Architecture Signature tooling with explicit claim boundaries.

Reading order

ArchSig is the observation and reporting layer for AAT / SFT. Start with the minimal scan path, then read command and artifact references before using reports for PR review or SFT forecasting.

• Getting Started Fixture scan, validation, repository scan, and Python import graph scan.
• Commands CLI reference for scan, reports, policy, dataset, operational feedback, repair, and SFT forecast commands.
• Artifacts Sig0, validation, snapshots, diff reports, AIR, Feature Extension Report, policy decision, datasets, and SFT forecasting artifacts.
• Boundaries Rules for measured zero, unmeasured axes, theorem promotion, non-conclusions, and lawfulness claims.
• Workflows First adoption, PR review, generated-change provenance, scheduled scan, architecture dynamics, and SFT forecast workflow.
• Schemas Schema version policy, compatibility outcomes, catalog families, and migration guardrails.
• Operational Feedback B10 daily ledger, calibration, threshold, ownership, repair adoption, incident correlation, and hypothesis refresh artifacts.
• Examples Good extension, hidden interaction witness, semantic witness, report reading, and dynamics reading.
• Roadmap B0-B13 phases, current implementation snapshot, and standardization targets.

Product surfaces

ArchSig is easier to use when the implemented product surface is separated from research, calibration, and adapter gaps. The same CLI family exposes four surfaces with different evidence and claim boundaries.

• ArchSig Core Lean and Python import graph scan, Sig0 validation, snapshots, and signature diffs. Remaining gaps include call graph, data dependency, dynamic import, plugin loading, and framework semantics adapters.
• ArchSig Review AIR, theorem precondition checks, Feature Extension Report, policy decision, PR comment summary, and baseline suppression. Policy calibration and arbitrary invariant judgment remain review-layer work, not Lean theorem output.
• ArchSig SFT Markdown PRD / Spec / Issue / AI proposal input can produce bounded forecast artifacts and a ConsequenceEnvelope. GitHub Issue JSON, AI proposal JSON, real dataset calibration, and framework semantics adapters remain gaps.
• ArchSig Operational PR history datasets, outcome linkage, daily ledgers, calibration records, threshold policies, ownership monitors, repair adoption, incident correlation, and hypothesis refresh artifacts support operational learning without causal theorem status.

Tool scope

The CLI scans Lean import graphs and Python import graphs, computes structural signature axes, applies optional policy and runtime evidence, and writes JSON artifacts for later validation, diffing, reporting, and forecasting commands.

scan
  -> validate
  -> snapshot
  -> signature-diff
  -> AIR
  -> theorem-check
  -> feature-report
  -> policy-decision
  -> pr-comment

Reference map

Website pages are summaries. The repository documents remain the source of truth for command behavior, schema details, artifact boundaries, and mutable roadmap status.

• CLI and artifact references tools/archsig/README.md, tools/archsig/docs/commands.md, and tools/archsig/docs/artifacts-and-boundaries.md.
• Tooling theory boundary docs/tool/README.md, docs/tool/claim_boundary.md, docs/tool/signature_artifacts.md, and docs/tool/reports.md.
• Operational and roadmap status tools/archsig/docs/operational-feedback.md and docs/tool/roadmap.md.

Claim discipline

Architecture Signature is a multi-axis diagnostic, not a scalar quality score. Every report must be read together with measurement status, coverage gaps, theorem preconditions, and non-conclusions.