.gitignore

# Python
__pycache__/
*.pyc
*.pyo
*.pyd
*.egg
*.egg-info/
pip-log.txt
pip-delete-this-directory.txt

# Build / distribution
dist/
dist-*/
build/
*.whl
*.tar.gz

# Optional infrastructure working state
**/.terraform/
*.tfstate
*.tfstate.*
*.tfplan
*.tfvars
*.tfvars.json
crash.log
crash.*.log

# Test / Coverage
.pytest_cache/
.hypothesis/
.coverage
.coverage.*
coverage.xml
htmlcov/
.tox/
.nox/
junit.xml
pytest-report.xml
test-results/
tests/validation/fixtures/
docs/accuracy.md
.ci-audit-requirements.txt

# Agent working files
/.agent/
/.grok/

# Local validation sweeps. Evaluated-target names and per-domain JSON output
# live here during accuracy work; never committed. Anything committed uses
# explicit synthetic identities under reserved namespaces or safe aggregates.
# Validation artifacts using real targets stay local.
# Generic tooling (runner, gap finder, diff, triage) is committed; the
# corpus and run outputs stay local so users can curate their own
# private test bed without leaking targets.
/validation/*
!/validation/audit_fingerprints.py
!/validation/run_corpus.py
!/validation/find_gaps.py
!/validation/diff_runs.py
!/validation/triage_candidates.py
!/validation/scan.py
!/validation/catalog_baseline.py
# Frozen, aggregate-only agent-portability preregistration and its network-free
# validator. Client transcripts, hidden reasoning, and run records remain under
# the ignored validation/agent-portability-local/ workspace.
!/validation/agent_portability_contract.py
!/validation/prepare_agent_portability_evaluation.py
!/validation/2026-08-14-agent-portability-preflight.md
# Generic catalog-round frame preparer; no target data is embedded.
!/validation/prepare_catalog_round.py
# Prior-sample drift contract and aggregate evaluator. Prior frames, retained
# results, comparison sidecars, and per-namespace differences remain private.
!/validation/prepare_catalog_drift_round.py
!/validation/evaluate_catalog_drift_round.py
# Generic private rank-band sampler and shared ranked-source primitives. Source,
# key, selected rows, and manifests remain under the ignored private roots.
!/validation/prepare_catalog_rank_frame.py
# Generic private ccTLD-delegation sampler. Official-source mappings, prior
# exclusions, keys, selected rows, and manifests remain private.
!/validation/prepare_catalog_region_frame.py
!/validation/prepare_catalog_region_sources.py
!/validation/ranked_sampling.py
!/validation/stratify_catalog_round.py
# Generic vendor-seed contract and aggregate evaluator. Provider-controlled
# sources, target membership, and per-domain results remain private.
!/validation/archive_vendor_seed_sources.py
!/validation/prepare_vendor_seed_round.py
!/validation/evaluate_vendor_seed_round.py
!/validation/summarize_ct_sessions.py
!/validation/corpus-example.txt
!/validation/README.md
# v1.9: publicly-reproducible calibration scripts. Both operate
# against either the synthetic-network distribution or a user-supplied
# corpus; neither contains target data.
!/validation/synthetic_calibration.py
!/validation/likelihood_sensitivity.py
!/validation/analyze_v19_calibration.py
# Differential verification of the inference core: independent full-joint
# reference, no target data (synthetic network only). Imported by the test.
!/validation/differential_verification.py
# Adversarial-property check: the suppression-monotonicity theorem as an
# exhaustive machine-checked invariant over the shipped network. No target
# data (network dataclasses only). Imported and gated by the test.
!/validation/adversarial_properties.py
# PV2 inference drift gate: deterministic CPT-implied marginals of the shipped
# network. No corpus / company data (node names + numbers only). Gated by the test.
!/validation/drift_check.py
!/validation/inference_baseline.json
# Credible-interval perturbation-coverage harness + memo: synthetic worlds only,
# aggregate numbers, no target data. Imported and gated by the test.
!/validation/interval_coverage.py
!/validation/interval-coverage.md
# Mutation-gate memo: scope, kill-set rationale, score floor, survivor
# dispositions. Counts and node ids only, no target data.
!/validation/mutation-gate.md
# Reference-calibration harness + memo: pure label/Wilson/calibration logic plus a
# maintainer-local network orchestration that prints aggregates only. The
# committed files carry no target data (synthetic worked example only). The pure
# logic is unit-tested.
!/validation/calibration_estimators.py
!/validation/reference_calibration.py
!/validation/reference-calibration.md
# Shared resolve-sweep progress helper: a throttled stderr heartbeat for the
# network harnesses (counts only, no target data). Pure async logic, unit-tested.
!/validation/progress.py
# Conformal-coverage harness: distribution-free split-conformal coverage on the
# labelable nodes (pure logic plus a maintainer-local orchestration that prints
# aggregates only). No target data; the pure logic is unit-tested.
!/validation/conformal_coverage.py
# Aggregate calibration memo renderer: reads private-run aggregate JSON, applies
# disclosure checks, and emits a reviewable aggregate memo. No target data.
!/validation/render_calibration_memo.py
# Private calibration bundle runner: orchestrates the reference, tenancy, and
# conformal harnesses into a gitignored run directory. No target data.
!/validation/run_calibration_bundle.py
# Tenancy reference-calibration harness: corroborates the M365 tenancy posterior
# (DNS channel only) against the provider's identity-endpoint attestation; GWS
# reported one-sided. Pure label/channel-split logic is unit-tested; the
# orchestration is maintainer-local and prints aggregates only. No target data.
!/validation/tenancy_reference_calibration.py
# Power analysis for the frozen product-quality decision rule: pure arithmetic
# (exact Clopper-Pearson bounds by binomial-CDF bisection), no network, no
# corpus, no third-party dependency. Reproduces the sample-size tables in
# docs/quality-baseline-preregistration.md. No target data.
!/validation/quality_power_analysis.py
# Network-free structural audit for the frozen v2.11 M365 arms. Enumerates
# abstract evidence-role states only and reads no corpus or target data.
!/validation/quality_arm_identifiability.py
!/validation/2026-08-13-quality-arm-identifiability.md
# Aggregate-only live product-quality characterization. The runner is public;
# its corpus and generated reports remain under the private roots below.
!/validation/characterize_live_quality.py
!/validation/2026-08-12-stable-v1-live-characterization.md
# Private evaluation-frame preparer. The generic HMAC sampler is public; source,
# key, membership, and collection outputs remain under the private roots below.
!/validation/prepare_quality_evaluation_frame.py
# Layer-ablation harness + memo: fully synthetic (model-sampled worlds, planted
# partitions, fictional hostnames), deterministic, publishable. No target data.
!/validation/layer_ablation.py
!/validation/layer-ablation.md
# Posture-distribution harness + memo: entropy-reduction and interval-width
# distributions stratified by observable posture. Pure aggregation logic is
# unit-tested; the orchestration is maintainer-local and prints aggregates
# only (quantiles, bucket means). No target data.
!/validation/posture_distributions.py
!/validation/public-list-calibration.md
!/validation/reproduce_paper_numbers.py
!/validation/2026-06-19-paper-reproduction-smoke.md
!/validation/2026-06-28-hybrid-interval-smoke.md
!/validation/2026-06-28-hybrid-interval-paper.md
!/validation/2026-06-28-adversarial-perturbation-paper.md
!/validation/2026-06-28-final-claim-audit.md
!/validation/2026-06-29-final-claim-audit-refresh.md
!/validation/2026-06-29-scorecard-gate-claim-audit.md
!/validation/2026-06-29-submission-freeze-local-proof.md
!/validation/2026-06-30-submission-freeze-local-proof.md
!/validation/2026-06-19-crowdstrike-match-mode.md
!/validation/2026-06-19-supabase-cname-target.md
# Full-corpus calibration run (2026-06-23): aggregate-only memo rendered through
# the disclosure-checked renderer. No apexes, org names, tenant IDs, or per-domain
# rows; strata below 10 domains suppressed.
!/validation/2026-06-23-full-corpus-calibration.md
!/validation/2026-06-28-full-corpus-calibration-refresh.md
!/validation/2026-06-19-msedge-triad-motif.md
# Exchange Online DKIM suffix-match validation (2026-06-26): aggregate counts
# only, no apexes/org names/tenant IDs/per-domain rows.
!/validation/2026-06-26-onmicrosoft-suffix-match.md
# Typed DNS catalog baseline (2026-07-17): aggregate counts and generic
# provider patterns only. No target names, target records, tenant IDs, or rows.
!/validation/2026-07-17-typed-catalog-baseline.md
# Regional catalog result (2026-08-13): aggregate counts, commitments,
# and provider-controlled references only. No selected namespaces or rows.
!/validation/2026-08-13-catalog-regional-round.md
# Vendor-seed catalog result (2026-08-14): disclosure-safe provider counts,
# exact reducer output, commitments, and no target rows or record values.
!/validation/2026-08-14-catalog-vendor-seed-round.md
!/validation/2026-08-14-catalog-vendor-seed-aggregate.json
# Prior-sample catalog drift result (2026-08-14): disclosure-safe aggregate
# outcomes and commitments only. No target names, records, candidates, or rows.
!/validation/2026-08-14-catalog-drift-round.md
!/validation/2026-08-14-catalog-drift-aggregate.json
# Phase 1 product-quality baseline (2026-08-05): generated by
# scripts/quality_scorecard.py from the registered MCP surface and the shipped
# catalog. Network-free and corpus-free, so it contains protocol byte counts and
# catalog counts only. No target names, records, tenant IDs, or rows.
!/validation/2026-08-05-quality-baseline-scorecard.md
# Per-pattern base-rate reducer and its aggregate memo (2026-07-25):
# provider-owned patterns and aggregate counts only; the reducer never
# reads target-identifying fields and its output stays in ignored paths.
!/validation/pattern_base_rates.py
!/validation/slug_cooccurrence.py
!/validation/2026-07-25-pattern-base-rates.md
# v2.1a: aggregate-state methodology example. Fully synthetic (numbered
# reserved identities, fabricated numbers); the reference reducer, generator, and synthetic
# cohort are public. Real cohorts stay local. The reducer run-output is a
# regenerable artifact (the doc shows how), so it stays out of the repo.
!/validation/aggregate/
/validation/aggregate/example_output.json
# v1.9.2.2: LLM-assisted triage runner for fingerprint candidates.
# Reads private candidates.json (gitignored) and emits generic
# YAML stanzas that are safe to commit. The runner itself is
# generic catalog-growth tooling.
!/validation/triage_llm.py
# Per-release corpus-validation reports: statistics + technical
# narrative; no per-domain data. The full result JSONs
# (v1.8-validation-results*.json, v1.9-validation-results*.json, ...)
# stay private under the /validation/* glob.
!/validation/v1.8-validation-summary.md
!/validation/v1.9-validation-summary.md
!/validation/v1.9.3-calibration.md
!/validation/v1.9.4-calibration.md
!/validation/analyze_v19_4_hardened.py
!/validation/v1.9.5-stability.md
!/validation/compute_node_stability.py
!/validation/v1.9.6-stability-update.md
!/validation/v1.9.8-metadata-audit.md
!/validation/v1.9.9-detection-gap-ux.md
!/validation/v1.9.9-corpus-run.md
!/validation/threshold_sensitivity.md
!/validation/threshold_sensitivity.py
!/validation/coverage_gap_audit.md
!/validation/performance_baseline.md
!/validation/invariant_audit.md
!/validation/agentic_ux_v199_runbook.md
!/validation/v1.9.10-pre-lock.md
!/validation/v1.9.10-bayesian-revalidation.md
!/validation/v1.9.10-mutation-status.md
!/validation/v2.0-prep-baseline.md
!/validation/v2.0-corpus-run-runbook.md
!/validation/v2.0-corpus-run.md
!/validation/v1.9.11-trend-table.md
!/validation/v2.0-validation-summary.md
!/validation/v2.0-release-notes-draft.md
!/validation/v2.0-lock-ceremony.md
!/validation/v1.9.24-candidates-triage.md
!/validation/v1.9.26-schema-contract.md
!/validation/2026-06-02-calibration.md
!/validation/v1.9.87-c2-corpus-batch.md
!/validation/2026-06-04-case-study-spot-check.md
!/validation/cal8-likelihood-sensitivity.md
!/validation/cal14-missingness-design.md
!/validation/corpus_aggregator.py
!/validation/synthetic_corpus/
!/validation/synthetic_corpus/**
/validation/synthetic_corpus/**/__pycache__/
# Adversarial DNS-record corpus (2026-08-17): hand-authored .invalid record sets,
# the build_source_result helper, the end-to-end runner, and its aggregate-only
# results. No real target data by construction; every value is a reserved-namespace
# or documented-vendor-reference record string.
!/validation/adversarial_corpus/
!/validation/adversarial_corpus/**
/validation/adversarial_corpus/**/__pycache__/
!/validation/2026-08-17-adversarial-corpus-round.md
# validation/__init__.py: added in v1.9.2 to make ``validation`` a
# proper package (previously implicit namespace) so static analysis
# resolves intra-package imports cleanly.
!/validation/__init__.py
# v1.9.2 agentic UX validation harness. Dev-only tooling. Fixtures use
# explicit synthetic identities under the reserved .invalid namespace. Detailed reports,
# transcripts, raw records, and any private-corpus outputs must stay under
# runs/ or local/ inside the harness, both of which are ignored below.
!/validation/agentic_ux/
!/validation/agentic_ux/**
# Re-ignore Python build/cache artifacts and any per-run output dirs
# inside the harness even though the parent is whitelisted above.
/validation/agentic_ux/**/__pycache__/
/validation/agentic_ux/runs/
/validation/agentic_ux/local/
!/validation/v1.9.2-agentic-ux.md
# User-curated private corpus + run outputs. Never commit.
/validation/corpus-private/
/validation/runs-private/
/validation/live_runs/
/validation/local/

# Per-company live JSON dumps from ad-hoc validation runs. Real
# company names belong in /validation/ (already gitignored above) or
# discarded after analysis. These patterns belt-and-suspenders the
# repo root so a stray `recon foo.com --json > foo.json` never
# accidentally lands in a commit.
/*.com.json
/*.org.json
/*.net.json
/*.io.json
/*.so.json
/*.gov.json
/*.edu.json
/*.co.json
/sc-*.json
/v1_*_dive/
/v1_*_validation*/

# Linting / Type checking
.ruff_cache/
.mypy_cache/
.dmypy.json
.pyre/
.pytype/
.pyright/

# Mutation-testing session state and any other tool-specific
# scratch files. Tools that need session state should write into
# .mutmut-cache/, .cosmic-ray/, or similar. Never the repo root.
# These globs catch the ones that ignore the convention.
cosmic-ray*.sqlite
cosmic-ray*.toml
.mutmut-cache/
.cosmic-ray/
.hypothesis/

# Stray scratch files that occasionally land in root from
# interrupted shell sessions or crashed tools. Each pattern matches
# garbage that should not be committed. (.stackdump, .DS_Store,
# Thumbs.db are already covered by the OS-artefact section below.)
core
*.core
nohup.out
test-output.txt
test-output-*.txt
scratch.py
scratch-*.py
_scratch*
tmp/
_tmp/

# IDE / editors
.vscode/
.vs/
.idea/
*.swp
*.swo
*~

# Merge / rebase artifacts
*.orig
*.rej

# Jupyter
.ipynb_checkpoints/

# direnv
.envrc
.direnv/

# Claude Code local settings
.claude/
CLAUDE.md
CLAUDE.local.md

# OS
.DS_Store
Thumbs.db
Desktop.ini
$RECYCLE.BIN/
*.stackdump

# Env
.env
.env.local
.env.*.local
.venv/
.venv-*/
venv/
env/

# Credentials. Defensive rules. A passive-recon tool should never
# commit keys or credentials, even local test ones. Prefer
# RECON_CONFIG_DIR or a secrets manager over dropping files in-tree.
*.pem
*.key
*.p12
*.pfx
*.jks
id_rsa
id_rsa.*
id_dsa
id_dsa.*
id_ecdsa
id_ecdsa.*
id_ed25519
id_ed25519.*
credentials.json
secrets.yaml
secrets.yml

# Logs
*.log
/logs/

# Kiro
.kiro/
/archive
docs/archive/
_local_gate.sh

# Mutation-testing session DB (regenerable; the config mutation.toml is committed).
/mutation.sqlite