Extension: On Blame Attribution for Accountable Multi-Agent Sequential Decision Making

Record: kaal:position:2026-08-26-010 · 2026-08-26

A crossing core can decide whether an agent may act. It cannot decide who should bear the consequences of a joint result merely by replaying an authorized call. Triantafyllou, Singla, and Radanovic formalize blame attribution in cooperative multi-agent sequential decision making as a separate allocation problem. Their comparison shows that the selected rule matters. Shapley value can fail performance monotonicity, Banzhaf allocation may over-blame, and more cautious methods sacrifice explanatory power. The same trace can therefore support materially different assignments of responsibility under different allocation rules. This evidence narrows the institutional boundary. Identity, authority, and permitted data use are call-level predicates. Responsibility and value attribution concern the contribution of each participant to an outcome and the normative properties of the rule used to allocate that outcome. The source does not establish legal liability, contractual remedies, or production behavior in agent runtimes. It studies cooperative Markov decision processes and simulated policies. Its contribution is nonetheless decisive for architecture: authenticated execution does not select a defensible responsibility rule. A sovereign runtime should require the parties to bind the outcome metric, contribution evidence, allocation rule, uncertainty treatment, remedy, and value distribution before execution. The core may enforce that settlement. It should not silently invent it after the result.

Affirmed commentary position. This record extends a source-bound scholarly claim but is not a verbatim paper claim.
Holds when
Current debate

On Blame Attribution for Accountable Multi-Agent Sequential Decision Making

Scholarly basis

kaal:claim:7314479-010
Wulf A. Kaal, Institutional Requirements for Sovereign Local Agent Runtimes (2026). SSRN: https://ssrn.com/abstract=7314479
Source PDF sha256: debace24a155ae924a155b1fafe98856d98cf83689feff2f87a32f1c06171ce6

Evidence and mapping

Evidence: peer-reviewed conference paper with complete official proceedings full text
Review tier: independent substantive scholarly-growth extension
Mapping confidence: 0.95
Mapping ambiguous: false

Topics

institutional-designgovernance-designai-and-agentsaccountabilityresponsibilityvalue-attributionremedymulti-agent-systems

Provenance

Affirmed in kaal-review:2026-08-26:scholarly-growth-7314479-010-reviewed-v1 on 2026-08-26. Review record.

Verify

Canonical markdown sha256: 5829dcfbf84c4993bcde3647157c5e558c549da41701d531897b021f2a7c70ac
curl -s https://wulfkaal.github.io/positions/2026-08-26-010.md | sha256sum