score = 100 × weighted mean of the signals below (weights renormalized). Freshness & attention are excluded on purpose.
- Creator consensus (evidence-weighted)0.867×0.833=0.722
- Reviewer authority0.067×0.500=0.033
- Agreement between reviewers0.067×0.945=0.063
Consensus shrinkage raw consensus 0.835 is pulled toward the catalog prior 0.83 (strength 10) over 17 reviews → 0.833. That shrinkage is why a thinly-unanimous shoe doesn’t outrank a deeply-loved one.
Excluded Freshness and attention are deliberately excluded from the score so the tier means quality, not recency or buzz. They are surfaced separately as the trending signal.
The highest rung whose cutoff the score clears.
- ○S≥ 89.3
- ○A≥ 86.3
- ●B≥ 81.0
- ○C≥ 76.0
- ○Dfloor
4.47 points from the next rung up.
Evaluated in order on the raw consensus (0.835) and the MED confidence band; first match wins.
- ○SKIPconsensus_numeric < 0.6
- ○BUY NOWconsensus_numeric >= 0.8 AND confidence == HIGH
- ●CONSIDERotherwise
The verdict reads the RAW consensus (share of creators recommending), not the score. A shoe can therefore be tier A yet CONSIDER rather than BUY_NOW when confidence is below HIGH.
A weighted mix of how the evidence stacks up:
- How many independent creators0.450×0.500=0.225
- Depth of each mention0.250×0.888=0.222
- Evidence recency0.200×0.318=0.064
- Reviewer authority0.100×0.500=0.050
A grade publishes only with enough independent creators AND at least one volume bar met.
- ●Independent creators5 / 3 needed
- ●Reviews where this shoe is the primary subject17 / 10 needed
- ●Total mentions47 / 30 needed
Behind it: 47 mentions (17 primary-subject) from 5 creators, 4 dedicated reviews.
Stages A/B/C (extract, synthesize, write) produce the editorial narrative only. No stage A/B/C value is an input to score, tier, verdict or confidence — those come solely from the consensus engine over stored mentions. Per-stage counts live on the `reasoning` block.