A.19.SelectorMechanism:3 - Forces

1. Set‑valued reality vs single‑winner convenience. Many admissible comparisons are partial orders. The kernel must preserve set‑valued semantics while still allowing single‑winner outcomes when explicitly requested by criteria.

2. Policy primacy vs method freedom. Criteria and defaults must be explicit and policy‑bound, while multiple method families and decision styles must remain add‑able without mutating the kernel.

3. No hidden thresholds vs usability pressure. Engineers often want “just pick one.” If the spec does not constrain this, hidden thresholds and tie‑breakers become de facto policy.

4. Evidence discipline vs delivery pressure. Under uncertainty, teams default to coercion (unknown → pass). The kernel must enforce tri‑state eligibility and fail‑closed discipline.

5. Replayability vs conceptual minimalism. The mechanism declaration stays small, while dated selection work, the actual Select application and its argument and SelectionSlot bindings, and the evidence-provenance path retain the effective editions, policies, candidates, and selected set needed for replay.

6. Evolvability vs didactic usability. The kernel must be stable enough to support SoTA wiring and specialisation chains, but also teachable: one place states the mechanism boundary, laws, eligibility behavior, and the neighboring replay basis for realized use.

7. Planned use and gate/guard separation. A.15.2 carries the intended baseline; A.15.3 typed filling is conditional on independently declared positions. Selection must not mutate into a gate pattern: no GateDecision or decision logs inside the mechanism boundary.

8. No competing defaults. If defaults exist for PortfolioMode, dominance regime, or archive policy, cite their declared sources rather than re-declaring them in the kernel.

9. Scope continuity vs legitimate reselection. Selection may narrow candidates or apply explicit policy, but it may not silently change the finite upstream comparison-application basis, its required pair coverage, any member’s predicate basis, claim scope, selected context slices, reference plane, or evaluation window. A justified change is a new selection application and may require new binary comparisons.