 50 tests | Tier-2 |
| 9 | deptry + fawltydeps for dependency wiring | auditing | 150 + 80 tests | Tier-2 |
| 10 | Hypothesis CRDT-law tests (Tier-2 — for Kimera's CRDTs, not Ophamin's) | (Kimera-side) | 300 + 200 tests | Tier-3 (owner-gated) |
| 11 | Schemathesis API-contract audit pillar | auditing | 200 + 100 tests | Tier-2 |
| 12 | Mutmut mutation-testing audit pillar | auditing | 200 + 100 tests | Tier-2 |
| 13 | Scalene per-cycle line-level profile | instrumenting | 200 + 100 tests | Tier-2 |
| 14 | STUMPY matrix-profile drift on Φ-trajectory | comparing | 250 + 150 tests | Tier-2 |
| 15 | Syft + Grype SBOM gen + vuln scan in interop | interop + auditing | 300 + 150 tests | Tier-2 |
| 16 | OSSF Scorecard upstream-dep health score | auditing | 200 + 100 tests | Tier-2 |
| 17 | NetworKit for SCAR-network analysis at scale | seeing | 250 + 150 tests | Tier-2 |
| 18 | sentence-transformers cross-encoder Jaccard scenario | measuring | 350 + 200 tests | Tier-2 |
| 19 | PyMC + ArviZ Bayesian Family-L style scenario | measuring | 400 + 250 tests | Tier-2 |
| 20 | Pingouin effect-size + multi-comparison helper | measuring | 100 + 60 tests | Tier-2 |
Total rough budget: ~5,000 LOC + ~2,800 tests across 20 PRs.
If sequenced one PR per 1-2 sessions, this is 4-6 weeks of work. Each PR is independently reviewable and ships value on its own.
23. What's deliberately out of scope¶
- Closed-source / commercial-only: Datadog, New Relic, Snyk Code, Splunk. Ophamin should stay OSS-first; an enterprise can add these as adapters but they're not core.
- JVM-only: JIDT (Java Information Dynamics Toolkit) — equivalent Python coverage exists.
- Heavy external infra: full Loki / Tempo deployments need K8s and go beyond what Ophamin can usefully wrap. Provide consumer adapters but don't ship the deployment.
- AI/ML model serving: NVIDIA Triton, KServe, Seldon — not relevant to Kimera-as-substrate.
- GPU-only tools: anything requiring CUDA without a CPU fallback would break MockSubstrate-only test runs.
24. Honest unknowns¶
- Trivy supply-chain compromise (March 2026): per the search results, Trivy was compromised twice. Verify current safety status before any integration. Syft+Grype is the cleaner Anchore-stack alternative.
- GeoMIP availability: 165-326× speedup over PyPhi sounds great but the search results don't surface a maintained pip-installable package. Verify maturity before substituting for PyPhi.
- Pyrefly (Pyre next-gen) maturity: Meta's replacement; depends on whether 2026 release supports the codebases Ophamin targets.
- License compatibility with Ophamin's proprietary license: GPL-3 tools (PyPhi, Tigramite, refurb, Pylint) need a usage-pattern audit before bundling vs invoking-via-subprocess.
- Performance on Kimera-scale inputs: most academic libraries benchmarked on smaller graphs / corpora. Validate at Kimera-SWM scale (3,363 modules, 1M+ scars potential) before committing to any one tool.
25. Sources¶
Per the WebSearch tool requirement, sources for this synthesis:
IIT / Φ¶
TDA¶
Information theory¶
Causal inference¶
Optimal transport¶
Drift detection¶
Profiling¶
Dependency analysis¶
SBOM / supply chain¶
Mutation + property testing¶
Static analysis¶
- Top 10 Python Code Analysis Tools 2026
- Best Code Review Tools for Python 2026
- Awesome static analysis
Quantum¶
Reaction-diffusion / PDE¶
Graph analysis¶
CRDT¶
Bayesian / PPL¶
MLOps¶
Time-series¶
Distributed tracing¶
Kuramoto / oscillators¶
Embeddings¶
Dimensionality reduction¶
SAT/SMT¶
Conformal prediction¶
Streaming / online¶
Acceleration¶
Supply chain (more)¶
Authored by Claude (Opus 4.7 1M context), 2026-05-15. Synthesized from 22 parallel WebSearch queries across the 20+ Kimera-SWM aspects. Subject to owner pick + sequencing decisions before implementation.