qualidade Ainteresse Sconf. medium
A substantial public legal-data infrastructure project that preserves Brazilian judicial communications, keeps provenance explicit across heterogeneous official sources, and exposes the resulting corpus through a public site, reconstructible DuckDB/Parquet data and a read-only MCP surface. Execution earns A because the archival pipeline, source separation, operational workflows, test discipline and recent maintenance are already unusually concrete. Interest reaches S because the combination of preservation, reproducibility, legal provenance and agent-readable access turns fragile judicial publications into reusable civic infrastructure. S quality is withheld because coverage remains intentionally incomplete and source-dependent, the stable public MCP endpoint is not yet proven, and some archive/publishing operations still have explicit external-credential and rollout dependencies.
2estrelas
3forks
Pythonlinguagem
01 de out. de 2026atualizado
Ver ficha do projeto
O que sustenta
- The data model preserves epistemic boundaries instead of flattening sources: archived publications, procedural state and decision text remain distinct, with provenance and source-specific freshness/coverage visible to consumers.
- The public archive is designed for reuse rather than only for the site: original ZIPs, manifests and Parquet derivatives live in the Internet Archive, while `catalog.sql` reconstructs the DuckDB view layer instead of shipping an opaque database artifact.
- The same underlying evidence is exposed through three useful product surfaces: a public search/dossier site, directly queryable public data, and a read-only MCP with product-level tools for publications, procedural state and decisions plus operational health/freshness diagnostics.
- Operational engineering is extensive and domain-aware: scheduled collection, backlog draining, manifest rendering, Parquet consolidation, catalog updates, deployment and canary workflows encode real archival failure modes such as distinguishing transient DJEN 403s, confirmed absence and malformed availability responses.
- Recent maintenance demonstrates active correctness work rather than cosmetic churn: current main includes review-driven fixes to legal segmentation labels, corpus governance checks and a provably equivalent pruning optimization that reduced a real corpus-scale near-duplicate scan from multi-minute hangs to roughly one minute without changing results.
O que segura
- Coverage is explicitly heterogeneous and incomplete across DJEN, TJRO JURIS, STJ and DataJud; the project correctly exposes this, but national completeness and equal maturity across sources remain unsolved product limitations.
- The repository documents the public HTTP MCP artifact but still withholds a stable public URL pending the required smoke proof, so the agent-facing product is operational locally before it is fully proven as a public service.
- A current archive-publishing handoff remains blocked on Internet Archive credentials, showing that part of the preservation pipeline still depends on external operational state that cannot be proven solely from repository code and CI.
- The supported product and the experimental Lab/segmenter surface coexist in one large repository; governance separates them explicitly, but the breadth of pipelines, workflows and experimental work creates a material maintenance and integration burden.
Histórico do tier
- 2026-09-24: initial placement -> quality A / interest S, medium confidence after reviewing main at 4709458a, product/architecture documentation, packaging and CLI/MCP contracts, archival/deployment workflows, current PR state and recent correctness/performance maintenance; S quality withheld for incomplete source coverage, the unproven stable public MCP endpoint and external archive-publishing dependencies, while interest is S for the unusually reusable fusion of public preservation, provenance-aware legal data, reconstructible analytics and agent access.
qualidade Ainteresse Sconf. medium
Operational open-data infrastructure for Brazil's public CNPJ corpus: Ficha combines a browser-native DuckDB-WASM explorer with a versioned public data contract and large snapshots preserved on Internet Archive. Execution is strong enough for A, while interest reaches S because the project turns a normally centralized public-data workload into reusable static civic infrastructure. S quality is withheld because the publication/ETL surface remains operationally complex, the current public manifest still selects the 2026-07 snapshot, and important identity/history work remains under active consolidation.
0estrelas
0forks
Pythonlinguagem
19 de set. de 2026atualizado
Ver ficha do projeto
O que sustenta
- The public contract is concrete rather than aspirational: the current manifest selects a 2026-07 snapshot and exposes multi-gigabyte Parquet relations with hashes and Internet Archive URLs, while the README documents both analytical and atomic access paths.
- The architecture has unusual leverage: DuckDB-WASM lets the browser query Parquet directly, while the same preserved artifacts and manifest can be reused by external analysis without depending on the Ficha UI or an application backend.
- Engineering discipline is broad for a civic-data project: separate web, ETL and ficha-py gates; Linux and Windows ETL tests; Ruff and pytest; Astro check/test/build; benchmark smokes; deploy, accessibility and visual-evidence workflows; ADRs; and fail-closed integrity work around checkpoints and key uniqueness.
- Maintenance is active and material, with September work on monthly key uniqueness, published-page accessibility, resilient remote-failure states, visual evidence, and safer PR-based automated manifest publication.
O que segura
- The current public manifest still selects snapshot 2026-07 as of this review. This is not treated as proof of failure, but it leaves freshness and publication cadence as an operational question rather than a fully settled strength.
- The ETL and publication system has a large operational surface, including many specialized workflows and long-running multi-gigabyte jobs; keeping provenance, checkpoint integrity, remote Internet Archive behavior and public-manifest updates coherent is a material maintenance burden.
- Some data-model questions remain under active investigation or RFC, notably canonical identity/cardinality for socios and derived cross-snapshot signals, so parts of the broader historical model are still consolidating.
- A fresh independent full-suite/site run and direct live Pages probe could not be reproduced in this review because Jatobá timed out and the fallback sandbox had no outbound DNS.
Histórico do tier
- 2026-09-24: initial placement -> quality A / interest S, medium confidence after reviewing main at 98c5c110, the public architecture and README, current manifest, verified monthly data releases, CI/test and workflow surfaces, recent hardening commits, and open investigation/RFC work; S quality withheld for ongoing operational/data-model consolidation, while interest is S for the unusually reusable browser-native plus preserved-open-data architecture.
qualidade Ainteresse Aconf. medium
Mature and unusually coherent OKF infrastructure: a strict authored-core parser and validator with relational Ibis/DuckDB projections, graph, CLI, MCP, GitHub Action and an optional Rust execution path. It earns A for both execution and interest because the project is already operationally broad, carefully specified and highly reusable, while S is withheld because the package is still beta and several open integration stacks are still consolidating core write, relation and native-engine contracts.
0estrelas
0forks
Pythonlinguagem
02 de out. de 2026atualizado
Ver ficha do projeto
O que sustenta
- The architecture makes a clear semantic boundary between strict authored OKF and source adapters, keeping one canonical downstream model instead of teaching every consumer each source dialect.
- The public surface is unusually complete for a beta infrastructure project: Python API, CLI, MCP, GitHub Action, DuckDB/Ibis relations, NetworkX projection and an optional Rust engine with documented fallback semantics.
- Release engineering is substantive rather than decorative: SemVer/changelog coupling, locked Python/Rust packaging, release dry-runs, OIDC-based PyPI publishing and a current v0.45.10 release are all present.
- Current maintenance remains active and technically meaningful, including FastMCP 4/Pydantic 2.12 migration and follow-up work on deterministic materialization, typed writes and native-engine routing.
O que segura
- The project is still explicitly beta and below 1.0, so the public contract remains subject to meaningful evolution.
- Maintaining parity across portable Python/TypeScript surfaces and the optional Rust core creates a significant integration and conformance burden; several open/stacked PRs show that this boundary is still being actively consolidated.
- A full local test run could not be independently reproduced in this review because Jatobá timed out and the fallback sandbox had no external DNS; the assessment therefore relies on repository structure, release artifacts, workflow definitions, current main and open development evidence.
Histórico do tier
- 2026-09-24: initial placement -> quality A / interest A, medium confidence after reviewing main at 50e1e91, README/architecture, package and release contracts, CI/release workflows, the v0.45.10 release and the current open development stack; S withheld for beta status, cross-runtime parity burden and still-active core consolidation.
qualidade Ainteresse Sconf. medium
A legally rigorous, unusually auditable system for reviewing Rondônia public-pension rules: Sisprev separates frozen source material, authored OKF rules and legal analysis, deterministic derived deployment artifacts, browser/PDF review surfaces and institutional approval states. Execution earns A because traceability, domain-specific regression gates and review evidence are already substantial. Interest reaches S because the project makes legal reasoning and operational rule maintenance coexist in one versioned evidence chain without pretending that CI is an act of legal authority. S quality is withheld for an aging Cycle 9 branch stack that still needs a clean restack onto current main, unresolved external confirmation of some Sisprev runtime semantics, and a few explicitly documented structural coverage gaps.
0estrelas
0forks
HTMLlinguagem
29 de set. de 2026atualizado
Ver ficha do projeto
O que sustenta
- The source-of-truth model is disciplined: received CSV material is preserved as a frozen baseline, the live rule catalog and legal analysis are authored in OKF Markdown, and operational CSVs, indexes and site snapshots are regenerated rather than hand-edited.
- CI protects concrete failure modes rather than generic ceremony: bundle conformance, exact regeneration of derived artifacts, frozen-source hashes, cycle-specific deployment loads, decision/spec consistency, append-only findings and high-cardinality rule-family regressions are checked independently.
- The project has unusually strong review evidence for a legal-audit repository: a generated site and institutional PDFs, commit-linked artifacts, real-browser visual checks, mobile overflow/reading-scale regressions, and semantic accessibility checks were added after observed defects rather than inferred from build success.
- Its epistemic and institutional boundaries are explicit: repository analysis, audit completion, implementation confirmation, PGE validation, IPERON approval and signature are modeled as distinct states, while detectors and textual extraction are treated as evidence that still requires substantive legal review.
- Recent maintenance is substantive: late-August legal/model corrections were followed by September work that made responsive rendering, keyboard/focus accessibility and published-surface evidence reproducible rather than leaving them as informal assumptions.
O que segura
- The open Cycle 9 stack no longer anchors cleanly to contemporary main and has an explicit restack plan; until that work is reconciled, a large body of already-developed audit work remains structurally separated from the current branch.
- Some operational semantics still require evidence from IPERON or the vendor, notably how the running Sisprev captures cause information and what particular calculation labels execute; the repository correctly records these as external confirmation gaps rather than filling them by inference.
- Known structural debts remain explicit, including the lack of a fully structured cycle-closure state/date contract and incomplete mechanical coverage of historical files under data/homologacao/.
- A fresh independent full-suite run could not be reproduced in this review because the Jatobá compute endpoint timed out; the assessment therefore relies on current main, repository contracts, committed workflows, recent merged PR evidence and open issue/PR state.
Histórico do tier
- 2026-09-24: initial placement -> quality A / interest S, medium confidence after reviewing main at d2828406, README/agent contracts, OKF source and derivation model, incident-driven CI, recent legal and browser-evidence work, and the current Cycle 9/external-confirmation debt; S quality withheld for unresolved integration and evidence gaps, while interest is S for the unusually reusable fusion of versioned legal audit, executable rule data, provenance and institutional-state separation.