top repos

{
  "id": "P15-R-001",
  "repo": "VowpalWabbit/vowpal_wabbit",
  "category": "bandits",
  "source_url": "https://github.com/VowpalWabbit/vowpal_wabbit",
  "observed_date": "2026-08-27",
  "evidence_class": "first_party_docs",
  "claim": "Vowpal Wabbit is a machine learning system which pushes the frontier of machine learning with techniques such as online, hashing, allreduce, reductions, learning2search, active, and interactive learning.",
  "signal_to_update": "Reward (click/revenue) + logged propensity updates online CB policy via reductions",
  "sparse_data_fit": "Poor at tens: asymptotic regret guarantees, built for high-throughput streams",
  "production_evidence": "MSR Decision Service paper: 3 live deployments (MSN, Complex, TrackRevenue), 10s of millions of users, thousands req/s, 14-25pct lift",
  "license_observed": "LICENSE read: BSD-3-clause-style text, Microsoft/Yahoo copyright; GitHub API says NOASSERTION because of custom header",
  "maintenance_state": "Active, pushed 2026-08 (stars=8706, archived=false, last_push=2026-08-26)",
  "limitations": "Needs volume; C++ build; propensity logging must be engineered in from day one",
  "disposition": "top10",
  "top10_rank": 1,
  "top10_rationale": "The Decision Service lineage is the canonical explore-log-learn-deploy loop with propensity logging, which is the single mechanism Actionist most needs to make build outcomes counterfactually reusable. Adopt the logging discipline and the reduction framing even though VW's own sample regime is far denser than tens of builds."
}

{
  "id": "P15-R-002",
  "repo": "fidelity/mabwiser",
  "category": "bandits",
  "source_url": "https://github.com/fidelity/mabwiser",
  "observed_date": "2026-08-27",
  "evidence_class": "first_party_docs",
  "claim": "MABWiser: Contextual Multi-Armed Bandits Library",
  "signal_to_update": "fit(decisions, rewards) then predict; reward per chosen arm updates that arm's estimator; partial_fit for warm updates",
  "sparse_data_fit": "Best sparse fit of the bandit set: context-free TS/UCB1 usable at tens; neighborhood policies (Radius, KNearest, TreeBandit, Clusters) pool across similar contexts by design",
  "production_evidence": "Fidelity AI Center of Excellence provenance + 2 peer-reviewed papers (IJAIT 2021, ICTAI 2019), but README self-describes as research/rapid-prototyping library; no named production system",
  "license_observed": "LICENSE read: Apache-2.0 full text",
  "maintenance_state": "Maintained but pushed 2024-09, slowing (stars=291, archived=false, last_push=2024-09-05)",
  "limitations": "No first-party named production deployment; do not assert Fidelity runs it in prod",
  "disposition": "top10",
  "top10_rank": 3,
  "top10_rationale": "The neighborhood policies are the direct answer to sparse asset reranking: they pool reward evidence across similar contexts instead of demanding per-arm volume, which is exactly the cold-start-by-similarity move for a new client or industry. Batch scikit-learn-style API fits tens-of-builds cadence with no streaming infrastructure."
}

{
  "id": "P15-R-003",
  "repo": "david-cortes/contextualbandits",
  "category": "bandits",
  "source_url": "https://github.com/david-cortes/contextualbandits",
  "observed_date": "2026-08-27",
  "evidence_class": "first_party_docs",
  "claim": "Python implementations of contextual bandits algorithms",
  "signal_to_update": "Reward per chosen arm updates online CB policies built over sklearn base learners",
  "sparse_data_fit": "Moderate: offline/batch friendly, works with small base learners, but still bandit-asymptotic",
  "production_evidence": "Academic implementations of published algorithms; no named production deployment",
  "license_observed": "LICENSE read: BSD 2-Clause full text",
  "maintenance_state": "Active, pushed 2026-06 (stars=838, archived=false, last_push=2026-06-28)",
  "limitations": "Reference implementation quality, not a serving system",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-004",
  "repo": "st-tech/zr-obp",
  "category": "bandits",
  "source_url": "https://github.com/st-tech/zr-obp",
  "observed_date": "2026-08-27",
  "evidence_class": "first_party_docs",
  "claim": "Open Bandit Pipeline: a python library for bandit algorithms and off-policy evaluation",
  "signal_to_update": "Logged (context, action, position, click, true propensity) enables off-policy evaluation of a new policy before deploying it",
  "sparse_data_fit": "Wrong shape for sparse: IPS-family estimator variance explodes at low N; needs thousands per arm",
  "production_evidence": "Real 7-day A/B test on ZOZOTOWN fashion e-commerce, ~26M rows, Uniform Random vs Bernoulli TS, true propensities known",
  "license_observed": "LICENSE read: Apache-2.0 full text",
  "maintenance_state": "Pushed 2024-06, slowing (stars=707, archived=false, last_push=2024-06-03)",
  "limitations": "OPE needs known propensities and volume; benchmark framework not a production service",
  "disposition": "top10",
  "top10_rank": 7,
  "top10_rationale": "It is the clearest worked example of evaluating a candidate ranking policy from logged data before shipping it, which is the risk control Actionist needs when changing shelf ranking. Include for the methodology and the known-propensity discipline, explicitly not for its sample regime."
}

{
  "id": "P15-R-005",
  "repo": "ray-project/ray",
  "category": "bandits",
  "source_url": "https://github.com/ray-project/ray",
  "observed_date": "2026-08-27",
  "evidence_class": "observed_behavior",
  "claim": "Ray is an AI compute engine. Ray consists of a core distributed runtime and a set of AI Libraries for accelerating ML workloads.",
  "signal_to_update": "RLlib contextual bandit algos update policy from reward stream",
  "sparse_data_fit": "Poor: distributed RL infrastructure, heavy for tens of observations",
  "production_evidence": "Ray widely deployed in production; bandits specifically are a small RLlib subset",
  "license_observed": "Apache-2.0 per API",
  "maintenance_state": "Very active (stars=43618, archived=false, last_push=2026-08-27)",
  "limitations": "Massive dependency for a small need",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-006",
  "repo": "tensorflow/agents",
  "category": "bandits",
  "source_url": "https://github.com/tensorflow/agents",
  "observed_date": "2026-08-27",
  "evidence_class": "observed_behavior",
  "claim": "TF-Agents: A reliable, scalable and easy to use TensorFlow library for Contextual Bandits and Reinforcement Learning.",
  "signal_to_update": "TF-Agents bandit agents update policy from reward",
  "sparse_data_fit": "Poor: designed for large-scale RL training",
  "production_evidence": "Google-authored library; bandits subset documented",
  "license_observed": "Apache-2.0 per API",
  "maintenance_state": "Pushed 2026-01, slowing (stars=3025, archived=false, last_push=2026-01-16)",
  "limitations": "TF dependency; heavyweight",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-007",
  "repo": "motiwari/BanditPAM",
  "category": "bandits",
  "source_url": "https://github.com/motiwari/BanditPAM",
  "observed_date": "2026-08-27",
  "evidence_class": "observed_behavior",
  "claim": "BanditPAM C++ implementation and Python package",
  "signal_to_update": "Bandit-based sampling accelerates k-medoids clustering",
  "sparse_data_fit": "Not applicable: bandits used as an optimization trick, not a feedback loop",
  "production_evidence": "Published algorithm (NeurIPS)",
  "license_observed": "MIT per API",
  "maintenance_state": "Pushed 2025-08 (stars=659, archived=false, last_push=2025-08-25)",
  "limitations": "Clustering accelerator, not a decision loop; miscategorized in most bandit lists",
  "disposition": "excluded",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-008",
  "repo": "SMPyBandits/SMPyBandits",
  "category": "bandits",
  "source_url": "https://github.com/SMPyBandits/SMPyBandits",
  "observed_date": "2026-08-27",
  "evidence_class": "observed_behavior",
  "claim": "🔬 Research Framework for Single and Multi-Players 🎰 Multi-Arms Bandits (MAB) Algorithms, implementing all the state-of-the-art algorithms for single-player (UCB, KL-UCB, Thompson...) and multi-player (MusicalChair, MEGA, rhoRand, MCTop/R...",
  "signal_to_update": "Simulation of MAB policies against synthetic environments",
  "sparse_data_fit": "Research simulation only",
  "production_evidence": "Academic research framework",
  "license_observed": "MIT per API",
  "maintenance_state": "Active, pushed 2026-06 (stars=424, archived=false, last_push=2026-06-19)",
  "limitations": "Simulation framework, no production path",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-009",
  "repo": "bgalbraith/bandits",
  "category": "bandits",
  "source_url": "https://github.com/bgalbraith/bandits",
  "observed_date": "2026-08-27",
  "evidence_class": "observed_behavior",
  "claim": "Python library for Multi-Armed Bandits",
  "signal_to_update": "Basic MAB policy implementations",
  "sparse_data_fit": "Educational scale",
  "production_evidence": "None; teaching library",
  "license_observed": "Apache-2.0 per API",
  "maintenance_state": "Dead, pushed 2020-02 (stars=771, archived=false, last_push=2020-02-11)",
  "limitations": "Unmaintained 6+ years",
  "disposition": "excluded",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-010",
  "repo": "johnmyleswhite/BanditsBook",
  "category": "bandits",
  "source_url": "https://github.com/johnmyleswhite/BanditsBook",
  "observed_date": "2026-08-27",
  "evidence_class": "secondary",
  "claim": "Code for my book on Multi-Armed Bandit Algorithms",
  "signal_to_update": "Book companion code for MAB algorithms",
  "sparse_data_fit": "Educational",
  "production_evidence": "None; book companion",
  "license_observed": "LICENSE not standard-detected (NOASSERTION); book companion code",
  "maintenance_state": "Dead, pushed 2020-01 (stars=923, archived=false, last_push=2020-01-09)",
  "limitations": "Pedagogical only",
  "disposition": "excluded",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-011",
  "repo": "alison-carrera/onn",
  "category": "bandits",
  "source_url": "https://github.com/alison-carrera/onn",
  "observed_date": "2026-08-27",
  "evidence_class": "observed_behavior",
  "claim": "Online Deep Learning: Learning Deep Neural Networks on the Fly / Non-linear Contextual Bandit Algorithm (ONN_THS)",
  "signal_to_update": "Online deep learning + contextual bandit (ONN_THS) update per sample",
  "sparse_data_fit": "Poor: neural online learning needs volume",
  "production_evidence": "Research implementation",
  "license_observed": "Apache-2.0 per API",
  "maintenance_state": "Pushed 2026-04 (stars=192, archived=false, last_push=2026-04-13)",
  "limitations": "Small project, neural approach needs data",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-012",
  "repo": "meta-pytorch/botorch",
  "category": "bayesian_preference",
  "source_url": "https://github.com/meta-pytorch/botorch",
  "observed_date": "2026-08-27",
  "evidence_class": "first_party_docs",
  "claim": "Bayesian optimization in PyTorch",
  "signal_to_update": "Noisy observation of an objective updates a Gaussian process posterior; acquisition function proposes next configuration",
  "sparse_data_fit": "Excellent: GP Bayesian optimization is explicitly sample-efficient, designed for tens of expensive evaluations",
  "production_evidence": "Meta: constrained BO with noisy experiments applied to real Facebook ranking system and compiler flags via live A/B tests",
  "license_observed": "LICENSE read: MIT, Meta Platforms copyright",
  "maintenance_state": "Very active, pushed 2026-08 (stars=3589, archived=false, last_push=2026-08-26)",
  "limitations": "Continuous parameter spaces are its home; discrete asset selection needs encoding work",
  "disposition": "top10",
  "top10_rank": 2,
  "top10_rationale": "Bayesian optimization is the one mature mechanism whose explicit design point is tens of expensive noisy observations, which is precisely Actionist's regime of tens of builds. The GP posterior gives calibrated uncertainty per candidate, turning explore-vs-exploit over shelf assets into a principled acquisition decision rather than a hand-tuned epsilon."
}

{
  "id": "P15-R-013",
  "repo": "facebook/Ax",
  "category": "bayesian_preference",
  "source_url": "https://github.com/facebook/Ax",
  "observed_date": "2026-08-27",
  "evidence_class": "first_party_docs",
  "claim": "Adaptive Experimentation Platform",
  "signal_to_update": "Adaptive experiment: each trial result updates the surrogate model and chooses the next arm",
  "sparse_data_fit": "Excellent: Meta states the tools work without large quantities of data",
  "production_evidence": "Meta Research announcement states use in research and production for adaptive experimentation",
  "license_observed": "LICENSE read: MIT, Meta Platforms copyright",
  "maintenance_state": "Very active, pushed 2026-08 (stars=2792, archived=false, last_push=2026-08-19)",
  "limitations": "Experiment-management layer; opinionated scaffolding",
  "disposition": "top10",
  "top10_rank": 4,
  "top10_rationale": "Ax is the productionized service layer over BoTorch and shows how to run adaptive experiments as a managed loop with trials, arms and metrics, which maps cleanly onto Actionist builds as trials. It is the most directly copyable orchestration shape for sequential decisions under small samples."
}

{
  "id": "P15-R-014",
  "repo": "lucasmaystre/choix",
  "category": "bayesian_preference",
  "source_url": "https://github.com/lucasmaystre/choix",
  "observed_date": "2026-08-27",
  "evidence_class": "first_party_docs",
  "claim": "Inference algorithms for models based on Luce's choice axiom",
  "signal_to_update": "Pairwise or partial-ranking comparison outcomes update one latent strength parameter per item (Bradley-Terry / Plackett-Luce)",
  "sparse_data_fit": "Strongest of the survey: O(n) parameters not O(n squared); roughly n log n comparisons suffice; EP variant returns per-item uncertainty",
  "production_evidence": "Inference library, not a service; no deployment claimed or needed",
  "license_observed": "LICENSE read: MIT full text, Lucas Maystre 2015",
  "maintenance_state": "Pushed 2025-09, stable/low-churn (stars=203, archived=false, last_push=2025-09-05)",
  "limitations": "Comparison graph must be connected or strengths are unidentifiable; needs regularization or unbeaten items diverge to infinity",
  "disposition": "top10",
  "top10_rank": 5,
  "top10_rationale": "Bradley-Terry-Luce is the most sample-efficient way to turn sparse relative outcomes (asset A adapted cheaper than asset B on this build) into a global ranking, because every comparison informs two items at once. The documented connectivity and regularization caveats are exactly the failure modes Actionist would hit with a thin build history, and they are cheap to design around."
}

{
  "id": "P15-R-015",
  "repo": "pymc-devs/pymc",
  "category": "bayesian_preference",
  "source_url": "https://github.com/pymc-devs/pymc",
  "observed_date": "2026-08-27",
  "evidence_class": "first_party_docs",
  "claim": "Bayesian Modeling and Probabilistic Programming in Python",
  "signal_to_update": "Bayesian posterior update from observed data",
  "sparse_data_fit": "Excellent: hierarchical partial pooling is the canonical sparse-data remedy",
  "production_evidence": "Widely used in industry; general-purpose PPL",
  "license_observed": "LICENSE read: header states PyMC is distributed under Apache-2.0, followed by the full Apache-2.0 text; also contains MIT-licensed AePPL code and historical Academic Free License copyrights (2006-2008), which is why the API reports NOASSERTION",
  "maintenance_state": "Very active (stars=9723, archived=false, last_push=2026-08-24)",
  "limitations": "General framework, requires modeling expertise",
  "disposition": "top10",
  "top10_rank": 6,
  "top10_rationale": "Hierarchical Bayesian partial pooling is the principled cold-start mechanism: a new client or industry inherits the population prior and shrinks toward its own data as builds accumulate. This is the one modeling idea that directly answers cold-start priors rather than working around them."
}

{
  "id": "P15-R-016",
  "repo": "stan-dev/stan",
  "category": "bayesian_preference",
  "source_url": "https://github.com/stan-dev/stan",
  "observed_date": "2026-08-27",
  "evidence_class": "observed_behavior",
  "claim": "Stan development repository. The master branch contains the current release. The develop branch contains the latest stable development.  See the Developer Process Wiki for details.",
  "signal_to_update": "Bayesian posterior update via HMC/NUTS",
  "sparse_data_fit": "Excellent for small N hierarchical models",
  "production_evidence": "Mature, widely used across research and industry",
  "license_observed": "BSD-3-Clause per API",
  "maintenance_state": "Very active (stars=2761, archived=false, last_push=2026-08-26)",
  "limitations": "Separate language; compilation step",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-017",
  "repo": "stan-dev/cmdstanpy",
  "category": "bayesian_preference",
  "source_url": "https://github.com/stan-dev/cmdstanpy",
  "observed_date": "2026-08-27",
  "evidence_class": "observed_behavior",
  "claim": "CmdStanPy is a lightweight interface to Stan for Python users which provides the necessary objects and functions to compile a Stan program and fit the model to data using CmdStan.",
  "signal_to_update": "Python interface to Stan inference",
  "sparse_data_fit": "Same as Stan",
  "production_evidence": "Interface layer",
  "license_observed": "BSD-3-Clause per API",
  "maintenance_state": "Very active (stars=198, archived=false, last_push=2026-08-03)",
  "limitations": "Thin wrapper",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-018",
  "repo": "scikit-optimize/scikit-optimize",
  "category": "bayesian_preference",
  "source_url": "https://github.com/scikit-optimize/scikit-optimize",
  "observed_date": "2026-08-27",
  "evidence_class": "observed_behavior",
  "claim": "Sequential model-based optimization with a  `scipy.optimize` interface",
  "signal_to_update": "Sequential model-based optimization updates surrogate from evaluations",
  "sparse_data_fit": "Good sample efficiency in principle",
  "production_evidence": "Was widely used; now archived",
  "license_observed": "BSD-3-Clause per API",
  "maintenance_state": "ARCHIVED, pushed 2024-02 (stars=2828, archived=true, last_push=2024-02-23)",
  "limitations": "Archived; do not adopt for new work",
  "disposition": "excluded",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-019",
  "repo": "optuna/optuna",
  "category": "bayesian_preference",
  "source_url": "https://github.com/optuna/optuna",
  "observed_date": "2026-08-27",
  "evidence_class": "observed_behavior",
  "claim": "A hyperparameter optimization framework",
  "signal_to_update": "Trial result updates TPE sampler which proposes next trial; pruning kills bad trials early",
  "sparse_data_fit": "Good: TPE works at tens-to-hundreds of trials",
  "production_evidence": "Very widely adopted in industry for HPO",
  "license_observed": "MIT per API",
  "maintenance_state": "Very active (stars=14713, archived=false, last_push=2026-08-27)",
  "limitations": "Hyperparameter search framing, not asset ranking",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-020",
  "repo": "yuki-koyama/sequential-gallery",
  "category": "bayesian_preference",
  "source_url": "https://github.com/yuki-koyama/sequential-gallery",
  "observed_date": "2026-08-27",
  "evidence_class": "observed_behavior",
  "claim": "Sequential Gallery for Interactive Visual Design Optimization [SIGGRAPH 2020]",
  "signal_to_update": "Human choice among a gallery of options updates a Bayesian preference model to pick the next gallery",
  "sparse_data_fit": "Designed for very few human decisions (SIGGRAPH interactive design)",
  "production_evidence": "Academic (SIGGRAPH 2020) prototype",
  "license_observed": "MIT per API",
  "maintenance_state": "Tiny project, 22 stars, pushed 2025-11 (stars=22, archived=false, last_push=2025-11-11)",
  "limitations": "Research prototype, C++, no production path",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-021",
  "repo": "sublee/trueskill",
  "category": "bayesian_preference",
  "source_url": "https://github.com/sublee/trueskill",
  "observed_date": "2026-08-27",
  "evidence_class": "first_party_docs",
  "claim": "An implementation of the TrueSkill rating system for Python",
  "signal_to_update": "Match outcomes update per-player Gaussian skill (mean, variance)",
  "sparse_data_fit": "Excellent: converges in very few matches, carries explicit uncertainty",
  "production_evidence": "Microsoft Xbox Live production lineage (the algorithm)",
  "license_observed": "LICENSE read: BSD for code BUT explicit caution — Microsoft permits TrueSkill(TM) only for Xbox Live games or non-commercial projects; verbatim \"If your project is commercial, you should find another rating system\"",
  "maintenance_state": "Pushed 2023-08, stable (stars=803, archived=false, last_push=2023-08-30)",
  "limitations": "LICENCE TRAP: patented algorithm, commercial use restricted despite BSD code badge",
  "disposition": "excluded",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-022",
  "repo": "philihp/openskill.js",
  "category": "bayesian_preference",
  "source_url": "https://github.com/philihp/openskill.js",
  "observed_date": "2026-08-27",
  "evidence_class": "observed_behavior",
  "claim": "A faster, open-license alternative to Microsoft TrueSkill",
  "signal_to_update": "Match/comparison outcomes update Gaussian skill ratings; patent-free TrueSkill alternative",
  "sparse_data_fit": "Excellent: Bayesian rating converges in few observations, uncertainty native",
  "production_evidence": "Used in gaming/leaderboard contexts; JS implementation",
  "license_observed": "MIT per API",
  "maintenance_state": "Active, pushed 2026-08 (stars=256, archived=false, last_push=2026-08-22)",
  "limitations": "Weaker theoretical pedigree than TrueSkill; JS only",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-023",
  "repo": "hyperopt/hyperopt",
  "category": "bayesian_preference",
  "source_url": "https://github.com/hyperopt/hyperopt",
  "observed_date": "2026-08-27",
  "evidence_class": "observed_behavior",
  "claim": "Distributed Asynchronous Hyperparameter Optimization in Python",
  "signal_to_update": "TPE updates from trial losses",
  "sparse_data_fit": "Moderate",
  "production_evidence": "Widely used historically",
  "license_observed": "LICENSE not standard-detected (NOASSERTION); BSD per project",
  "maintenance_state": "Active but low churn (stars=7592, archived=false, last_push=2026-08-24)",
  "limitations": "Largely superseded by Optuna",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-024",
  "repo": "automl/SMAC3",
  "category": "bayesian_preference",
  "source_url": "https://github.com/automl/SMAC3",
  "observed_date": "2026-08-27",
  "evidence_class": "observed_behavior",
  "claim": "SMAC3: A Versatile Bayesian Optimization Package for Hyperparameter Optimization",
  "signal_to_update": "Bayesian optimization with random forests for algorithm configuration",
  "sparse_data_fit": "Good at small budgets",
  "production_evidence": "AutoML research group, competition-proven",
  "license_observed": "LICENSE not standard-detected (NOASSERTION); BSD-3-Clause per project",
  "maintenance_state": "Active (stars=1243, archived=false, last_push=2026-08-24)",
  "limitations": "Research-oriented API",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-025",
  "repo": "growthbook/growthbook",
  "category": "experimentation",
  "source_url": "https://github.com/growthbook/growthbook",
  "observed_date": "2026-08-27",
  "evidence_class": "first_party_docs",
  "claim": "Open Source Feature Flags, Experimentation, and Product Analytics",
  "signal_to_update": "Experiment exposure + metric events produce Bayesian or frequentist lift estimates that gate rollout",
  "sparse_data_fit": "Bayesian engine reports probability-to-beat-control, more usable at low N than p-values",
  "production_evidence": "Open-source product with real commercial deployments",
  "license_observed": "LICENSE read: MIT Expat for core, BUT enterprise dirs (packages/back-end/src/enterprise, front-end/enterprise, shared/src/enterprise) under GrowthBook Enterprise licence",
  "maintenance_state": "Very active (stars=8160, archived=false, last_push=2026-08-27)",
  "limitations": "Mixed licence: enterprise directories are not MIT",
  "disposition": "top10",
  "top10_rank": 10,
  "top10_rationale": "Its Bayesian statistics engine reports probability-that-a-variant-wins rather than significance, which is the honest way to make ship decisions on tens of builds instead of pretending to power a t-test. Also the cleanest open reference for the exposure-to-metric plumbing Actionist must build; note the enterprise directory carve-out."
}

{
  "id": "P15-R-026",
  "repo": "Unleash/unleash",
  "category": "experimentation",
  "source_url": "https://github.com/Unleash/unleash",
  "observed_date": "2026-08-27",
  "evidence_class": "first_party_docs",
  "claim": "Open-source feature management platform",
  "signal_to_update": "Flag evaluation + metrics; targeting rules gate exposure",
  "sparse_data_fit": "Flag infrastructure, no learning",
  "production_evidence": "Widely deployed feature-flag service",
  "license_observed": "LICENSE read: AGPL-3.0 full text (core; enterprise features separate)",
  "maintenance_state": "Very active (stars=13762, archived=false, last_push=2026-08-27)",
  "limitations": "AGPL is a real constraint for a hosted client product; no learning loop",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-027",
  "repo": "Flagsmith/flagsmith",
  "category": "experimentation",
  "source_url": "https://github.com/Flagsmith/flagsmith",
  "observed_date": "2026-08-27",
  "evidence_class": "observed_behavior",
  "claim": "Flagsmith is an open source feature flagging and remote config service. Self-host or use our hosted version at https://app.flagsmith.com.",
  "signal_to_update": "Flag evaluation and segmentation",
  "sparse_data_fit": "No learning loop",
  "production_evidence": "Deployed feature-flag service",
  "license_observed": "BSD-3-Clause per API",
  "maintenance_state": "Very active (stars=6531, archived=false, last_push=2026-08-27)",
  "limitations": "Delivery only",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-028",
  "repo": "PostHog/posthog",
  "category": "experimentation",
  "source_url": "https://github.com/PostHog/posthog",
  "observed_date": "2026-08-27",
  "evidence_class": "first_party_docs",
  "claim": ":hedgehog: PostHog is the leading platform for building self-driving products. Our developer tools – AI observability, analytics, session replay, flags, experiments, error tracking, logs, and more – capture all the context agents need to...",
  "signal_to_update": "Product events feed experiments, funnels and feature-flag decisions",
  "sparse_data_fit": "Needs event volume for its analytics; experiment layer needs power",
  "production_evidence": "Large real deployments as a product analytics platform",
  "license_observed": "LICENSE read: MIT Expat for core, BUT everything under ee/ is licensed separately (enterprise)",
  "maintenance_state": "Very active (stars=39288, archived=false, last_push=2026-08-27)",
  "limitations": "Mixed licence; heavy platform",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-029",
  "repo": "Eppo-exp/eppo-multiplatform",
  "category": "experimentation",
  "source_url": "https://github.com/Eppo-exp/eppo-multiplatform",
  "observed_date": "2026-08-27",
  "evidence_class": "observed_behavior",
  "claim": "SDKs and platform artifacts to support Eppo Flagging.",
  "signal_to_update": "Flag/experiment assignment SDK",
  "sparse_data_fit": "No learning loop in OSS part",
  "production_evidence": "Commercial product; OSS is SDK only",
  "license_observed": "MIT per API",
  "maintenance_state": "Active but 5 stars; SDK component only (stars=5, archived=false, last_push=2026-08-24)",
  "limitations": "OSS surface is thin; the statistics engine is not here",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-030",
  "repo": "facebookarchive/planout",
  "category": "experimentation",
  "source_url": "https://github.com/facebookarchive/planout",
  "observed_date": "2026-08-27",
  "evidence_class": "first_party_docs",
  "claim": "PlanOut is a library and interpreter for designing online experiments.",
  "signal_to_update": "Deterministic hash-based random assignment from an experiment definition language",
  "sparse_data_fit": "Assignment only, no learning",
  "production_evidence": "Facebook production experimentation lineage, published paper",
  "license_observed": "LICENSE read: BSD licence text, Facebook 2014",
  "maintenance_state": "ARCHIVED (verified), pushed 2021-03 (stars=1688, archived=true, last_push=2021-03-19)",
  "limitations": "Archived; assignment DSL only, no analysis or learning",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-031",
  "repo": "intuit/wasabi",
  "category": "experimentation",
  "source_url": "https://github.com/intuit/wasabi",
  "observed_date": "2026-08-27",
  "evidence_class": "observed_behavior",
  "claim": "Wasabi A/B Testing service is an open source project that is no longer under active development or being supported",
  "signal_to_update": "A/B assignment and metrics service",
  "sparse_data_fit": "No learning loop",
  "production_evidence": "Was an Intuit production service",
  "license_observed": "Apache-2.0 per API",
  "maintenance_state": "DEAD despite archived=false flag: README states \"no longer under active development or being supported\"; pushed 2023-05 (stars=1139, archived=false, last_push=2023-05-26)",
  "limitations": "GitHub archived flag is false but the project declares itself unsupported; a badge-vs-body mismatch",
  "disposition": "excluded",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-032",
  "repo": "thomaspoignant/go-feature-flag",
  "category": "experimentation",
  "source_url": "https://github.com/thomaspoignant/go-feature-flag",
  "observed_date": "2026-08-27",
  "evidence_class": "observed_behavior",
  "claim": "GO Feature Flag is a simple, complete and lightweight self-hosted cloud native feature flag solution 100% Open Source — built on OpenFeature 🎛️",
  "signal_to_update": "Flag evaluation with rollout strategies",
  "sparse_data_fit": "No learning",
  "production_evidence": "Active OSS flag service",
  "license_observed": "MIT per API",
  "maintenance_state": "Very active (stars=2090, archived=false, last_push=2026-08-27)",
  "limitations": "Delivery only",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-033",
  "repo": "open-feature/spec",
  "category": "experimentation",
  "source_url": "https://github.com/open-feature/spec",
  "observed_date": "2026-08-27",
  "evidence_class": "observed_behavior",
  "claim": "OpenFeature specification",
  "signal_to_update": "Vendor-neutral flag evaluation spec (CNCF)",
  "sparse_data_fit": "N/A specification",
  "production_evidence": "CNCF-backed, adopted across vendors",
  "license_observed": "Apache-2.0 per API",
  "maintenance_state": "Very active (stars=1245, archived=false, last_push=2026-08-27)",
  "limitations": "Specification, not an implementation",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-034",
  "repo": "langfuse/langfuse",
  "category": "llm_eval_loop",
  "source_url": "https://github.com/langfuse/langfuse",
  "observed_date": "2026-08-27",
  "evidence_class": "first_party_docs",
  "claim": "🪢 Open source AI engineering platform: LLM evals, observability, metrics, prompt management, playground, datasets. Integrates with OpenTelemetry, LangChain, OpenAI SDK, LiteLLM, and more. 🍊YC W23",
  "signal_to_update": "Traces + scores (human, model, or code) attach to LLM runs; datasets enable regression runs",
  "sparse_data_fit": "Scoring and observability; the rerank step is not automated",
  "production_evidence": "Widely deployed LLM observability platform",
  "license_observed": "LICENSE read: MIT Expat for core, BUT ee/, web/src/ee/, worker/src/ee/ are enterprise-licensed; note copyright now reads ClickHouse, Inc.",
  "maintenance_state": "Very active (stars=33792, archived=false, last_push=2026-08-27)",
  "limitations": "Closes the observe-score loop but does not itself rerank; ownership now ClickHouse",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-035",
  "repo": "Arize-ai/phoenix",
  "category": "llm_eval_loop",
  "source_url": "https://github.com/Arize-ai/phoenix",
  "observed_date": "2026-08-27",
  "evidence_class": "first_party_docs",
  "claim": "AI Observability & Evaluation",
  "signal_to_update": "Traces and evals over LLM runs",
  "sparse_data_fit": "Evaluation tooling",
  "production_evidence": "Real deployments as LLM observability",
  "license_observed": "LICENSE read: Elastic License 2.0 (ELv2) — NOT open source; explicitly forbids providing the software to third parties as a hosted or managed service, and forbids circumventing licence-key functionality",
  "maintenance_state": "Very active (stars=11210, archived=false, last_push=2026-08-27)",
  "limitations": "LICENCE TRAP: ELv2 hosting restriction directly conflicts with offering it inside a hosted client product",
  "disposition": "excluded",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-036",
  "repo": "promptfoo/promptfoo",
  "category": "llm_eval_loop",
  "source_url": "https://github.com/promptfoo/promptfoo",
  "observed_date": "2026-08-27",
  "evidence_class": "observed_behavior",
  "claim": "Test your prompts, agents, and RAGs. Red teaming/pentesting/vulnerability scanning for AI. Compare performance of GPT, Claude, Gemini, DeepSeek, and more. Simple declarative configs with command line and CI/CD integration.  Used by OpenA...",
  "signal_to_update": "Test cases + assertions produce pass/fail and scores per prompt/model variant",
  "sparse_data_fit": "Deterministic eval harness; no automated rerank",
  "production_evidence": "Widely used eval tool",
  "license_observed": "MIT per API",
  "maintenance_state": "Very active (stars=24611, archived=false, last_push=2026-08-27)",
  "limitations": "Evaluation, not a feedback-driven ranking loop",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-037",
  "repo": "confident-ai/deepeval",
  "category": "llm_eval_loop",
  "source_url": "https://github.com/confident-ai/deepeval",
  "observed_date": "2026-08-27",
  "evidence_class": "observed_behavior",
  "claim": "The LLM Evaluation Framework",
  "signal_to_update": "Metric scores over LLM outputs in a pytest-style harness",
  "sparse_data_fit": "Eval only",
  "production_evidence": "Popular eval framework",
  "license_observed": "Apache-2.0 per API",
  "maintenance_state": "Very active (stars=17895, archived=false, last_push=2026-08-26)",
  "limitations": "Scoring only; no closed rerank loop",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-038",
  "repo": "openai/evals",
  "category": "llm_eval_loop",
  "source_url": "https://github.com/openai/evals",
  "observed_date": "2026-08-27",
  "evidence_class": "observed_behavior",
  "claim": "Evals is a framework for evaluating LLMs and LLM systems, and an open-source registry of benchmarks.",
  "signal_to_update": "Eval registry and harness producing scores",
  "sparse_data_fit": "Eval only",
  "production_evidence": "OpenAI-published harness",
  "license_observed": "LICENSE not standard-detected (NOASSERTION); MIT per repo",
  "maintenance_state": "Pushed 2026-04, slowing (stars=19280, archived=false, last_push=2026-04-14)",
  "limitations": "Benchmark harness only",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-039",
  "repo": "braintrustdata/autoevals",
  "category": "llm_eval_loop",
  "source_url": "https://github.com/braintrustdata/autoevals",
  "observed_date": "2026-08-27",
  "evidence_class": "observed_behavior",
  "claim": "AutoEvals is a tool for quickly and easily evaluating AI model outputs using best practices.",
  "signal_to_update": "Scoring functions for LLM outputs",
  "sparse_data_fit": "Scoring primitives only",
  "production_evidence": "OSS component of a commercial eval platform",
  "license_observed": "MIT per API",
  "maintenance_state": "Active (stars=1009, archived=false, last_push=2026-07-29)",
  "limitations": "Library of scorers; the loop is in the closed product",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-040",
  "repo": "wandb/weave",
  "category": "llm_eval_loop",
  "source_url": "https://github.com/wandb/weave",
  "observed_date": "2026-08-27",
  "evidence_class": "observed_behavior",
  "claim": "Weave is a toolkit for developing AI-powered applications, built by Weights & Biases.",
  "signal_to_update": "Traces and evaluations logged for LLM apps",
  "sparse_data_fit": "Eval and tracking",
  "production_evidence": "W&B commercial platform component",
  "license_observed": "Apache-2.0 per API",
  "maintenance_state": "Very active (stars=1121, archived=false, last_push=2026-08-27)",
  "limitations": "Tracking layer; rerank not automated",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-041",
  "repo": "truera/trulens",
  "category": "llm_eval_loop",
  "source_url": "https://github.com/truera/trulens",
  "observed_date": "2026-08-27",
  "evidence_class": "observed_behavior",
  "claim": "Evaluation and Tracking for LLM Experiments and AI Agents",
  "signal_to_update": "Feedback functions score LLM app runs on groundedness/relevance",
  "sparse_data_fit": "Feedback-function framing is closest to a loop, but rerank is manual",
  "production_evidence": "Research-backed, commercial lineage",
  "license_observed": "MIT per API",
  "maintenance_state": "Active (stars=3525, archived=false, last_push=2026-08-27)",
  "limitations": "Scores app runs; no asset ranking mechanism",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-042",
  "repo": "vibrantlabsai/ragas",
  "category": "llm_eval_loop",
  "source_url": "https://github.com/vibrantlabsai/ragas",
  "observed_date": "2026-08-27",
  "evidence_class": "observed_behavior",
  "claim": "Supercharge Your LLM Application Evaluations 🚀",
  "signal_to_update": "Reference-free metrics score RAG pipeline outputs",
  "sparse_data_fit": "Eval only",
  "production_evidence": "Widely adopted RAG eval library",
  "license_observed": "Apache-2.0 per API",
  "maintenance_state": "Pushed 2026-02; ownership moved to vibrantlabsai (stars=15495, archived=false, last_push=2026-02-24)",
  "limitations": "Scoring only; note org transfer from explodinggradients",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-043",
  "repo": "comet-ml/opik",
  "category": "llm_eval_loop",
  "source_url": "https://github.com/comet-ml/opik",
  "observed_date": "2026-08-27",
  "evidence_class": "observed_behavior",
  "claim": "Debug, evaluate, and monitor your LLM applications, RAG systems, and agentic workflows with comprehensive tracing, automated evaluations, and production-ready dashboards.",
  "signal_to_update": "Traces, evals and scores for LLM apps",
  "sparse_data_fit": "Eval and observability",
  "production_evidence": "Comet commercial platform OSS component",
  "license_observed": "Apache-2.0 per API",
  "maintenance_state": "Very active (stars=21631, archived=false, last_push=2026-08-27)",
  "limitations": "Observability; loop closure is manual",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-044",
  "repo": "uptrain-ai/uptrain",
  "category": "llm_eval_loop",
  "source_url": "https://github.com/uptrain-ai/uptrain",
  "observed_date": "2026-08-27",
  "evidence_class": "observed_behavior",
  "claim": "UpTrain is an open-source unified platform to evaluate and improve Generative AI applications. We provide grades for 20+ preconfigured checks (covering language, code, embedding use-cases), perform root cause analysis on failure cases an...",
  "signal_to_update": "Scores LLM outputs and flags regressions",
  "sparse_data_fit": "Eval only",
  "production_evidence": "Small project",
  "license_observed": "Apache-2.0 per API",
  "maintenance_state": "Pushed 2024-08, likely dormant (stars=2363, archived=false, last_push=2024-08-18)",
  "limitations": "Appears dormant",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-045",
  "repo": "benfred/implicit",
  "category": "recommender",
  "source_url": "https://github.com/benfred/implicit",
  "observed_date": "2026-08-27",
  "evidence_class": "first_party_docs",
  "claim": "Fast Python Collaborative Filtering for Implicit Feedback Datasets",
  "signal_to_update": "Implicit feedback (views, clicks) updates matrix factorization via ALS/BPR",
  "sparse_data_fit": "Poor cold start: pure collaborative filtering cannot score an unseen item",
  "production_evidence": "Widely used production recommender library",
  "license_observed": "LICENSE read: MIT full text",
  "maintenance_state": "Active, pushed 2026-05 (stars=3815, archived=false, last_push=2026-05-08)",
  "limitations": "No content features, so no cold start; wrong for a new asset with no history",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-046",
  "repo": "lyst/lightfm",
  "category": "recommender",
  "source_url": "https://github.com/lyst/lightfm",
  "observed_date": "2026-08-27",
  "evidence_class": "first_party_docs",
  "claim": "A Python implementation of LightFM, a hybrid recommendation algorithm.",
  "signal_to_update": "Interactions train embeddings of CONTENT FEATURES; user/item vector is the sum of its feature embeddings",
  "sparse_data_fit": "Good: a brand-new item with known metadata scores immediately at zero interactions; solves per-item sparsity",
  "production_evidence": "Explicit first-party production statement in RecSys 2015 paper: deployed at Lyst over ~8M fashion items, tens of thousands added daily, ANN serving",
  "license_observed": "LICENSE read: Apache-2.0 full text",
  "maintenance_state": "Pushed 2024-07, maintenance-mode (stars=5111, archived=false, last_push=2024-07-24)",
  "limitations": "Still needs aggregate interactions to train feature embeddings; corpus-level sparsity not solved",
  "disposition": "top10",
  "top10_rank": 9,
  "top10_rationale": "Representing an asset only through its features, never an opaque id, is the single most transferable cold-start mechanism here: a newly added shelf asset inherits statistical strength from every asset sharing its tags. Lyst's is the strongest first-party production claim in the recommender category, and the hybrid framing matches Actionist's metadata-rich, interaction-poor regime exactly."
}

{
  "id": "P15-R-047",
  "repo": "NicolasHug/Surprise",
  "category": "recommender",
  "source_url": "https://github.com/NicolasHug/Surprise",
  "observed_date": "2026-08-27",
  "evidence_class": "observed_behavior",
  "claim": "A Python scikit for building and analyzing recommender systems",
  "signal_to_update": "Explicit ratings update MF/neighborhood models",
  "sparse_data_fit": "Poor cold start; explicit ratings assumed",
  "production_evidence": "Teaching and research standard",
  "license_observed": "BSD-3-Clause per API",
  "maintenance_state": "Pushed 2026-05, low churn (stars=6808, archived=false, last_push=2026-05-30)",
  "limitations": "Explicit-rating framing does not match observed build outcomes",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-048",
  "repo": "RUCAIBox/RecBole",
  "category": "recommender",
  "source_url": "https://github.com/RUCAIBox/RecBole",
  "observed_date": "2026-08-27",
  "evidence_class": "observed_behavior",
  "claim": "A unified, comprehensive and efficient recommendation library",
  "signal_to_update": "Unified training of many recommender models",
  "sparse_data_fit": "Research benchmarking",
  "production_evidence": "Academic benchmark suite",
  "license_observed": "MIT per API",
  "maintenance_state": "Pushed 2025-02 (stars=4541, archived=false, last_push=2025-02-24)",
  "limitations": "Benchmark framework, heavy",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-049",
  "repo": "recommenders-team/recommenders",
  "category": "recommender",
  "source_url": "https://github.com/recommenders-team/recommenders",
  "observed_date": "2026-08-27",
  "evidence_class": "observed_behavior",
  "claim": "Best Practices on Recommendation Systems",
  "signal_to_update": "Reference implementations and best practices across recommender algorithms",
  "sparse_data_fit": "Varies by algorithm; includes cold-start discussion",
  "production_evidence": "Microsoft-originated, Linux Foundation project; widely referenced",
  "license_observed": "MIT per API",
  "maintenance_state": "Very active (stars=21865, archived=false, last_push=2026-08-25)",
  "limitations": "Collection of notebooks and utilities, not a service",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-050",
  "repo": "gorse-io/gorse",
  "category": "recommender",
  "source_url": "https://github.com/gorse-io/gorse",
  "observed_date": "2026-08-27",
  "evidence_class": "observed_behavior",
  "claim": "AI powered open source recommender system engine supports classical/LLM rankers and multimodal content via embedding",
  "signal_to_update": "User feedback events update recommendation models automatically in a running service",
  "sparse_data_fit": "Needs event volume",
  "production_evidence": "Self-hostable recommender service with real users",
  "license_observed": "Apache-2.0 per API",
  "maintenance_state": "Very active (stars=9808, archived=false, last_push=2026-08-25)",
  "limitations": "Closed-loop service but built for consumer event volume",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-051",
  "repo": "lenskit/lkpy",
  "category": "recommender",
  "source_url": "https://github.com/lenskit/lkpy",
  "observed_date": "2026-08-27",
  "evidence_class": "first_party_docs",
  "claim": "Python recommendation toolkit",
  "signal_to_update": "Offline evaluation of recommender algorithms",
  "sparse_data_fit": "Research evaluation",
  "production_evidence": "Academic (Boise State) research toolkit",
  "license_observed": "LICENSE read: MIT text with note that data/ has separate terms; NOASSERTION from the note header",
  "maintenance_state": "Active, pushed 2026-08 (stars=314, archived=false, last_push=2026-08-26)",
  "limitations": "Research evaluation focus",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-052",
  "repo": "tensorflow/recommenders",
  "category": "recommender",
  "source_url": "https://github.com/tensorflow/recommenders",
  "observed_date": "2026-08-27",
  "evidence_class": "observed_behavior",
  "claim": "TensorFlow Recommenders is a library for building recommender system models using TensorFlow.",
  "signal_to_update": "Retrieval and ranking model training",
  "sparse_data_fit": "Needs volume",
  "production_evidence": "Google-authored",
  "license_observed": "Apache-2.0 per API",
  "maintenance_state": "Pushed 2026-07 (stars=2027, archived=false, last_push=2026-07-08)",
  "limitations": "Deep models need data",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-053",
  "repo": "NVIDIA-Merlin/Merlin",
  "category": "recommender",
  "source_url": "https://github.com/NVIDIA-Merlin/Merlin",
  "observed_date": "2026-08-27",
  "evidence_class": "observed_behavior",
  "claim": "NVIDIA Merlin is an open source library providing end-to-end GPU-accelerated recommender systems, from feature engineering and preprocessing to training deep learning models and running inference in production.",
  "signal_to_update": "GPU-accelerated recommender training and inference",
  "sparse_data_fit": "Needs very large data",
  "production_evidence": "NVIDIA production-oriented stack",
  "license_observed": "Apache-2.0 per API",
  "maintenance_state": "Pushed 2026-07 (stars=907, archived=false, last_push=2026-07-22)",
  "limitations": "Massive scale assumption; wrong regime",
  "disposition": "excluded",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-054",
  "repo": "PreferredAI/cornac",
  "category": "recommender",
  "source_url": "https://github.com/PreferredAI/cornac",
  "observed_date": "2026-08-27",
  "evidence_class": "observed_behavior",
  "claim": "A Comparative Framework for Multimodal Recommender Systems",
  "signal_to_update": "Multimodal recommender comparison including cold-start-capable hybrids",
  "sparse_data_fit": "Some hybrid models handle cold start",
  "production_evidence": "Academic (Preferred.AI) framework",
  "license_observed": "Apache-2.0 per API",
  "maintenance_state": "Very active (stars=1053, archived=false, last_push=2026-08-11)",
  "limitations": "Research comparison framework",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-055",
  "repo": "ossf/scorecard",
  "category": "asset_scoring",
  "source_url": "https://github.com/ossf/scorecard",
  "observed_date": "2026-08-27",
  "evidence_class": "first_party_docs",
  "claim": "OpenSSF Scorecard - Security health metrics for Open Source",
  "signal_to_update": "18+ deterministic security/health checks produce per-check 0-10, aggregated by documented risk weights (Critical 10, High 7.5, Medium 5, Low 2.5)",
  "sparse_data_fit": "Not applicable: deterministic per-repo heuristics, no learning, no N",
  "production_evidence": "Weekly scan of the 1 million most critical OSS projects, published to a public BigQuery dataset; used by Tensorflow, Angular, Flutter, deps.dev",
  "license_observed": "LICENSE read: Apache-2.0 full text",
  "maintenance_state": "Very active (stars=5654, archived=false, last_push=2026-08-24)",
  "limitations": "Project explicitly disclaims its own aggregate: \"Aggregate scores in particular tells you nothing about what individual behaviors a repository is or is not doing\"; checks are heuristics with false positives",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-056",
  "repo": "ossf/criticality_score",
  "category": "asset_scoring",
  "source_url": "https://github.com/ossf/criticality_score",
  "observed_date": "2026-08-27",
  "evidence_class": "observed_behavior",
  "claim": "Gives criticality score for an open source project",
  "signal_to_update": "Weighted formula over repo signals (age, contributors, commit frequency, dependents) yields a criticality score",
  "sparse_data_fit": "Deterministic formula, no learning",
  "production_evidence": "Google/OpenSSF, run across large repo populations",
  "license_observed": "Apache-2.0 per API",
  "maintenance_state": "Active (stars=1452, archived=false, last_push=2026-08-27)",
  "limitations": "Fixed hand-tuned weights; measures importance not quality or fitness",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-057",
  "repo": "librariesio/libraries.io",
  "category": "asset_scoring",
  "source_url": "https://github.com/librariesio/libraries.io",
  "observed_date": "2026-08-27",
  "evidence_class": "first_party_docs",
  "claim": ":books: The Open Source Discovery Service",
  "signal_to_update": "SourceRank: transparent additive/subtractive rule points over metadata, docs, tests, dependents",
  "sparse_data_fit": "Deterministic rules; works on a single package with no history",
  "production_evidence": "Powered a real public service across many registries",
  "license_observed": "LICENSE read: AGPL-3.0 full text",
  "maintenance_state": "Active (stars=1156, archived=false, last_push=2026-08-05)",
  "limitations": "AGPL; SourceRank rules are hand-tuned heuristics, never learned",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-058",
  "repo": "chaoss/grimoirelab",
  "category": "asset_scoring",
  "source_url": "https://github.com/chaoss/grimoirelab",
  "observed_date": "2026-08-27",
  "evidence_class": "first_party_docs",
  "claim": "GrimoireLab: platform for software development analytics and insights",
  "signal_to_update": "Ingests git/issue/CI activity into community-health metrics",
  "sparse_data_fit": "Descriptive metrics, no learning",
  "production_evidence": "CHAOSS/Bitergia, used for real community dashboards",
  "license_observed": "LICENSE read: GPL-3.0 full text",
  "maintenance_state": "Active, pushed 2026-08; the live CHAOSS tooling now that Augur is archived (stars=621, archived=false, last_push=2026-08-17)",
  "limitations": "GPL-3.0 is a meaningful constraint for a commercial hosted product",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-059",
  "repo": "chaoss/augur",
  "category": "asset_scoring",
  "source_url": "https://github.com/chaoss/augur",
  "observed_date": "2026-08-27",
  "evidence_class": "secondary",
  "claim": "The Augur project is no longer part of CHAOSS. Use CollectOSS instead!",
  "signal_to_update": "Community health metric collection",
  "sparse_data_fit": "No learning",
  "production_evidence": "Was the CHAOSS metrics service",
  "license_observed": "No LICENSE file resolvable via API (404); repo shows NONE",
  "maintenance_state": "ARCHIVED (verified): archived=true, 0 stars; the entire augur-* family under chaoss is archived. GrimoireLab is the live successor (stars=0, archived=true, last_push=2026-07-23)",
  "limitations": "Archived and licence-unresolvable; do not adopt",
  "disposition": "excluded",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-060",
  "repo": "ossf/allstar",
  "category": "asset_scoring",
  "source_url": "https://github.com/ossf/allstar",
  "observed_date": "2026-08-27",
  "evidence_class": "observed_behavior",
  "claim": "GitHub App to set and enforce security policies",
  "signal_to_update": "Enforces security policies on repos, files issues on violation",
  "sparse_data_fit": "Policy enforcement, no learning",
  "production_evidence": "OpenSSF, deployed as a GitHub app",
  "license_observed": "Apache-2.0 per API",
  "maintenance_state": "Active (stars=1444, archived=false, last_push=2026-08-24)",
  "limitations": "Enforcement not scoring",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-061",
  "repo": "google/deps.dev",
  "category": "asset_scoring",
  "source_url": "https://github.com/google/deps.dev",
  "observed_date": "2026-08-27",
  "evidence_class": "first_party_docs",
  "claim": "Resources for the deps.dev API",
  "signal_to_update": "Dependency graph, licence and security metadata across ecosystems via API",
  "sparse_data_fit": "Deterministic metadata, no learning",
  "production_evidence": "Google-run public service with real API traffic",
  "license_observed": "LICENSE read: Apache-2.0 full text",
  "maintenance_state": "Active (stars=438, archived=false, last_push=2026-08-27)",
  "limitations": "Metadata source, not a ranker; useful as a feature supplier",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-062",
  "repo": "google/osv-scanner",
  "category": "asset_scoring",
  "source_url": "https://github.com/google/osv-scanner",
  "observed_date": "2026-08-27",
  "evidence_class": "observed_behavior",
  "claim": "Vulnerability scanner written in Go which uses the data provided by https://osv.dev",
  "signal_to_update": "Matches dependencies against the OSV vulnerability database",
  "sparse_data_fit": "Deterministic matching",
  "production_evidence": "Google-run, widely adopted",
  "license_observed": "Apache-2.0 per API",
  "maintenance_state": "Very active (stars=10926, archived=false, last_push=2026-08-26)",
  "limitations": "Vulnerability signal only; a useful negative feature for asset ranking",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-063",
  "repo": "chaoss/wg-metrics-models",
  "category": "asset_scoring",
  "source_url": "https://github.com/chaoss/wg-metrics-models",
  "observed_date": "2026-08-27",
  "evidence_class": "observed_behavior",
  "claim": "Working Group for Metrics Model",
  "signal_to_update": "Defines composite community-health metrics models",
  "sparse_data_fit": "Methodology definitions",
  "production_evidence": "CHAOSS working group output",
  "license_observed": "MIT per API",
  "maintenance_state": "Pushed 2025-02 (stars=17, archived=false, last_push=2025-02-13)",
  "limitations": "Documentation of metric models, no implementation",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-064",
  "repo": "npms-io/npms-analyzer",
  "category": "registry_ranking",
  "source_url": "https://github.com/npms-io/npms-analyzer",
  "observed_date": "2026-08-27",
  "evidence_class": "first_party_docs",
  "claim": "The analyzer behind https://npms.io",
  "signal_to_update": "Collectors produce quality/popularity/maintenance; scores normalized against whole-corpus min/max/mean through a Bezier curve with control points (0,0),(mean,0.75),(mean,0.75),(1,1)",
  "sparse_data_fit": "Requires the full corpus by construction: population statistics are needed to normalize any single score",
  "production_evidence": "Powered npms.io search ranking; architecture doc published",
  "license_observed": "LICENSE read: MIT full text, npms 2016",
  "maintenance_state": "EFFECTIVELY DEAD: repo archived=false and pushed 2023-02, but the live API returns HTTP 200 with analysis frozen at ~Jan 2023 (lodash analyzedAt 2022-12-03, next pinned at 13.1.2) (stars=330, archived=false, last_push=2023-02-27)",
  "limitations": "TRAP: a live-looking endpoint serving 3.5-year-stale scores; the multi-factor + corpus-relative normalization methodology is the value, not the service",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-065",
  "repo": "rust-lang/crates.io",
  "category": "registry_ranking",
  "source_url": "https://github.com/rust-lang/crates.io",
  "observed_date": "2026-08-27",
  "evidence_class": "observed_behavior",
  "claim": "The Rust package registry",
  "signal_to_update": "Registry search ranking over downloads and metadata",
  "sparse_data_fit": "Deterministic ranking",
  "production_evidence": "Real registry serving the Rust ecosystem",
  "license_observed": "Apache-2.0 per API",
  "maintenance_state": "Very active (stars=3683, archived=false, last_push=2026-08-27)",
  "limitations": "Ranking is a small part of a registry app; download-count-driven",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-066",
  "repo": "pypi/warehouse",
  "category": "registry_ranking",
  "source_url": "https://github.com/pypi/warehouse",
  "observed_date": "2026-08-27",
  "evidence_class": "observed_behavior",
  "claim": "The Python Package Index",
  "signal_to_update": "Registry search relevance",
  "sparse_data_fit": "Deterministic relevance",
  "production_evidence": "Real registry serving Python",
  "license_observed": "Apache-2.0 per API",
  "maintenance_state": "Very active (stars=4137, archived=false, last_push=2026-08-27)",
  "limitations": "Search relevance, not quality ranking",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-067",
  "repo": "npm/cli",
  "category": "registry_ranking",
  "source_url": "https://github.com/npm/cli",
  "observed_date": "2026-08-27",
  "evidence_class": "observed_behavior",
  "claim": "the package manager for JavaScript",
  "signal_to_update": "Registry client; search surfaces registry-side ranking",
  "sparse_data_fit": "No learning client-side",
  "production_evidence": "The npm client itself",
  "license_observed": "LICENSE not standard-detected (NOASSERTION); Artistic-2.0 per repo",
  "maintenance_state": "Very active (stars=10062, archived=false, last_push=2026-08-26)",
  "limitations": "Ranking logic is server-side and not open",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-068",
  "repo": "opendp/opendp",
  "category": "privacy_telemetry",
  "source_url": "https://github.com/opendp/opendp",
  "observed_date": "2026-08-27",
  "evidence_class": "first_party_docs",
  "claim": "The core library of differential privacy algorithms powering the OpenDP Project.",
  "signal_to_update": "Differentially private computations over sensitive data with a formal privacy budget",
  "sparse_data_fit": "Poor at small N: DP noise overwhelms signal in tiny cohorts",
  "production_evidence": "Harvard/OpenDP, used in real statistical disclosure settings",
  "license_observed": "LICENSE read: MIT full text, President and Fellows of Harvard College",
  "maintenance_state": "Very active (stars=433, archived=false, last_push=2026-08-26)",
  "limitations": "DP noise vs tens of builds is a losing trade; the accounting framing still transfers",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-069",
  "repo": "google/differential-privacy",
  "category": "privacy_telemetry",
  "source_url": "https://github.com/google/differential-privacy",
  "observed_date": "2026-08-27",
  "evidence_class": "first_party_docs",
  "claim": "Google's differential privacy libraries.",
  "signal_to_update": "DP aggregation primitives with privacy accounting",
  "sparse_data_fit": "Poor at small N",
  "production_evidence": "Google-authored, used in Google products",
  "license_observed": "LICENSE read: Apache-2.0 full text",
  "maintenance_state": "Very active (stars=3347, archived=false, last_push=2026-08-24)",
  "limitations": "Same small-N tension",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-070",
  "repo": "opendp/smartnoise-sdk",
  "category": "privacy_telemetry",
  "source_url": "https://github.com/opendp/smartnoise-sdk",
  "observed_date": "2026-08-27",
  "evidence_class": "observed_behavior",
  "claim": "Tools and service for differentially private processing of tabular and relational data",
  "signal_to_update": "DP SQL queries and synthetic data",
  "sparse_data_fit": "Poor at small N",
  "production_evidence": "OpenDP/Microsoft collaboration",
  "license_observed": "MIT per API",
  "maintenance_state": "Pushed 2026-07 (stars=302, archived=false, last_push=2026-07-01)",
  "limitations": "Query-layer DP; heavy for this use",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-071",
  "repo": "google/rappor",
  "category": "privacy_telemetry",
  "source_url": "https://github.com/google/rappor",
  "observed_date": "2026-08-27",
  "evidence_class": "first_party_docs",
  "claim": "RAPPOR: Privacy-Preserving Reporting Algorithms",
  "signal_to_update": "Randomized response: Bloom filter + permanent memoization + per-report randomization; server decodes population frequencies by regression",
  "sparse_data_fit": "Terrible at small N: noise cancels only in aggregate; real regime was ~14M respondents/day",
  "production_evidence": "Deployed in Chrome to ~14M opted-in respondents daily to detect settings hijacking (CCS 2014 paper)",
  "license_observed": "LICENSE read via API: Apache-2.0",
  "maintenance_state": "ARCHIVED (verified), pushed 2022-07; repo warns its fastrand module is not cryptographically strong and must not be used in production. Chrome supersession is inferred, not first-party confirmed (stars=869, archived=true, last_push=2022-07-14)",
  "limitations": "Archived precedent only; wrong scale entirely",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-072",
  "repo": "divviup/janus",
  "category": "privacy_telemetry",
  "source_url": "https://github.com/divviup/janus",
  "observed_date": "2026-08-27",
  "evidence_class": "observed_behavior",
  "claim": "Experimental implementation of the Distributed Aggregation Protocol (DAP) specification.",
  "signal_to_update": "Client secret-shares each metric to two non-colluding aggregators; only the population aggregate is reconstructable (Prio/DAP)",
  "sparse_data_fit": "Bad by design at small N: secure aggregation is only meaningful over large populations; small cohorts leak",
  "production_evidence": "Production DAP deployment for Mozilla Firefox live user metrics; Divvi Up names Firefox, Horizontal and Tinfoil",
  "license_observed": "MPL-2.0 per API",
  "maintenance_state": "Very active (stars=70, archived=false, last_push=2026-08-26)",
  "limitations": "Requires two non-colluding operators; needs large cohorts",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-073",
  "repo": "divviup/libprio-rs",
  "category": "privacy_telemetry",
  "source_url": "https://github.com/divviup/libprio-rs",
  "observed_date": "2026-08-27",
  "evidence_class": "observed_behavior",
  "claim": "Implementation of Prio in Rust.",
  "signal_to_update": "Prio VDAF primitives",
  "sparse_data_fit": "Same",
  "production_evidence": "Underlying library for the above deployments",
  "license_observed": "MPL-2.0 per API",
  "maintenance_state": "Very active (stars=122, archived=false, last_push=2026-08-26)",
  "limitations": "Cryptographic primitives",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-074",
  "repo": "divviup/prio-server",
  "category": "privacy_telemetry",
  "source_url": "https://github.com/divviup/prio-server",
  "observed_date": "2026-08-27",
  "evidence_class": "observed_behavior",
  "claim": "A Prio server implementation.",
  "signal_to_update": "Earlier Prio aggregator server",
  "sparse_data_fit": "Same",
  "production_evidence": "Was used for exposure-notification-era aggregation",
  "license_observed": "MPL-2.0 per API",
  "maintenance_state": "ARCHIVED (verified), pushed 2023-07 (stars=73, archived=true, last_push=2023-07-17)",
  "limitations": "Superseded by janus",
  "disposition": "excluded",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-075",
  "repo": "plausible/analytics",
  "category": "privacy_telemetry",
  "source_url": "https://github.com/plausible/analytics",
  "observed_date": "2026-08-27",
  "evidence_class": "first_party_docs",
  "claim": "Open source, privacy-first web analytics. Lightweight, cookie-free Google Analytics alternative. Self-hosted or cloud.",
  "signal_to_update": "Aggregate site analytics without per-user identifiers or cookies",
  "sparse_data_fit": "Aggregate counts",
  "production_evidence": "Real hosted and self-hosted deployments",
  "license_observed": "LICENSE read: AGPL-3.0 full text",
  "maintenance_state": "Very active (stars=28766, archived=false, last_push=2026-08-27)",
  "limitations": "AGPL; product analytics, not asset ranking",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-076",
  "repo": "umami-software/umami",
  "category": "privacy_telemetry",
  "source_url": "https://github.com/umami-software/umami",
  "observed_date": "2026-08-27",
  "evidence_class": "observed_behavior",
  "claim": "Umami is a privacy-first analytics platform. Traffic, campaigns, behavior, conversions, and revenue in one place — no cookies, no surveillance, self-hosted or in the cloud.",
  "signal_to_update": "Privacy-first aggregate web analytics",
  "sparse_data_fit": "Aggregate counts",
  "production_evidence": "Widely self-hosted",
  "license_observed": "MIT per API",
  "maintenance_state": "Very active (stars=38389, archived=false, last_push=2026-08-27)",
  "limitations": "Web analytics framing only",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-077",
  "repo": "opendp/tumult-analytics",
  "category": "privacy_telemetry",
  "source_url": "https://github.com/opendp/tumult-analytics",
  "observed_date": "2026-08-27",
  "evidence_class": "observed_behavior",
  "claim": "Tumult Analytics is a Python library for privately computing aggregate queries on tabular data. It is built atop the Tumult Core library.",
  "signal_to_update": "DP aggregate queries with a privacy budget",
  "sparse_data_fit": "Poor at small N",
  "production_evidence": "Tumult Labs work, now under the opendp org; used in real government/statistical deployments",
  "license_observed": "Apache-2.0 per API",
  "maintenance_state": "ACTIVE (verified), pushed 2026-08, but only 16 stars under opendp/tumult-analytics (stars=16, archived=false, last_push=2026-08-24)",
  "limitations": "Verified alive but low-visibility after the org move",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-078",
  "repo": "IBM/differential-privacy-library",
  "category": "privacy_telemetry",
  "source_url": "https://github.com/IBM/differential-privacy-library",
  "observed_date": "2026-08-27",
  "evidence_class": "observed_behavior",
  "claim": "Diffprivlib: The IBM Differential Privacy Library",
  "signal_to_update": "DP machine learning primitives",
  "sparse_data_fit": "Poor at small N",
  "production_evidence": "IBM research library",
  "license_observed": "MIT per API",
  "maintenance_state": "Pushed 2025-09 (stars=919, archived=false, last_push=2025-09-17)",
  "limitations": "Research library",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-079",
  "repo": "meta-pytorch/opacus",
  "category": "privacy_telemetry",
  "source_url": "https://github.com/meta-pytorch/opacus",
  "observed_date": "2026-08-27",
  "evidence_class": "observed_behavior",
  "claim": "Training PyTorch models with differential privacy",
  "signal_to_update": "DP-SGD: clipped per-sample gradients plus noise during training",
  "sparse_data_fit": "Poor at small N",
  "production_evidence": "Meta, used for DP model training",
  "license_observed": "Apache-2.0 per API",
  "maintenance_state": "Active, pushed 2026-07 (stars=1952, archived=false, last_push=2026-07-13)",
  "limitations": "Training-time DP for neural nets; wrong layer",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-080",
  "repo": "evidentlyai/evidently",
  "category": "drift_monitoring",
  "source_url": "https://github.com/evidentlyai/evidently",
  "observed_date": "2026-08-27",
  "evidence_class": "first_party_docs",
  "claim": "Evidently is ​​an open-source ML and LLM observability framework. Evaluate, test, and monitor any AI-powered system or data pipeline. From tabular data to Gen AI. 100+ metrics.",
  "signal_to_update": "Compares a current data/prediction window against a reference to emit drift and quality metrics; can trigger alerts",
  "sparse_data_fit": "Statistical tests need reasonable window sizes but the reference-vs-current framing works at modest scale",
  "production_evidence": "Widely adopted ML monitoring library with real deployments",
  "license_observed": "LICENSE read: Apache-2.0 full text",
  "maintenance_state": "Very active (stars=7849, archived=false, last_push=2026-08-05)",
  "limitations": "Detects drift; does not itself rerank or decide",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-081",
  "repo": "whylabs/whylogs",
  "category": "drift_monitoring",
  "source_url": "https://github.com/whylabs/whylogs",
  "observed_date": "2026-08-27",
  "evidence_class": "observed_behavior",
  "claim": "An open-source data logging library for machine learning models and data pipelines. 📚 Provides visibility into data quality & model performance over time. 🛡️ Supports privacy-preserving data collection, ensuring safety & robustness. 📈",
  "signal_to_update": "Lightweight statistical profiles of data enable drift comparison without moving raw data",
  "sparse_data_fit": "PROFILE-BASED AGGREGATION IS THE KEY IDEA: profiles are small mergeable sketches, not raw records",
  "production_evidence": "Commercial lineage (WhyLabs), real deployments",
  "license_observed": "Apache-2.0 per API",
  "maintenance_state": "Pushed 2025-01, slowing (stars=2831, archived=false, last_push=2025-01-10)",
  "limitations": "Profiling only; the privacy-friendly mergeable-sketch pattern is the transferable part",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-082",
  "repo": "fivetran/great_expectations",
  "category": "drift_monitoring",
  "source_url": "https://github.com/fivetran/great_expectations",
  "observed_date": "2026-08-27",
  "evidence_class": "observed_behavior",
  "claim": "Always know what to expect from your data.",
  "signal_to_update": "Declarative expectations validate data and fail loudly when violated",
  "sparse_data_fit": "Deterministic assertions, works at any N",
  "production_evidence": "Very widely deployed in production data pipelines",
  "license_observed": "Apache-2.0 per API",
  "maintenance_state": "Very active; note ownership now fivetran (stars=11742, archived=false, last_push=2026-08-27)",
  "limitations": "Validation not learning; useful as an incident signal source",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-083",
  "repo": "deepchecks/deepchecks",
  "category": "drift_monitoring",
  "source_url": "https://github.com/deepchecks/deepchecks",
  "observed_date": "2026-08-27",
  "evidence_class": "secondary",
  "claim": "Deepchecks: Tests for Continuous Validation of ML Models & Data. Deepchecks is a holistic open-source solution for all of your AI & ML validation needs, enabling to thoroughly test your data and models from research to production.",
  "signal_to_update": "Suites of data and model checks produce pass/fail plus metrics",
  "sparse_data_fit": "Checks work at modest N",
  "production_evidence": "Real adoption for ML validation",
  "license_observed": "LICENSE not standard-detected (NOASSERTION); repo indicates AGPL-3.0 for parts",
  "maintenance_state": "Pushed 2025-12 (stars=4047, archived=false, last_push=2025-12-28)",
  "limitations": "Mixed licence; validation not ranking",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-084",
  "repo": "SeldonIO/alibi-detect",
  "category": "drift_monitoring",
  "source_url": "https://github.com/SeldonIO/alibi-detect",
  "observed_date": "2026-08-27",
  "evidence_class": "secondary",
  "claim": "Algorithms for outlier, adversarial and drift detection",
  "signal_to_update": "Drift, outlier and adversarial detectors over data streams",
  "sparse_data_fit": "Detectors need windows; some work at modest N",
  "production_evidence": "Seldon commercial lineage, real MLOps deployments",
  "license_observed": "LICENSE not standard-detected (NOASSERTION); Business Source Licence applies to parts of Seldon's stack",
  "maintenance_state": "Pushed 2025-12 (stars=2548, archived=false, last_push=2025-12-11)",
  "limitations": "Licence needs care; detection not ranking",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-085",
  "repo": "NannyML/nannyml",
  "category": "drift_monitoring",
  "source_url": "https://github.com/NannyML/nannyml",
  "observed_date": "2026-08-27",
  "evidence_class": "observed_behavior",
  "claim": "nannyml: post-deployment data science in python",
  "signal_to_update": "Estimates model performance WITHOUT ground truth labels, then detects silent failure",
  "sparse_data_fit": "Directly relevant idea: estimating outcome quality when labels are delayed or missing",
  "production_evidence": "Commercial OSS with real users",
  "license_observed": "Apache-2.0 per API",
  "maintenance_state": "Pushed 2025-07 (stars=2150, archived=false, last_push=2025-07-12)",
  "limitations": "Tabular ML focus; performance estimation assumptions may not transfer",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-086",
  "repo": "IFCA-Advanced-Computing/frouros",
  "category": "drift_monitoring",
  "source_url": "https://github.com/IFCA-Advanced-Computing/frouros",
  "observed_date": "2026-08-27",
  "evidence_class": "observed_behavior",
  "claim": "Frouros: an open-source Python library for drift detection in machine learning systems.",
  "signal_to_update": "Concept and data drift detection algorithms",
  "sparse_data_fit": "Streaming drift detectors",
  "production_evidence": "Academic (IFCA) library",
  "license_observed": "BSD-3-Clause per API",
  "maintenance_state": "Pushed 2026-04 (stars=261, archived=false, last_push=2026-04-23)",
  "limitations": "Small research library",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-087",
  "repo": "sodadata/soda-core",
  "category": "drift_monitoring",
  "source_url": "https://github.com/sodadata/soda-core",
  "observed_date": "2026-08-27",
  "evidence_class": "observed_behavior",
  "claim": "Data Contracts engine for the modern data stack. https://www.soda.io",
  "signal_to_update": "Data quality checks in a declarative language",
  "sparse_data_fit": "Deterministic checks",
  "production_evidence": "Commercial data-quality product OSS core",
  "license_observed": "LICENSE not standard-detected (NOASSERTION); Apache-2.0 per repo",
  "maintenance_state": "Very active (stars=2417, archived=false, last_push=2026-08-26)",
  "limitations": "Data quality only",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-088",
  "repo": "huggingface/trl",
  "category": "edit_learning",
  "source_url": "https://github.com/huggingface/trl",
  "observed_date": "2026-08-27",
  "evidence_class": "observed_behavior",
  "claim": "Train transformer language models with reinforcement learning.",
  "signal_to_update": "Preference pairs (chosen vs rejected) directly update a policy via DPO; also PPO/RLHF from a reward model",
  "sparse_data_fit": "Preference-pair framing is sample-efficient relative to RL, but LLM training still needs thousands of pairs",
  "production_evidence": "The standard OSS implementation behind many real preference-tuned models",
  "license_observed": "Apache-2.0 per API",
  "maintenance_state": "Very active (stars=19158, archived=false, last_push=2026-08-27)",
  "limitations": "Trains model weights, not an asset ranker; the accept/reject pair as the atomic learning unit is what transfers",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-089",
  "repo": "continuedev/continue",
  "category": "edit_learning",
  "source_url": "https://github.com/continuedev/continue",
  "observed_date": "2026-08-27",
  "evidence_class": "observed_behavior",
  "claim": "open-source coding agent",
  "signal_to_update": "Telemetry can log accept/reject of AI code suggestions",
  "sparse_data_fit": "Accept/reject at the edit level is a real production signal, but the OSS product does not close it into a rerank",
  "production_evidence": "Widely installed AI coding extension with documented telemetry",
  "license_observed": "Apache-2.0 per API",
  "maintenance_state": "Very active (stars=35647, archived=false, last_push=2026-08-26)",
  "limitations": "Logs the signal; no demonstrated closed rerank loop in the OSS repo",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-090",
  "repo": "dyad-sh/dyad",
  "category": "edit_learning",
  "source_url": "https://github.com/dyad-sh/dyad",
  "observed_date": "2026-08-27",
  "evidence_class": "first_party_docs",
  "claim": "Local, open-source AI app builder for power users ✨ v0 / Lovable / Replit / Bolt alternative 🌟 Star if you like it!",
  "signal_to_update": "Local AI app builder; the closest analogue to Actionist's own build surface",
  "sparse_data_fit": "No demonstrated learning loop over build outcomes",
  "production_evidence": "Popular OSS AI app builder",
  "license_observed": "LICENSE read: Apache-2.0 core BUT everything under src/pro/ is separately licensed",
  "maintenance_state": "Very active (stars=21322, archived=false, last_push=2026-08-27)",
  "limitations": "Mixed licence; valuable as a domain analogue, not as a learning mechanism",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-091",
  "repo": "LAION-AI/Open-Assistant",
  "category": "edit_learning",
  "source_url": "https://github.com/LAION-AI/Open-Assistant",
  "observed_date": "2026-08-27",
  "evidence_class": "observed_behavior",
  "claim": "OpenAssistant is a chat-based assistant that understands tasks, can interact with third-party systems, and retrieve information dynamically to do so.",
  "signal_to_update": "Human rankings of assistant replies train a reward model",
  "sparse_data_fit": "Crowd ranking at scale; the ranking-to-reward-model pipeline is documented",
  "production_evidence": "Large real crowdsourcing effort, now concluded",
  "license_observed": "Apache-2.0 per API",
  "maintenance_state": "Concluded, pushed 2024-08 (stars=37408, archived=false, last_push=2024-08-17)",
  "limitations": "Project ended; useful as a documented ranking-collection precedent",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-092",
  "repo": "argilla-io/argilla",
  "category": "edit_learning",
  "source_url": "https://github.com/argilla-io/argilla",
  "observed_date": "2026-08-27",
  "evidence_class": "observed_behavior",
  "claim": "Argilla is a collaboration tool for AI engineers and domain experts to build high-quality datasets",
  "signal_to_update": "Human feedback and annotations collected then exported to train or evaluate",
  "sparse_data_fit": "Designed for small curated datasets, which suits sparse regimes",
  "production_evidence": "Real adoption for data curation; now part of Hugging Face",
  "license_observed": "Apache-2.0 per API",
  "maintenance_state": "Very active (stars=5085, archived=false, last_push=2026-08-24)",
  "limitations": "Collection and curation layer; the learning happens downstream",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-093",
  "repo": "HumanSignal/label-studio",
  "category": "edit_learning",
  "source_url": "https://github.com/HumanSignal/label-studio",
  "observed_date": "2026-08-27",
  "evidence_class": "observed_behavior",
  "claim": "Label Studio is a multi-type data labeling and annotation tool with standardized output format",
  "signal_to_update": "Human labels on any data type feed downstream training",
  "sparse_data_fit": "Works at small N by design",
  "production_evidence": "Very widely deployed labeling platform",
  "license_observed": "Apache-2.0 per API",
  "maintenance_state": "Very active (stars=28156, archived=false, last_push=2026-08-27)",
  "limitations": "Labeling infrastructure; no ranking mechanism",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-094",
  "repo": "vivekjoshy/openskill.py",
  "category": "bayesian_preference",
  "source_url": "https://github.com/vivekjoshy/openskill.py",
  "observed_date": "2026-08-27",
  "evidence_class": "first_party_docs",
  "claim": "Multiplayer Rating System. No Friction.",
  "signal_to_update": "Comparison outcomes update per-item Gaussian rating (mu, sigma) with explicit uncertainty",
  "sparse_data_fit": "Excellent for sparse: designed to rank from few games; sigma encodes confidence directly",
  "production_evidence": "Python implementation of published Weng-Lin Bayesian rating models",
  "license_observed": "LICENSE read: MIT full text",
  "maintenance_state": "Active, pushed 2026-05 (stars=368, archived=false, last_push=2026-05-12)",
  "limitations": "Rating models assume comparison structure; needs outcome pairs",
  "disposition": "top10",
  "top10_rank": 8,
  "top10_rationale": "It delivers TrueSkill-class Bayesian ranking with per-item uncertainty under a clean MIT licence and no patent restriction, which is the practical escape from the TrueSkill commercial trap this survey found. The mu/sigma pair is directly usable as both a ranking score and an exploration signal for rarely-used shelf assets."
}

{
  "id": "P15-R-095",
  "repo": "splitio/javascript-client",
  "category": "experimentation",
  "source_url": "https://github.com/splitio/javascript-client",
  "observed_date": "2026-08-27",
  "evidence_class": "observed_behavior",
  "claim": "NodeJS and Browser SDK client for Split Software",
  "signal_to_update": "Flag/experiment assignment with impression tracking",
  "sparse_data_fit": "No learning in SDK",
  "production_evidence": "Commercial experimentation vendor SDK",
  "license_observed": "Apache-2.0 per API",
  "maintenance_state": "Active (stars=49, archived=false, last_push=2026-08-04)",
  "limitations": "SDK only; the statistics live server-side and closed",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-096",
  "repo": "ecosyste-ms/packages",
  "category": "asset_scoring",
  "source_url": "https://github.com/ecosyste-ms/packages",
  "observed_date": "2026-08-27",
  "evidence_class": "first_party_docs",
  "claim": "An open API service providing package, version and dependency metadata of many open source software ecosystems and registries.",
  "signal_to_update": "Aggregates package, version and dependency metadata across many registries into an open API",
  "sparse_data_fit": "Deterministic metadata, no learning; usable for a single package",
  "production_evidence": "Runs as a real public API service across many ecosystems",
  "license_observed": "LICENSE read: AGPL-3.0 full text",
  "maintenance_state": "Active, pushed 2026-08 (stars=108, archived=false, last_push=2026-08-25)",
  "limitations": "AGPL-3.0 is a real constraint for a commercial hosted product; metadata supplier not a ranker",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}

{
  "id": "P15-R-097",
  "repo": "Homebrew/homebrew-formula-analytics",
  "category": "registry_ranking",
  "source_url": "https://github.com/Homebrew/homebrew-formula-analytics",
  "observed_date": "2026-08-27",
  "evidence_class": "observed_behavior",
  "claim": "💀 Query Homebrew's analytics from the command-line (deprecated)",
  "signal_to_update": "Opt-in install/usage telemetry aggregated into public package analytics",
  "sparse_data_fit": "Aggregate counts, no learning",
  "production_evidence": "Real opt-in telemetry across the Homebrew user base; a genuine consent-based precedent",
  "license_observed": "No LICENSE file (NONE per API)",
  "maintenance_state": "ARCHIVED (verified): archived=true, pushed 2025-02, description marked deprecated (stars=17, archived=true, last_push=2025-02-09)",
  "limitations": "Archived and unlicensed; valuable only as an opt-in telemetry precedent",
  "disposition": "census",
  "top10_rank": null,
  "top10_rationale": null
}