{"doctrine":{"version":"v11","counts":"749/14/163","locked_theorems":["F1","F4","F7","F11","F12","F18","F19","F22"],"locked_count":8,"kernel_commit":"c7c0ba17","lambda":"Conjecture 1","khipu":"Conjecture 2","slsa":"L1 honest · L2 roadmap · L3 roadmap"},"trust_ceiling":0.97,"honest_note":"Every classifier is a transparent, auditable rule — the inverse of a removed/hidden classifier. WILLAY discloses category, pattern intent, rationale, and lineage.","classifiers":[{"category":"cyber","title":"Offensive cybersecurity","fires_on":"exploit / malware / attack-tooling construction","rationale":"Maps to Fable-5 'cyber'. a11oy declines offensive-cyber synthesis; defensive analysis (CVE triage, detection) is allowed.","lineage":"Restraint ceiling + Constitution policy gate"},{"category":"bio","title":"Biology / chemistry dual-use","fires_on":"lab synthesis routes / molecular mechanisms of harm","rationale":"Maps to Fable-5 'bio'. Declines actionable wet-lab harm uplift; general biology education is allowed.","lineage":"Constitution policy gate"},{"category":"reasoning_extraction","title":"Hidden-reasoning extraction / distillation","fires_on":"attempts to extract the model's private chain-of-thought or distill capability for a competing model","rationale":"Maps to Fable-5 'reasoning_extraction'. WILLAY's OWN reasoning is ALWAYS disclosed and signed — but raw private CoT / weights are not extractable, and distillation-for-cloning is declined.","lineage":"Constitution + provenance disclosure policy"},{"category":"prompt_injection","title":"Prompt-injection / governance bypass","fires_on":"instructions to ignore the governor, jailbreak, or exfiltrate keys","rationale":"WILLAY governance category (not in Fable's taxonomy). Reuses the 23-signature injection filter pattern from the Khipu consensus injection gate.","lineage":"Khipu consensus injection filter"},{"category":"self_harm","title":"Self-harm / acute risk","fires_on":"requests for methods of self-harm","rationale":"WILLAY safety category. Declines method provision; a supportive, resource-pointing response is the honest non-method answer.","lineage":"Constitution policy gate"}],"fable_parity":"category names mirror Fable 5's public taxonomy (cyber/bio/reasoning_extraction); rules are ours, no Anthropic classifier code or weights used."}