MOCK PDF — Annex IV §1–§9 placeholder
=====================================

NOTE: text-fixture mock. See `crates/silver-annex-iv-v2/MOCK.md`.

Model:        fixture-vertical-model-B
Vertical:     legal summarization (high-risk, Annex III §6 (law-enforcement))
Generated:    2026-05-04 (mock)
Spec ref:     EU AI Act Art. 11 + Annex IV; Art. 26 (deployer obligations)

§1 — General description
------------------------
Vertical model "fixture-vertical-model-B" is a domain-specialised
fine-tune of fixture-foundation-model-A targeting legal-document
summarisation. Marketed to law-firm clients in EU jurisdictions.
Classified high-risk under Annex III §6.

§2 — Detailed description
-------------------------
§2(d) Datasheets: legal-summary fine-tune corpus declared in
       `legal-summary-pii-leak.json`. Note: this fixture intentionally
       contains a residual T6 PII-leak refutation surfaced via the
       erasure-completeness check (see `/vault/erasure/<id>`).
§2(e) Validation/testing: ROUGE-L 0.41; human evaluation on 200 EU
       case-law summaries; deployer-side calibration wizard in app.

§3 — Monitoring, functioning and control
----------------------------------------
Deployer-side prompt logging; PII-redaction wrapper; firm-level
access control.

§4 — Performance metrics
------------------------
ROUGE-L: 0.41
Hallucination rate (claim-level): 1.8%
PII-leak rate (post-redaction): < 0.01% (target); residual cases
flagged in incident replay ledger.

§5 — Risk-management system
---------------------------
Annex IV §5 + Art. 9 risk-management process; post-market monitoring
per Art. 72.

§6 — Lifecycle changelog
------------------------
2026-04-15  Fine-tune completed.
2026-04-30  Deployer rollout to 12 EU law firms.

§7 — Harmonised standards applied
---------------------------------
ISO/IEC 23894:2023; ISO/IEC 42001:2023.

§8 — Declaration of conformity
------------------------------
DoC reference: DOC-2026-MOCK-002.

§9 — Post-market monitoring plan
--------------------------------
Continuous incident replay; quarterly erasure-completeness re-prove
on the deployer's residual cu