gap_evidence_matrix.csv

gap_id,gap_statement,supporting_paper_ids,counterevidence_paper_ids,novelty,significance,feasibility,evaluability,data_availability,engineering_risk,publication_risk,ablation_clarity,negative_result_value,status,next_check
G1,"conditional downstream contact utility and selective abstention","P07;P08;P19;P20;P21;P22;P40","P07;P08;P22;P40",7,9,8,9,7,5,5,9,10,"superseded by factorized G13","flat utility target failed Gate-B V1"
G2,"fixed-candidate residual value of VLM evidence beyond geometry","P07;P08;P10;P14;P15","P01;P15;P26",6,9,8,10,8,3,6,10,10,"Gate-B V1 No-Go; retain semantic-axis test only","do not run LOCKED80 or retune current cohort"
G3,"transfer boundary across initializer loss and domain","P02;P07;P14;P15;P21","P21",7,8,6,8,6,7,5,8,9,"selected full-version","choose one feasible cross-context"
G4,"marginal utility and non-additivity of contact sets","P01;P07",,8,8,5,7,6,8,6,7,7,"full/stretch","top-1 MVP then pairwise audit"
G5,"person-permutation and mirror-consistent contact grounding","P07;P08;P14",,7,7,8,9,8,4,7,9,6,"feature/stress test","compare standard augmentation and enumeration"
G6,"ambiguity-aware facet or landmark bilateral correspondence","P01;P16;P18","P01",6,7,4,5,5,9,8,6,6,"defer","remove 75x75 claim; assess facet annotations"
G7,"time-varying trust and persistent video contact","P03;P11;P14","P14",7,8,4,7,7,9,7,7,8,"defer","only after image selector works"
G8,"open-vocabulary contact ontology mapping","P07;P16;P18",,6,7,6,6,8,7,8,6,7,"defer","ontology error audit"
G9,"fast surrogate for paired optimizer utility","P20;P22",,8,8,5,8,7,8,6,8,7,"stretch","requires utility dataset first"
G10,"grouped or shift-aware contact risk control","P21;P22",,7,8,5,7,5,8,6,8,8,"stretch","formalize exchangeability and grouped calibration"
G11,"contact-specific equivariance residual as a predictor of downstream harm","P27;P28","P27",7,8,8,9,8,4,6,9,9,"selected feature and mechanism test","test whether flip and person-swap residual predicts utility after controlling for geometry"
G12,"outcome calibration when VLM evidence is stable but wrong","P27;P28;P22","P26",8,9,7,9,7,6,5,9,10,"selected safety framing","compare self-consistency confidence against paired optimizer utility and selective-risk metrics"
G13,"vector-valued human-human contact intervention outcomes separating semantic validity local contact benefit global geometry and safety","P07;P08;P17;P20;P26;P40","P40",8,10,7,10,8,6,4,10,10,"selected innovation one problem","Gate-BR0 candidate-local target audit"
G14,"reversible short optimizer response probe for predicting non-convex contact intervention outcome before commitment","P20;P40;P41","P41",8,9,7,9,8,6,5,10,9,"selected innovation one mechanism","Gate-BR1 fixed-budget trajectory smoke"