If ambiguity resolution rules are made user-visible (e.g., recency wins, explicit priority tags win, or project-owner instructions win), how do different default ambiguity policies change users’ perceptions of fairness and blame when an assistant must choose between conflicting instructions from the same user versus from different users in a shared project?

Answer

Visible ambiguity rules shift blame from the assistant to the stated policy, but different defaults move perceived fairness and blame differently for intra‑user vs inter‑user conflicts.

Same user, conflicting instructions

• Recency‑wins
  • Fairness: Seen as simple but sometimes careless ("you forgot my long‑term preference"). Fairness is higher when the policy is shown and users can pin persistent instructions (soft exception).
  • Blame: Users blame themselves more ("I said something newer") and the policy a bit, and the assistant less, if explanations reuse the rule label (e.g., "Used: recency rule").
• Explicit‑priority‑tags‑win
  • Fairness: Perceived as most fair and controllable if tagging is easy. Users see mistakes as their own for not tagging or for tagging inconsistently.
  • Blame: Shifts heavily to the user’s tagging choices and only lightly to the assistant.

Different users in a shared project

• Project‑owner‑wins
  • Fairness: Seen as procedurally fair when roles match existing org norms; non‑owners accept the owner’s primacy if this rule is clearly surfaced in the shared space.
  • Blame: Non‑owners mostly blame project governance (who the owner is), not the assistant; owners blame themselves or their defaults.
• Most‑restrictive‑instruction‑wins
  • Fairness: Read as safety‑oriented and impartial but can feel unfair if one cautious user blocks others’ work and there is no scoped override (e.g., local exception or separate workspace).
  • Blame: Users blame the conservative teammate and, secondarily, the policy that empowered them.

Comparative effects of making these rules visible

• Visibility of which ambiguity rule fired reduces perceptions of personal bias and arbitrary behavior across both intra‑ and inter‑user conflicts, aligning with prior findings on explicit conflict rules and chain‑of‑command cues (c306, c307, c308, c309, c310; c264, c265, c266, c267).
• For same‑user conflicts, recency is tolerated but seen as less fair than explicit‑priority tagging when users care about long‑lived policies; fairness rises if users can mark durable instructions that override recency.
• For multi‑user conflicts, role‑based (owner‑wins) rules are usually judged fairer than pure recency and often more acceptable than most‑restrictive unless safety is salient; most‑restrictive feels fairer if paired with visible ways to carve out less‑restricted sub‑spaces.
• Across all variants, if the visible ambiguity rule routinely clashes with side‑effect controls or hard rules without clear labeling, users revert to blaming the assistant and see the policy as fake; using the same labels in the legible behavior policy and in explanations maintains fairness and places blame on the chosen rule or governance, not on the assistant personally.