Skip to main content
Glama

achlab

Analysis of Competing Hypotheses with auditable provenance. Full documentation in Portuguese: README.pt-BR.md.

Status: M2 — release candidate. CLI, read-only MCP server and 53 tests.

The design rule is verify, don't generate: every evidence item carries a citable source, every assessment carries a rationale, and a report with incomplete provenance is refused (exit 3) unless it is explicitly stamped NÃO CITÁVEL.

Install (development)

python3 -m venv .venv
.venv/bin/pip install -e ".[dev]"
.venv/bin/ach demo

Related MCP server: MCP Cobranza

Workflow

ach init "Who disclosed the memo?" --dir case-001
ach hyp add "H1: an insider did it" --dir case-001
ach hyp add "H2: an external contractor did it" --dir case-001
ach ev add --title "Badge log" --source "Synthetic badge log" \
  --reliability A --credibility 2 --relevance alta \
  --excerpt "22:14 access, the only one after 20:00" \
  --locator "fixture://badge/2026-01-01" --dir case-001
ach assess E1 H1 --consistency ++ --rationale "..." --dir case-001
ach assess E1 H2 --consistency -- --rationale "..." --dir case-001
ach ev accept E1 --dir case-001
ach score --dir case-001
ach report --dir case-001 --md case-001/report.md
ach validate --dir case-001

Exit codes: 0 ok · 1 operational error · 2 integrity/schema violation · 3 report refused for incomplete provenance.

MCP (read-only)

pip install "achlab[mcp]"
ach-mcp --root ~/analyses

Tools: ach_list_projects, ach_get_project, ach_score, ach_report. None of them writes; the analysis is written through the CLI, with a human decision.

Documentation

Test

.venv/bin/python -m pytest

53 tests: matrix, tie, non-diagnostic evidence, decisive-evidence search, absolute-margin floor, integrity validation, CLI, MCP and symlink containment.

License

MIT. Contributions: CONTRIBUTING.md. Security: SECURITY.md.

Gate de segredo (pre-commit)

Todo clone precisa ativar o hook uma vez — core.hooksPath é config local e não viaja com o checkout (worktrees compartilham, porque leem o mesmo config do repo):

git config core.hooksPath .githooks

.githooks/pre-commit roda gitleaks com as regras de .gitleaks.toml sobre o conteúdo staged e falha fechado: sem o binário no PATH ou sem o .toml, o commit para. Bypass consciente é git commit --no-verify, com o motivo registrado no commit.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      Not graded
      quality
      B
      maintenance
      Enables deterministic analysis of a single Agent tool-catalog snapshot and comparison of two snapshots through read-only JSON-RPC tools, reporting digests, schema statistics, duplicates, collisions, budget checks, and diffs.
      Apache 2.0
    • F
      license
      Not graded
      quality
      B
      maintenance
      Provides read-only tools for analyzing a collections portfolio, including summaries, aging, account details, cashflow, and collector performance, with on-demand HTML reporting via natural language.
      -
    • A
      license
      C
      quality
      B
      maintenance
      Provides read-only, stdio-based MCP tools for inspecting locally captured C/C++ fault artifacts, grouping processes and sessions, querying the evidence store, and resolving explicitly supported RPC/IPC relationships. Access is restricted to a configured artifact root, with observed, static-candidate, and unresolved evidence kept strictly separate.
      24
      1
      Apache 2.0
    • A
      license
      B
      quality
      B
      maintenance
      Enables any MCP client to plan and run verifiable analysis of engine hot-fire and bench test data through tools that open runs, segment phases, check sensor health, redlines, valve response and oscillations, compare simulations, verify every reported number against an evidence ledger, and write HTML reports. All findings stay traceable and reproducible from deterministic tool calls rather than raw samples.
      5
      26 PyPI
      MIT