Methodology
Scores
Anatomy of a score
- Value — 0–100 (Divergence is signed, −100…+100).
nullwhen there is not enough data. - Drivers — named components with
points(they sum to the value, ± rounding), the raw feature (raw,display) and, when a reference distribution exists, a percentile. - Confidence —
level,sample_sizeand a humanbasissuch as “14 qualifying authors · 386 historical calls”. - Model version — e.g.
momentum-v1.0. Coefficients never change silently; a change ships as a new version and history stays reproducible. - Stale flag — when inputs were delayed (Fomo or market data), the score is marked
staleand rendered dimmed with the reason.
The example below is the real engine (momentum-v1.0) applied to an illustrative feature vector for a token called $XYZ — the same component the app shows when you click a score.
Example · illustrative inputs
momentum-v1.0Attention is accelerating.
| Driver | Value | Points |
|---|---|---|
Thesis velocity 14 theses in the last hour vs a 7-day baseline of 4.4/h | +218% | +23 |
Unique author growth 12 authors this hour vs 7 the hour before | +71% | +13 |
Acceleration 14 vs 6 theses in the prior hour | +133% | +15 |
Mention persistence | 18 / 24 hours active | +15 |
| Total | 66 | |
Based on: 37 theses · 26 authors in 24h
Model momentum-v1.0 · sample size 37
Confidence
| Level | Meaning |
|---|---|
INSUFFICIENT | Not enough data to compute the score at all. The value is null and shown as “—” with the reason. |
LOW | Computed, but from a thin sample. Treat as indicative. |
MEDIUM | A reasonable sample for this score. |
HIGH | A large sample by this score's own thresholds. |
Each score defines its own thresholds (below). Statistics elsewhere — backtests, event studies, analogs — always show N, confidence and the date range too.
Momentum momentum-v1.0
Measures how fast Fomo attention on a token is accelerating right now. Points = 35·sat(thesis velocity, 2) + 25·sat(unique author growth, 1)·min(authors this hour / 5, 1) + 20·sat(acceleration, 1) + 20·(active hours in 24h / 24), where sat(x, k) = 1 − e^(−x/k) for x > 0 and 0 otherwise.
Inputs
| Input | Description |
|---|---|
Thesis velocity (35 pts)thesis_velocity | Theses in the last hour versus the token's own 7-day hourly baseline: (theses1h − baseline) / max(baseline, 0.5). +218% means 3.18× the usual hourly rate. |
Unique author growth (25 pts)unique_author_growth | Distinct authors this hour versus the previous hour, damped until at least 5 authors are active so one or two accounts cannot max it out. |
Acceleration (20 pts)acceleration | Theses this hour versus the previous hour. Shown as a percentile of the trailing cross-section when a reference distribution exists. |
Mention persistence (20 pts)persistence | Share of the last 24 hourly buckets with at least one thesis — sustained mentions rather than a single burst. |
Range
0–100. 0 = no acceleration; 100 = every component saturated.
Confidence
From theses in the last 24h: fewer than 3 → INSUFFICIENT (no value), fewer than 10 → LOW, fewer than 30 → MEDIUM, otherwise HIGH.
Limitations
- Fomo activity is a sample of social attention; momentum describes attention, not value or future price.
- Hourly windows are noisy for low-activity tokens — read the confidence level and sample size.
- New tokens have a thin 7-day baseline (floored at 0.5 theses/hour), so early velocity can look extreme.
Conviction conviction-v1.0
Separates sustained attention from one-off noise. Points = 35·repeat-author ratio + 25·(active days / 7) + 20·sat(mean author days − 1, 1.5) + 20·revisit rate.
Inputs
| Input | Description |
|---|---|
Repeat authors (35 pts)repeat_authors | Share of authors active in the last 24h who posted on the token at least twice within 7 days. |
Active days (25 pts)multi_window | Days (of the last 7) with at least one thesis on the token. |
Author persistence (20 pts)author_persistence | Average number of distinct days each 7-day author posted on the token; one day contributes nothing. |
Revisit rate (20 pts)revisit | Share of the last 24h's theses written by authors who had already posted on the token earlier in the week. |
Range
0–100. Low = attention from first-time, one-off posts; high = the same authors keep coming back.
Confidence
From distinct authors in 7 days: fewer than 3 → INSUFFICIENT (no value), fewer than 10 → LOW, fewer than 25 → MEDIUM, otherwise HIGH.
Limitations
- Fomo activity is a sample of social attention; conviction describes posting behaviour, not value.
- A small group posting repeatedly also raises conviction — read it together with Crowding.
- Uses a 7-day window, so it reacts slowly to a brand-new token.
Crowding crowding-v1.0
Concentration of the last 24h of attention. High = a few accounts dominate; low = broad independent participation. Points = 30·min(top-1 author share / 50%, 1) + 25·top-5 author share + 25·Gini of per-author thesis counts (small-sample corrected; a single author = 1) + 20·(1 − min(distinct authors / 20, 1)).
Inputs
| Input | Description |
|---|---|
Top author share (30 pts)top1_share | Share of 24h theses written by the single most active author; 50% or more earns full points. |
Top 5 author share (25 pts)top5_share | Share of 24h theses written by the five most active authors. |
Concentration (25 pts)concentration | Gini coefficient of thesis counts across the 24h authors with the n/(n−1) small-sample correction (0 = everyone posted equally; one author alone = 1). |
Thin breadth (20 pts)thin_breadth | Fewer than 20 distinct authors in 24h adds crowding; 20 or more adds none. |
Range
0–100. High = concentrated (a caution state), low = broad.
Confidence
From theses in the last 24h: fewer than 3 → INSUFFICIENT (no value), fewer than 10 → LOW, fewer than 30 → MEDIUM, otherwise HIGH.
Limitations
- Fomo activity is a sample of social attention; crowding describes who is posting, not whether they are right.
- Related accounts operated by one person cannot be detected from public activity, so true crowding can be higher than shown.
- With very few theses, top-5 share is mechanically 100% — read the confidence level.
Early Quality early_quality-v1.0
Historical track record of the accounts posting early on a token. Takes the first ≤ 10 distinct authors in the last 24h who have ≥ 5 resolved calls (as of the score time). Their raw (un-shrunk) statistics are pooled — each author weighted by min(resolved calls, 40), so one prolific account cannot outweigh the rest — and shrunk ONCE: weight w = E / (E + 20), where E is the pooled evidence. Points = 50·research + 20·hit rate + 15·low false positives + 15·lead time, each component linear between pooled-set anchors (research composite 25 → 0, 70 → full; hit rate 20% → 0, 45% → full; false positives 62% → 0, 35% → full; median lead 30h → 0, 8h → full) and shrunk toward 0.5 with the same w. Never uses follower counts.
Inputs
| Input | Description |
|---|---|
Research score (50 pts)research | Pooled un-shrunk research composite of the qualifying early authors (each weighted by min(calls, 40)), computed only from outcomes resolved before the score time; 25 or less earns nothing, 70 or more earns full points. |
Hit rate (20 pts)hit_rate | Pooled share of their resolved calls that touched +25% within 72 hours; 20% or less earns nothing, 45% or more earns full points. |
Low false positives (15 pts)low_false_positives | Pooled share of calls with no major move and a negative 3-day return; 62% or more earns nothing, 35% or less earns full points. |
Lead time (15 pts)lead_time | Pooled median time from call to the first +25% touch; 30h or more earns nothing, 8h or less earns full points. |
Range
0–100. Around 50 = no evidence either way (the prior); above 70 = historically strong early posters.
Confidence
From qualifying authors and their resolved calls: none → INSUFFICIENT (no value); fewer than 3 authors or 60 calls → LOW; fewer than 8 authors or 250 calls → MEDIUM; otherwise HIGH. Shown as e.g. “14 qualifying authors · 386 historical calls”.
Limitations
- Fomo activity is a sample of social attention; past outcomes of these authors describe history, not what happens next.
- Only authors with at least 5 resolved calls count; brand-new accounts are invisible to this score.
- Outcomes use available market data; tokens without price history cannot contribute calls.
Narrative Velocity narrative_velocity-v1.0
How fast the token's narrative (AI Agents, Gaming, DePIN, …) is spreading through Fomo. Points = 35·sat(new narrative authors per hour, 3) + 25·sat(24h narrative growth, 1) + 20·network spread + 20·sat(other tokens in the narrative with theses in 24h, 5).
Inputs
| Input | Description |
|---|---|
New authors per hour (35 pts)new_author_rate | Authors posting in the narrative for the first time in 48h, over the last 6 hours, per hour. |
Narrative growth (25 pts)narrative_growth | Narrative theses in the last 24h versus the previous 24h. |
Network spread (20 pts)network_spread | Share of this token's 24h authors who are posting on it for the first time this week. |
Cross-token mentions (20 pts)cross_token | Other tokens in the same narrative that received theses in the last 24h. |
Range
0–100. Tokens without a classified narrative have no value.
Confidence
From the narrative's theses in the last 24h: fewer than 5 → INSUFFICIENT (no value), fewer than 20 → LOW, fewer than 60 → MEDIUM, otherwise HIGH. No classified narrative → INSUFFICIENT.
Limitations
- Fomo activity is a sample of social attention; narrative velocity describes spread of attention, not value.
- Narratives come from transparent rules (narrative-v1.1) over thesis text and the token's identity (symbol/name, stock or VIRTUAL pair, GeckoTerminal categories, launch venue); unusual wording can be misclassified, and a token with no signal at all lands in “Other”.
- All tokens of one narrative share the narrative-level components at the same moment.
Divergence divergence-v1.0
Signed gap between social attention and market activity: clamp(25·(z_social − max(z_price, z_volume)), −100, 100). Positive = social activity is ahead of the market; negative = the market is ahead of social activity. Also returns a state: EARLY SOCIAL (z_social > 1, |z_price| < 0.5), CONFIRMED (both > 1), LAGGING SOCIAL (z_price > 1, z_social < 0.5), DIVERGING (z_price > 1 while thesis activity falls > 30% hour-on-hour, or z_social > 1 while z_price < −1), NEUTRAL, or NO MARKET DATA.
Inputs
| Input | Description |
|---|---|
Social zsocial_z | Last-hour theses minus the token's median hourly count over the prior 7 days, divided by the hourly standard deviation (floored at 1 thesis/hour). Centring on the median keeps a typical hour near 0 because hourly counts are skewed. |
Price zprice_z | 1-hour log price change divided by the token's own trailing 48h hourly volatility (floored at 0.5%). |
Volume zvolume_z | Last-hour log volume versus the trailing 48h distribution of hourly volumes. The market side uses the larger of price z and volume z. |
Range
−100 to +100. 0 = social and market activity are in line.
Confidence
From hourly price returns available in the trailing 48h: fewer than 3 → INSUFFICIENT (no value), fewer than 12 → LOW, fewer than 36 → MEDIUM, otherwise HIGH. No market points → NO MARKET DATA.
Limitations
- Fomo activity is a sample of social attention; divergence describes timing differences, not value or direction of future price.
- Needs market data: tokens on chains without price history show NO MARKET DATA rather than a guessed value.
- Hourly volatility from 48h of data is a rough estimate; thin or illiquid pools produce noisy z-scores.
Smart Cohort smart_cohort-v1.0
Whether an algorithmic (auto) cohort is converging on the token: at least max(3, min(5, ⌈15% of the cohort⌉)) members posting on it within 60 minutes during the last 24h. For the strongest such event, points = 35·participation + 25·(1 − window minutes / 60) + 25·(cohort quality / 100) + 15·(1 − historical co-occurrence), where participation = members / min(cohort size, 10). 0 when no cohort converged.
Inputs
| Input | Description |
|---|---|
Participation (35 pts)participation | Members who posted in the window divided by the cohort size, counting 10 members as full participation for larger cohorts. |
Time clustering (25 pts)time_clustering | Tighter windows score higher: 1 − minutes / 60. |
Cohort quality (25 pts)cohort_quality | Mean point-in-time research score of the cohort's members. |
Rarity (15 pts)rarity | 1 − share of the trailing 7 days' hourly windows in which at least the cohort's convergence minimum of members co-posted on one token. |
Range
0–100. 0 = no convergence in 24h.
Confidence
From members across tracked cohorts (cohorts with ≥ 3 members and a known quality): no tracked cohort → INSUFFICIENT (no value); fewer than 10 members → LOW; fewer than 30 → MEDIUM; otherwise HIGH.
Limitations
- Fomo activity is a sample of social attention; a cohort converging describes coordination of attention, not value.
- Only algorithmic (auto) cohorts feed this public score; their membership is snapshotted daily and used point-in-time. Your saved cohorts drive cohort pages and alerts, never public scores.
- Cohorts with no quality history are counted for convergence but add no quality points.
Model versions
Active versions. Every quant object returns the versions it was computed with in model_versions, and GET /v1/models serves these definitions plus the registered coefficients.
| Model | Version |
|---|---|
| features | features-v1.0 |
| Momentum | momentum-v1.0 |
| Conviction | conviction-v1.0 |
| Crowding | crowding-v1.0 |
| Early Quality | early_quality-v1.0 |
| Narrative Velocity | narrative_velocity-v1.0 |
| Divergence | divergence-v1.0 |
| Smart Cohort | smart_cohort-v1.0 |
| signals | signals-v1.0 |
| outcome | outcome-v1.0 |
| research | research-v1.0 |
| narrative | narrative-v1.1 |
| analog | analog-v1.0 |
| backtest | backtest-v1.0 |
Scores feed the signal rules, the scanner and the backtest engine.
FomoQuant provides analytics and historical/statistical context, not financial advice or guaranteed predictions.