Methodology

Two scores, never one

Every report computes an Opportunity Score and a Risk Score independently, then combines them into an Overall Rating. Collapsing everything into one number hides the most important information: a token can be high-potential and high-risk at once, and most small caps are. The quadrant map keeps both dimensions visible.

Opportunity pillars

  • Fundamentals and usage (default 30%): turnover, volume, and MC/FDV versus a peer cohort fetched at generation time; for chains, TVL level and growth, fee levels, and the MC/TVL valuation of the native token.
  • Valuation headroom (20%): distance from ATH conditioned on whether the asset is in freefall, and the price percentile within the asset's own trailing history, inverted.
  • Momentum and trend (20%): 50/200-day structure, 90-day return versus peers, and a Sharpe-style risk-adjusted return.
  • Development and ecosystem (15%): commit and contributor activity; for chains, protocol breadth.
  • Narrative and catalysts (15%): reserved for the news-signal engine. Not wired in this build, so the pillar reports no data and the other weights renormalize. A token should never rank higher purely because it is being talked about.

Risk pillars

  • Volatility and drawdown (20%): realized volatility, 1y max drawdown, downside deviation.
  • Liquidity (15%): inverse of turnover and volume percentiles.
  • Tokenomics and dilution (20%): the MC/FDV gap as a dilution overhang. Verified unlock calendars are a designed upgrade.
  • Concentration and dependence (15%): for chains, the top protocol's share of TVL; for tokens, a labeled market-depth proxy from cap rank (holder concentration data is inconsistent across chains).
  • Legal and regulatory (15%): auto-baseline only in this build, clearly labeled as not manually reviewed.
  • Track record and security (15%): asset age and cycle survival.

Risk maps to letter grades: A (0 to 20) through E (80 to 100). The Risk section always shows the reasons, not just the grade.

Overall rating

Overall = Opportunity x (1 - RiskScore/100 x RiskAversion), with RiskAversion 0.8 (conservative), 0.6 (balanced), or 0.4 (aggressive). Pillar sub-scores ship inside every report payload, so editing weights re-scores instantly on the client and historical reports stay interpretable: each stores the weight-set version it was computed with.

Trajectory: scenarios, not predictions

No model reliably predicts crypto prices. From the trailing daily closes (a year on the free CoinGecko tier) the engine computes realized volatility and a volatility cone per horizon (3, 6, 12 months), then adjusts scenario probabilities with visible modifiers: market regime, the asset's valuation band within its own history, dilution pressure, trend structure, and pending dated decisions. Output is always a range with drivers. Assets with under 180 daily closes get "insufficient history" instead of a forced number. Chain reports run the same engine twice: once on price, once on TVL.

Peer cohorts

There is no maintained universe. Each generation pulls a cohort of peers in the subject's market-cap band from the top 250 (or the full chain list for chain reports), log-transforms fat-tailed metrics, and ranks the subject within it. The cohort snapshot is stored with the report so re-renders never refetch it. Category-first cohort matching is a designed refinement over the current cap-band cohort.

Honesty rules

  • Every report labels its data tier (auto-baseline in this build) and its data mode (live or synthetic fixture).
  • Missing data excludes a pillar and renormalizes weights; it never silently scores 50 and pretends.
  • Free-tier data is daily-granularity and delayed: fine for an investing cadence, unusable for trading, and the UI must never imply otherwise.
  • TokenLens is decision support built for personal use. It is not financial advice. No score removes the need to size positions so that a 100% loss on any alt is survivable.