Add v2_a11 product_shape_gravity_resistance axis to filter_score.js

Append 'product_shape_gravity_resistance' as the 6th entry in the FILTERS array (after line ~19). Scoring rubric 0-3: 0 = core value delivery is claim-verification, audit-trail generation, compliance-checking, or source-attribution auditing; 1 = primary mechanic is adjacent to the verification cluster (monitoring dashboards, trust-scoring pipelines, attestation workflows); 2 = distinct primary mechanic that incidentally touches verification contexts; 3 = primary mechanic is structurally distant from the audit/verification cluster. Add this 0-3 rubric verbatim to both the HIGH and LOW scoring system prompts (lines ~72-114). The composite formula (medianTotal - 0.5*iqrTotal + 0.3*signal_count) iterates FILTERS dynamically — no formula change required. Leave ABSOLUTE_FLOOR=9 unchanged for the first 30 scored proposals post-deploy to observe composite distribution shift before recalibrating the floor.

The 4 Commander-killed hypotheses (a38d31 = AI Control Failure Forecast Audit, e9cb5c = Reality-Graded Upgrade Gate, c89a71 = ClaimGate for B2B SaaS, 6bf9c5 = AI Tool Claim Verification) all score 0 on v2_a11. Graduated candidates score 2-3. The axis adds a 2-3 point composite spread between gravity-well proposals and structurally-distinct proposals at equivalent scores on existing axes, reducing gravity-well proposals' rank in the Commander queue and pushing those at the graduation margin below the floor.

Retroactively score the 4 Commander-killed proposals using only the v2_a11 rubric. If any score >1, the rubric is too permissive and needs tightening. Independently score 5 graduated proposals on v2_a11; if any score <2, the axis is incorrectly penalizing valid proposals.

Rejected by filter_score. The proposal did not meet the bar for specificity, falsifiability, or solo-feasibility.