Clean-Room derivation of 195 controls from BSI QUAIDAL (10 criteria + 15
building blocks + 30 measures + 140 metrics) for EU AI Act Art. 10
training-data quality compliance.
- ingest_bsi_quaidal.py parses YAML frontmatter into a structural index
(no protected prose stored on disk).
- derive_quaidal_mcs.py rewrites each entry via local LLM (qwen3.5:35b-a3b)
with a hard 4-gram plagiarism gate < 20%; achieved mean overlap 0.5%.
- Migration 011 adds compliance.derived_controls table with full source
provenance (framework, section, url, commit SHA, license note).
- apply_quaidal_to_db.py UPSERTs YAML into DB.
- Source repo (legal-sources/bsi-quaidal/) gitignored.
Same pattern as IACE module DIN-reference handling: name the norm and
section, never quote.
Backed by BSI license clarification 2026-05: § 5 UrhG anwendbar,
share:true im Frontmatter; Clean-Room derivation is the safe path.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Marketing-Website
- Neue SavingsSection auf Homepage: "Compliance entdeckt sechsstellige
Einsparungen". Pitch-Position der Cookie-Audit-Cost-Optimization-Story
fuer DAX-Konzern-Sales (BMW-Case-Style: 90 Vendors -> 25 nach
Konsolidierung, EUR 500k-3M / Jahr).
- /savings-scan: Kostenloser 5-Min-Saving-Scan-Form (URL + E-Mail).
Form-Submit ist Placeholder, soll an Compliance-Backend gehaengt werden.
- /savings-methodik: 4-Stufen-Erklaerung der Cookie-Tier-Inferenz +
ehrliche Caveats (Listpreise != Vertragspreise, Media-Spend nicht
enthalten) + Datenquellen.
- Content-de + Content-en in content.ts beide um savings-Block ergaenzt
und Section-Numerierung angepasst (03=Savings, 04=Deterministic).
- LOC-Split: savings-Inhalte (DE+EN, ~100 LOC) in content.savings.ts
ausgelagert damit content.ts unter 500-LOC-Hard-Cap bleibt.
Control-Pipeline
- LESSONS-LEARNED-mc-check-types.md fuer die parallele CRA-MC-Generation.
Erklaert die TEXT/PROCESS/REVIEW-Klassifikation die im Compliance-Repo
retrofitted wurde. Verhindert dass CRA-MCs denselben Defekt bekommen.
Mapping-Heuristik fuer verification_method -> check_type, plus
Backfill-Workflow fuer ~62 ambiguous Eintraege.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- 20 manually verified golden controls with expected MC topics
- Structural quality tests: min 10K MCs, max 300/MC, no orphans
- Doc-check controls tests: 8 doc types covered, no empty questions
- Quality thresholds: 90% accuracy, enforced by regression tests
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Splits master controls >200 members by re-clustering their object groups
with k=4-20 per group. First round: 38 groups → 325 sub-groups → 253 new MCs.
25 generic MCs remain (monitoring, procedure, etc.) — need regulation-source split.
Session summary: Block F complete, Control Generation (1,599+), Pass 0a/0b,
Production Sync, G-pre1/2/3 Object Clustering + Master Controls + API,
G1-G4 Compliance Execution Layer (Decision Trace, Commit Ledger, Decision Memory,
Pre-Deployment Enforcement).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
New table: deployment_checks (verdict, blocking/warning controls, risk score)
New API:
POST /v1/deployment-checks (SDK asks: "can I deploy?")
GET /v1/deployment-checks/{id} (check result)
POST /v1/deployment-checks/{id}/override (manual override with justification)
GET /v1/deployment-checks/stats (approval/block rate)
Check logic: queries G1 decision_traces + G3 open failures per affected control.
Verdict: approved (0 blocking) or blocked (with fix recommendations).
454 tests pass, 0 regressions.
Block G complete: G1-G4 all implemented.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
New table: decision_events (assessment→decision→fix→verification→failure cycle)
New API:
POST /v1/decision-events (record lifecycle event)
GET /v1/decision-events (list with filters)
GET /v1/decision-events/timeline/{control_id} (full chronological timeline)
GET /v1/decision-events/stats (failure rate, cycle times)
Each event captures input_state, output_state, actor, evidence.
454 tests pass, 0 regressions.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
New table: compliance_commits (commit hash, affected controls, risk level)
New API:
POST /v1/compliance-commits (SDK registers commit + impact)
GET /v1/compliance-commits (list with filters)
GET /v1/compliance-commits/by-control/{id} (all commits for a control)
GET /v1/compliance-commits/stats (dashboard)
GET /v1/compliance-commits/{id} (detail)
GIN index on affected_control_ids for fast @> containment queries.
454 tests pass, 0 regressions.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
New table: decision_traces (status, reason, evidence, fix plan per control)
New API:
POST/GET/PUT /v1/decision-traces (CRUD for decisions)
GET /v1/decision-traces/stats (compliance dashboard)
GET /v1/controls/{id}/full-trace (Regulation→Obligation→Control→Decision→Evidence)
454 tests pass, 0 regressions.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
G-pre1: 144k objects clustered into 7,466 groups via Mini-Batch K-Means
on bge-m3 embeddings. Two-stage: k=5000 base + sub-cluster groups >50.
G-pre2: 5,114 Master Controls from lifecycle phase chains
(define→implement→test→monitor), linking 172,504 atomic controls.
G-pre3: REST API for Master Controls
GET /v1/master-controls (list, search, filter)
GET /v1/master-controls/stats
GET /v1/master-controls/{mc_id} (detail with phase-controls)
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
8 tests confirm all REGULATION_LICENSE_MAP, ACTION_TYPES, _NEGATIVE_PATTERNS,
_ACTION_SYNONYMS, and _OBJECT_SYNONYMS entries are correctly migrated to DB.
Dicts kept as fallback for DB-unavailability resilience.
Block F complete: F1-F5 all done.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Uses Ollama (qwen3.5:35b-a3b, think:false) to generate additional
German synonyms for action types and object tokens. Results stored
with source='llm' in action_synonyms/object_synonyms tables.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Registers /generate/run-pass0a and /generate/pass0a-status/{job_id}
on the core control-pipeline (port 8098). Previously Pass 0a was only
available on the compliance backend which connects to Production DB,
causing a split-brain when controls are generated locally.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Migrates ACTION_TYPES (26+8 types), _NEGATIVE_PATTERNS (22), _ACTION_SYNONYMS
(65), and _OBJECT_SYNONYMS (75) from hardcoded dicts to DB tables.
- SQL migration: 003_action_object_ontology.sql (3 tables)
- Migration scripts: f2_migrate_actions.py (34 types, 145 synonyms), f3_migrate_objects.py (75 objects)
- OntologyRegistry cache: 5min TTL, raises RuntimeError if empty (safe fallback to dicts)
- control_ontology.classify_action/get_phase delegate to DB with dict fallback
- control_dedup.normalize_action/normalize_object delegate to DB with dict fallback
- 25 new tests, 446 total pass, 0 regressions
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Next: F1 Regulation Registry (DB + API + Frontend + Auto-Create)
Frontend at /sdk/regulation-registry in breakpilot-compliance admin
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
_SECTION_NUMBER_RE only had patterns for §/Art/Section/Kapitel/Annex
but missed NIST-style identifiers (AC-1, GV.OC-01, 3.1, A01:2021).
This caused 0% section rate for all NIST/BSI/ENISA documents even
though sections were correctly detected — the section NUMBER wasn't
extracted from the header.
Also adds:
- reupload_legal_strategy.py: re-upload with legal chunking
- extract_and_upload_nist.py: local PDF extraction workaround
- qdrant-snapshot.sh: backup mechanism for Qdrant collections
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Complete instructions for next session including:
- Current quality metrics per document type
- Prioritized action items (NIST fix, citation backfill, missing laws)
- Full Block E-G roadmap with details
- All critical files, DB state, test commands
- Known issues (3 lost NIST PDFs, frontend 500s, D5 script safety)
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Session achieved: structural metadata end-to-end (D2-D4), overlap bug
fix, HTML stripping with charset detection, 430/436 docs re-ingested.
Remaining: ~40 EU Official Journal PDFs need HTML from EUR-Lex (broken
multi-column PDF extraction), 3 missing EDPB PDFs, 1 corrupt PDF.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
EU Official Journal PDFs (AI Act, CRA, NIS2, DSGVO, etc.) use
multi-column layouts that pypdf breaks into fragmented words
("Ar tik el" instead of "Artikel"). pdfplumber handles these correctly.
Backend priority: unstructured > pdfplumber > pypdf (auto mode).
Also increases D5 re-ingestion timeout to 3600s for large PDFs.
58 embedding-service tests passing. pdfplumber: MIT license.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
HTML files from gesetze-im-internet.de were decoded as raw UTF-8, keeping
<div>/<p> tags intact. The legal chunker regex requires § at line start,
which never matched inside HTML tags → 0% section metadata for HTML docs.
Fix: detect HTML content and strip tags before sending to embedding
service. Block elements become newlines, entities are decoded.
§ signs now appear at line starts → section detection works.
Also adds D5 re-ingestion scripts (reingest_d5.py + config) for
batch re-processing of all documents in Qdrant collections.
27 rag-service tests passing.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
D2: RAG service stores section/section_title/paragraph/paragraph_num/page
from embedding service chunks_with_metadata into Qdrant payloads.
D3: Control generator prefers section > article > section_title from
Qdrant, adds page to source_citation and generation_metadata.
D4: Validated with real BGB §§ 312-312k text. Found and fixed critical
bug where Phase 3 overlap destroyed the [§ ...] section prefix, causing
only the first chunk per document to have metadata. All subsequent
chunks lost section info.
Also fixes pre-existing lint issues (unused imports, ambiguous variable
names, duplicate dict key, bare except).
456 tests passing (58 embedding + 387 pipeline + 11 rag-service).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Sends 67k review candidates to Haiku Batch API in pairs.
Each pair gets a DUPLIKAT/VERSCHIEDEN decision with reasoning.
Results stored in control_dedup_reviews.review_status.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Stores last_control_id in canonical_generation_jobs after each page.
On restart, resumes from checkpoint instead of starting over.
Checkpoint is deleted on completion.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Paginated DB queries (100 rows/page) instead of loading all 166k rows
- Individual timeout (30s) per embedding + qdrant call
- Per-control try/except — one failure doesn't kill the job
- Sequential processing (no asyncio.gather) for stability
- Progress logging every 500 controls
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
153k of 160k merge groups have only 1 control — no intra-group
dedup possible. Skip them in Phase 1, they become masters automatically.
Phase 2 (cross-group) still checks them via Qdrant embeddings.
Reduces Phase 1 from ~96h to ~2h.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Prevents duplicate batch submissions that caused ~$170 in extra costs.
Refuses new submit if a batch was submitted in the last 10 minutes.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Remove applicability/scanner_hint/evidence_type/provides_context from
Pass 0b prompt to reduce output tokens (~40% less). These 6 fields are
added via cheap Haiku backfill afterwards (~$1.50 per 10k controls).
Saves ~$200 over the remaining 160k obligations.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Prompt v4 adds 6 new fields to Pass 0b output:
- applicability: condition rules (same format as dependency engine)
- check_type: expanded to 10 granular types
- scanner_hint: search_terms + negative_indicators for MCP
- manual_review_required_if: escalation conditions
- evidence_type: code/process/hybrid
- provides_context: context variables this control creates
New endpoint POST /generate/backfill-extended:
- Backfills existing 9k controls via Haiku Batch API (~$1.50)
- Adds all 6 new fields to generation_metadata
- Supports dry_run mode
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
LLM merge_key phases (e.g. "submission") don't always match PHASE_ORDER
keys. Derive phase order from action_type via get_phase_order() instead.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Fix truncated title detection: only flag near-200-char titles or mid-word cutoffs
- Fix evidence leak detection: check title start patterns, not keyword substring
("nachweisen" verb is valid action, "Nachweis vorliegen" is evidence)
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Benjamin Admin