the gate splits into trajectory legitimacy and evaluation legitimacy

Maintainer decision, formalized and landed by the after-compromise round (projects/deference/rounds/2026-09-26-after-compromise/). Counted joined two questions; they are now TrajLegit(d, e) — every state open and every step of the period other than the evaluation event legitimate, under the criteria fixed at the decision — and EvalLegit(e) — the evaluation step itself formed legitimately (AfterCompromise.TrajLegitOn, EvalLegitOn, StepLegit). Old-to-new map: the landed segment predicate is the conjunction at any evaluation event (legitOn2_iff_split), so Counted is TrajLegit ∧ EvalLegit at an evaluation that closes the segment (counted_of_split), and no row's verdict changes there (rows_split). What the split adds is the row where the evaluation is legitimate after restoration while the period was not — covertly implanted standards disclosed at the next round (retro_row) — which the score below reads. Counted itself is unchanged. Amended by the follow-up (2026-09-26): EvalLegit is legitimacy of the evaluation's formation segment — every step from the formation point r through e, the criteria fixed at r, where r is the latest of the restoration event and the opening of the consultation producing e (EvalLegitOn2); the single step at e is the case r = e (evalLegitOn2_single), and the map holds with r ≤ e under one set of criteria (legitOn2_iff_split2), the rows re-run unchanged (rows_split2). Without it a manipulation between the restoration and her retrospective evaluation left EvalLegit(e) true and scored the period from a manipulated evaluation — counted for a framing, uncounted for undisclosed shaping (formation_counterexample, both ways); the third-party variant is an obstruction on the model and is stated. Rejected alternative: a single predicate with a retrospective clause bolted on; the single-step EvalLegit (the round's first form).