Methodology
This page explains how the Smart Cohort Radar (Lite) is computed and why each column appears on the table.
What is shown on Lite?
Lite is a once-per-day snapshot of on-chain momentum across supported chains. It’s intentionally delayed to differentiate it from Pro. The snapshot is timestamped as as of the chosen UTC date.
- Delay: default 1 day (configurable), i.e., today − delay_days. If that date doesn’t exist in the cache, the latest available date is used.
- Ranking: tokens are sorted by Momentum (desc) then Radar (desc).
These rules mirror the Lite emitter script used to produce the cache read by the front page.
Column definitions
- Smart Share — Fraction of sampled “smart cohort” addresses with a non-zero balance of the token on a given chain (0–1, shown as %).
- Δ d/d — Day-over-day change in Smart Share for the same symbol+chain (today vs. previous UTC day).
- Sample — Number of addresses sampled for the token on that chain on the snapshot date.
- Momentum — Percentile rank of the token’s Δ d/d relative to all tokens in the snapshot (0–100). Higher = stronger day’s change.
- Radar — Composite score combining level and confidence: round(100 × [0.7 × level_percentile + 0.3 × confidence]), where level_percentile is the percentile of Smart Share and confidence = min(1, Sample / 60).
- Tier — Category derived from Smart Share and Δ thresholds (see below). If Sample < 60, tier is LOW_SAMPLE.
- Address — Contract / token address passed through from the source data for convenience.
Tiering logic
Tier | Rule |
---|---|
HOT | Smart Share ≥ 0.35 and Δ d/d ≥ 0.10 |
WARM | Smart Share ≥ 0.25 and Δ d/d ≥ 0.05 |
RADAR | Δ d/d ≥ 0.02 or Smart Share ≥ 0.15 |
QUIET | Otherwise |
LOW_SAMPLE | Sample < 60 (confidence is capped by sample size) |
Numbers above are current defaults; they may be tuned as coverage evolves.
Notes & limitations
- “Smart cohort” represents a curated set of wallets chosen for signal quality; Lite does not expose cohort composition.
- Percentiles are computed within the snapshot day so rankings reflect that day’s cross-section, not a global history.
- Lite excludes some real-time corrections used by Pro (e.g., stricter outlier clipping and alerting); that’s by design.
- Scores are not investment advice. Use them to triage attention, not as standalone trading signals.