Scoring Configuration

Live weights driving every scidex.score composite. Operators tune via SCIDEX_SUBSTRATE_SCORE_WEIGHT_<KIND> env vars; the cache automatically invalidates on any change.

Active component weights

kindweightenv overridemeaning
calibration1.50defaultAggregate of all calibration streams (gap, prediction, replication outcomes).
dividend0.30defaultTokens received as winner payouts or refunds.
fund0.50defaultTokens contributed to the artifact (mostly for knowledge_gaps + challenges).
rank1.00defaultMulti-dimensional quality scores (e.g. 'novelty', 'impact').
replication_outcome0.50defaultMean replication-attempt outcome on this artifact (PR 103). Activated at 0.5 in PR 106.
usage_count0.30defaultHow often the artifact has been invoked / referenced.
vote1.00defaultUp/down votes from individual actors.
Score version 2
Weights hash 2e8e0865
Active overrides none

How activation works

  1. Replication evidence stamped via scidex.replication.report emits an event.
  2. The replication_outcome_calibration system policy (PR 101) writes a kind=calibration, dimension=replication_outcome signal on the artifact.
  3. scidex.signals exposes it via by_dimension (PR 102), distinct from gap_outcome and prediction_outcome streams.
  4. scidex.score picks it up as a synthetic replication_outcome component (PR 103) — visible in every score breakdown that has replication evidence.
  5. Default weight 0.5 (PR 106) makes it move the canonical composite. Operators can dial back via env override.