ADR 0002: Personas pair 1:1 with task types

Status

Superseded by 0020 (and informed by 0019).

The 1:1 task-type → persona pairing now stands as a suggested default route, not a deterministic assignment. The agent may re-assess and load a different persona skill when the work in front of it doesn't match the suggested default (recording the divergence in its task file's ## Decisions). Determinism is no longer gatekeeper-enforced — the flow graph is recommended routing (a launcher may apply it deterministically, but nothing in-session blocks a re-assessment). The original decision text below is kept as history.

Context

Free-form persona choice lets models pick whichever stance flatters continuation bias ("helpful reviewer who approves"). That collapses empirical discipline and mixes incompatible proofs in one session.

Decision

Every task type maps to exactly one default lead persona. The launcher sets persona from task type + flow graph; the agent does not self-assign.

Overlay personas remain project-local additions; they extend, not splinter, the canonical map.

Consequences

• Positive: deterministic conditioning; Skeptic proofs never blend with Builder improvisation in the same task by default.
• Negative: occasional misfit when reality straddles types — resolved by re-tasking, not persona blending mid-session.

Ledger note (2026-06-11): partially superseded by ADR-0064 (persona shipping model).