You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
docs/guides/iso_soc2_deployer_guide.md: false claim — evaluation.require_human_approval: true for every high-risk/unacceptable run — 'F-compliance-110 raises ConfigError if you forget.' (DOC-D2B03-003) #90
Wave 0 · theme D0-5: Documented controls that are not implemented · highest severity Critical · 1 finding(s) · effort docs fix
Summary
Operators relying on the claimed guard can run high-risk models through to final_model/ with no Article 14 human-oversight gate and no error.
Primary finding
DOC-D2B03-003 — docs/guides/iso_soc2_deployer_guide.md: false claim — evaluation.require_human_approval: true for every high-risk/unacceptable run — 'F-compliance-110 raises ConfigError if you forget.'
The document says.evaluation.require_human_approval: true for every high-risk/unacceptable run — 'F-compliance-110 raises ConfigError if you forget.'
What the code does. F-compliance-110 (_enforce_safety_gate_for_strict_tier, called from _warn_high_risk_compliance) raises ConfigError only when evaluation.safety is disabled. It never checks require_human_approval and does not even warn about it; the only strict-tier warning concerns auto_revert. Reproduced: risk_classification high-risk or unacceptable with safety enabled and require_human_approval=false is accepted.
User impact. Operators relying on the claimed guard can run high-risk models through to final_model/ with no Article 14 human-oversight gate and no error.
Proposed correction. "Set evaluation.require_human_approval: true for every risk_classification ∈ {high-risk, unacceptable} run. ForgeLM does NOT enforce this: F-compliance-110 only refuses these tiers when evaluation.safety.enabled is false. Enforce human approval in your config review or a CI policy check."
Turkish mirror. the Turkish page has the same error; fix both in the same PR
Acceptance criteria
Each statement is re-checked against the code at fix time.
Documentation listed above is corrected in English and in its Turkish mirror in the same PR.
Every finding in this issue is resolved or explicitly closed with a reason.
The release note names the contract this fix restores.
Found by the October 2026 ForgeLM code and documentation review against commit cab2947. Code links point at that commit; line numbers may have moved since. Finding IDs (CR-…, DOC-…, TECH-…) are stable references for this backlog.
Activity
docs: compliance and privacy wording, SBOM and Article 12/14, and the…