Tighten fast_feedback_loops rubric to penalize multi-causal outcome attribution

In filter_score.js, extend the existing `fast_feedback_loops` axis rubric in both the highSystem and lowSystem adversarial prompt strings. The current rubric asks 'Can the system learn from real outcomes within minutes/hours, not weeks?' — append the following disambiguation block: 'Score 2-3 ONLY when the resolution mechanism grades the output against external, objective ground truth that exists independently of the buyer's perception: a citation verifiable against an external database, a timestamp in a public system of record (Jira, GitHub, Zendesk), a classification against a known outcome (ATS hire/no-hire, regulatory ruling, market event). Score 0-1 when the primary feedback signal is: CSAT or NPS score, stakeholder satisfaction rating, managerial judgment, or any outcome where the AI output is one of multiple causal inputs (e.g., coaching + prompt change + agent output all contributed to the outcome — attribution is impossible). Anti-pattern to catch: when the hypothesis describes learning from whether "support quality improved" or "customers are happier," score 0-1. When it describes learning from whether a specific promise timestamp matched a Zendesk ticket resolution event, score 2-3.' Add this block to both highSystem (which argues high) and lowSystem (which argues low) in identical form so neither model gains an asymmetric anchor.

hyp-2026-05-06-847f7e (Support Promise Calibration Console — killed because 'CSAT/SLA outcomes are multi-causal') scores 0-1 on fast_feedback_loops under the revised rubric. hyp-2026-05-13-47730e (AI Portfolio Claim Auditor — killed because 'board verdicts multi-causal') scores 0-1. hyp-2026-05-06-ec4507 (Support Escalation — Zendesk timestamp as resolution mechanism) scores 2-3. Net effect: the NBJ Q4 describability test's 'external objective ground truth' criterion is enforced at filter_score rather than waiting for council to identify the flaw.

Apply revised fast_feedback_loops scoring retroactively to 43 S157 candidates. Required: the 4 FRAGILE candidates (hyp-47730e, hyp-dc26ff, hyp-a81ff4, hyp-a38d31) all score 0-1 on this axis. Required: at least 3 of the 5 ROBUST 5/5 candidates score 2-3. If average score gap between FRAGILE and ROBUST candidates on this axis is less than 1.0 point, the rubric is not discriminating on resolution objectivity.

Rejected at the council verdict. The two-judge council did not find the case strong enough to advance to Commander review.