Function reduce_to_fit 

pub fn reduce_to_fit<F>(
    full_msgs: &[ChatMessage],
    base_policy: &ReductionPolicy,
    prior_log: &ReductionLog,
    fits: F,
) -> (Vec<ChatMessage>, ReductionLog, ReductionPolicy)
where
    F: Fn(&[ChatMessage]) -> bool,

PARITY-18 v3 — the “aggressive reduction” half of the fix (SPEC.md scaling-context-guard): apply project_messages with base_policy first, then, if fits says the projected view still doesn’t pass, retry with progressively tighter policies (see tighten) up to MAX_AGGRESSIVE_LEVELS times, then escalate to A10 turn-clearing, keeping whichever attempt is smallest. Never fails and never loops unboundedly — it always returns SOME projection (the caller’s own preflight guard is responsible for deciding whether even the tightest attempt still exceeds the target model’s context limit and refusing to send in that case, PARITY-18 dev/01).

fits is a closure over the REDUCED VIEW ALONE (this function has no knowledge of the system prompt, tool schemas, or a trailing user prompt — those live with the caller). It exists to close a v2 defect a skeptical review caught (the flagship rescue-path regression): v2 had two INDEPENDENTLY-derived boundaries — this function stopped escalating once estimate_view_tokens(view) <= target_tokens where target_tokens = context_limit - CONTEXT_RESPONSE_RESERVE_TOKENS, while tokens::context_guard only ACCEPTS a request when with_guard_margin(view + tools) + CONTEXT_RESPONSE_RESERVE_TOKENS <= context_limit — a materially tighter bound (no margin, no tools, no system-prompt overhead counted by the reducer’s stop condition at all). Any session whose reduced view landed in the band between those two boundaries was declared “fits” here and then refused by the guard, with D6 turn-clearing never triggered because this function had already stopped. Passing tokens::context_guard itself (wrapped with the caller’s real system prompt/tools/prompt) as fits makes that impossible BY CONSTRUCTION: the reducer’s stopping condition and the guard’s acceptance are now the same boundary, so a session this function says it reduced-to-fit is a session the guard will actually send. (resume_cmd is the sole real caller; see its fits closure there.)

Returns (view, log, policy_actually_applied) so the caller can seed a runtime agent’s reduction policy/log setters with the exact policy that produced the returned view (subsequent turns keep re-applying it, resume_cmd’s existing re-reduce path).