Atlas scores observable market structure and activity. It does not predict returns or certify token safety.
Heat V2.0 is neutral market attention, bounded 0–100: 35 × clamp(log10(1 + one-hour USD volume) / 6), 30 × clamp(log10(1 + one-hour transactions) / 4), 20 × clamp(log10(1 + USD liquidity) / 7), and 15 × clamp(abs(one-hour price change %) / 50). Every clamp is 0–1; round the final sum. Paid boosts contribute zero. Reversing buys and sells leaves Heat unchanged.
Missing or invalid required inputs produce no Heat score. Zero is known zero activity. Promotion is labelled separately as paid attention.
Velocity compares actual observations at least 30 seconds apart within a 20-minute window. FAST: Heat change ≥10; BUILDING: ≥3; COOLING: ≤−3; STEADY otherwise. Heat V1 observations are excluded from V2 comparisons. Optional durable storage can retain compatible observations across restarts; Status shows actual continuity. No history is fabricated.
Atlas Score measures evidence alignment, never probability of profit. Called values, identity, original reasons and risks are locked at publication. Latest values are the most recently sampled observations. Sampled peak is the highest value captured by Atlas after publication and may differ from intraperiod exchange highs. Returns use comparable price pairs first, otherwise comparable market-cap pairs; those baselines are never mixed. Both favorable and unfavorable records persist. These are research observations, not audited investment performance. Recent ledger views are bounded and do not establish an all-history win rate.
Market score, 0–100. Five components: liquidity 25, structure 20, activity 20, attention 20, evidence 15. No paid-promotion or buy-direction input. Security is a separate, unverified state.
Each signal below has a maximum weight and a bounded 0–1 value. A component displays its observed points divided by available weight, scaled to that component’s maximum. Coverage is the sum of available signal weights, out of 100. The headline total is the sum of observed points divided by available weight ×100, rounded, then capped. Missing signals remain unknown; they are not zero-filled. Inputs and partial coverage are shown in the score breakdown.
15: log10(1+USD liquidity)/log10(1+$10M). 6: liquidity/valuation, scaled to 10%. 4: liquidity/max(one-hour volume,1). Valuation uses known positive FDV, otherwise known positive market cap. Every ratio is clamped 0–1.
6: log10(1+pool age in hours)/log10(169). 7: 1−clamp(log10(1+valuation/liquidity)/3). 4: 1−clamp(max(0,24h volume/liquidity−10)/90), so extreme turnover does not earn more points. 3: independent price agreement when at least two providers corroborate.
10: log one-hour transactions, capped at 10,000. 6: log one-hour USD volume, capped at $1M. 4: clamp(12×five-minute volume/one-hour volume) when the baseline is positive. Trading is direction neutral: heavy selling is still activity.
10: Heat V2/100. 5: clamp((observed Heat change+10)/30). 2: clamp((observed liquidity percent change+10)/30). 2: observed duration scaled to ten minutes. 1: up to three actual, recent non-promotional Tape events, available only when that context is loaded. Missing history never invents acceleration. Compact Radar analysis uses already-loaded data; deep Token Lab analysis can include recorded Tape evidence without new provider requests.
6: fresh source within 90 seconds=1, up to five minutes=0.4, older=0. 4: completeness of price, liquidity, valuation, 5m/1h/24h volume, one-hour transactions and pool age. 3: pool identity plus independent provider count capped at two. 2: independent price agreement. A single source cannot establish independent agreement. Atlas never averages away conflicting prices.
Coverage below 60% hides the headline total. Positive known price/liquidity, pool identity and a source timestamp no older than five minutes are also required. Below 80% coverage, maximum score is 74. Liquidity below $10K caps at 39; below $25K caps at 59. Price disagreement above 5% caps at 59 and reduces evidence quality. Very young pools under two hours have LOW evidence confidence. HIGH requires at least 90% coverage, a fresh source, two providers and ten minutes of actual history. Limited history: until Atlas has observed a pool for at least 10 minutes across at least 3 snapshots, the headline is labeled LIMITED HISTORY and capped at 79. The full range becomes available after sufficient history, including history restored after a restart. These are research heuristics, not calibrated return forecasts.
Labels: 0–24 very thin; 25–44 fragile market; 45–59 developing; 60–74 active; 75–89 strong market structure; 90–100 deep/high activity. A label describes current observable conditions, not safety or investment merit. Security data, honeypot status, mint/freeze rights, holders, LP locks and developer wallets are not established by these market inputs. External scores remain separately attributed and are not enabled in this release.
Future material formula changes require a version change. Heat V1 and V2 are never compared. Legacy observations expire within their bounded retention windows and are excluded from V2 Velocity and Tape views.
Tape uses repeated real snapshots at least 30 seconds apart. Thresholds: Heat ±10, rank ±5, transactions +40%, liquidity ±10%, volume +50%. Ranks depend on a changing sample. Retained history lasts at most 20 minutes; events at most 24 hours/500 records. Optional persistence does not invent missing observations.
Attention vs Price uses price flat within ±0.5% and Heat flat within ±3 points. It compares compatible observations only. Source timestamps and unknown values remain visible. No interpolated prices or security findings.