§02f Jupiter — Post-Hoc Forensics — Causal attribution
Mars® Spec › §02f Jupiter — Post-Hoc Forensics › Causal attribution
← Inter-AI analyst orchestration · Section index · Adaptive posture →
4. Causal attribution
A forensic finding produced by the post-hoc forensic layer may be attributed to a causal dimension — one or more of {data, user, prompt, model} — as an order-typed, evidenced, recomputable judgment.
Order-typed attribution. The causal attribution is tagged by the order(s) under which the finding and its cause are expressed. Attribution to a causal dimension is not a flat label; it is an order-typed claim that inherits the finding’s order structure.
Evidenced claim. The attribution carries the cited evidence from the retained witnesses that supports the assignment to its causal dimension: the methods applied, the divergence signals consumed, and the derivation steps. The attribution is a claim with its proof, not a bare label.
Recomputability. The causal attribution is recomputable: an independent inspector re-derives the causal-dimension assignment from the finding and its retained witnesses, without trust in the deployment. A mismatch reveals an unsound attribution. Recomputation witnesses consumed by this section must satisfy the §02a §5a.2 multi-stage witness chain structural requirements; a witness that does not satisfy those requirements is structurally incomplete as a causal attribution witness regardless of how its provenance is labeled.
Multi-dimension attribution. Where a finding admits more than one cause, the forensic layer emits an ordered set of dimension attributions, each evidenced and recomputable, with a declared basis for ordering. A finding with multiple causes is not collapsed to a single-dimension label.
Causal-dimension taxonomy. The governed taxonomy is {data, user, prompt, model}, extensible. The taxonomy is not exhaustive by design; future causal dimensions may be declared and registered without structural change to the attribution mechanism.
Attribution as the composition interface. The causal attribution is the interface between the forensic finding and the downstream forensic consumers — the reinforcement-action layer, the adaptive posture layer, and the temporal-chain forensic layer. Each downstream consumer takes the attributed finding as its input; the attribution is the linchpin of the post-hoc forensic layer. An implementation of any downstream consumer that does not ground its operation in a causal attribution produced under the requirements of this section is not consuming the post-hoc forensic layer’s composition interface.
← Inter-AI analyst orchestration · Section index · Adaptive posture →