breakpilot-compliance

Author SHA1 Message Date

Author	SHA1	Message	Date
Benjamin Admin	c5b22e0c99	fix: derive intake flags from DETECTED SERVICES, not from text content Fundamental architecture fix: data processing happens through APIs/scripts/ cookies — NOT through visible page text. A news site about healthcare does NOT process health data. Before: Qwen reads website text → guesses "health_data: true" (WRONG) After: Google Analytics detected → tracking: true (CORRECT, deterministic) New flow: detect services from HTML → map service categories to flags → feed flags into UCCA assessment. No LLM needed for flag extraction. SERVICE_TO_FLAGS maps categories: tracking→tracking, marketing→marketing+ third_party_sharing, payment→payment_data, heatmap→profiling, etc. SPECIFIC_SERVICE_FLAGS for Klarna (Art.22), Stripe (US transfer), etc. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>	2026-05-02 08:37:51 +02:00
Benjamin Admin	0f3ec9061e	fix: false positive findings + restore docs-src + §312k ecommerce filter 1. Intake prompt: "BETREIBER verarbeitet" statt "Text erwaehnt". IHK berichtet ueber Gesundheitsdaten → false. Vorher: true. 2. §312k Check: nur bei E-Commerce/Abo-Websites (Warenkorb, Shop, PayPal etc.) IHK hat keine Vertraege → kein Kuendigungsbutton noetig. 3. docs-src/ restored from commit `9824304` Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>	2026-05-02 08:26:59 +02:00
Benjamin Admin	4298ae17ab	feat: Phase 0+1 — LLM intake extraction + control relevance filter Phase 0: Qwen extracts 14 structured intake flags (personal_data, marketing, profiling, ai_usage, etc.) instead of keyword matching. Fallback to keywords if LLM unavailable. Flags feed into UCCA for accurate scoring. Phase 1: Control relevance filter removes false positives. C_TRANSPARENCY only recommended if AI/ML keywords found in text. 7 control rules with keyword lists + intake flag fallback. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>	2026-04-29 11:36:24 +02:00

Benjamin Admin

c5b22e0c99

fix: derive intake flags from DETECTED SERVICES, not from text content

Fundamental architecture fix: data processing happens through APIs/scripts/
cookies — NOT through visible page text. A news site about healthcare does
NOT process health data.

Before: Qwen reads website text → guesses "health_data: true" (WRONG)
After: Google Analytics detected → tracking: true (CORRECT, deterministic)

New flow: detect services from HTML → map service categories to flags →
feed flags into UCCA assessment. No LLM needed for flag extraction.

SERVICE_TO_FLAGS maps categories: tracking→tracking, marketing→marketing+
third_party_sharing, payment→payment_data, heatmap→profiling, etc.
SPECIFIC_SERVICE_FLAGS for Klarna (Art.22), Stripe (US transfer), etc.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

2026-05-02 08:37:51 +02:00

Benjamin Admin

0f3ec9061e

fix: false positive findings + restore docs-src + §312k ecommerce filter

1. Intake prompt: "BETREIBER verarbeitet" statt "Text erwaehnt".
   IHK berichtet ueber Gesundheitsdaten → false. Vorher: true.
2. §312k Check: nur bei E-Commerce/Abo-Websites (Warenkorb, Shop, PayPal etc.)
   IHK hat keine Vertraege → kein Kuendigungsbutton noetig.
3. docs-src/ restored from commit 9824304

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

2026-05-02 08:26:59 +02:00

Benjamin Admin

4298ae17ab

feat: Phase 0+1 — LLM intake extraction + control relevance filter

Phase 0: Qwen extracts 14 structured intake flags (personal_data,
marketing, profiling, ai_usage, etc.) instead of keyword matching.
Fallback to keywords if LLM unavailable. Flags feed into UCCA for
accurate scoring.

Phase 1: Control relevance filter removes false positives.
C_TRANSPARENCY only recommended if AI/ML keywords found in text.
7 control rules with keyword lists + intake flag fallback.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

2026-04-29 11:36:24 +02:00

3 Commits