Methodology · Public

How we score. In public.

Call Score is a 0–100 score for one matured market call. Creator ranking is a separate eligibility-and-ranking contract that uses sample size, exclusions, freshness, and bucket states before anyone appears as official. If a number looks wrong, audit me.

components: 5
candles: 18.7M
creator seed list: 123
extraction floor: 70%

The seed-list count is the repo-maintained creator admission baseline. The homepage tracked-creator count is live HH PostgreSQL coverage and may be higher after runtime ingestion.

The pipeline.

Scrape → Extract → Match → Score → Rank

step 01
Scrape
Auto-generated subtitles, daily
step 02
Extract
AI identifies actionable predictions
step 03
Match
Each call ↔ Binance candles
step 04
Score
5 components → 0–100 Call Score
step 05
Rank
Creator Rank Score + eligibility buckets

The score.

5 components · max 100 · floor 70%

Component	Max	How it's earned
Direction correct	40	30d move must clear the direction threshold: bullish > +2%, bearish < -2%, neutral within ±10%.
Alpha over BTC	25	BTC-relative 30d alpha contributes up to 25 points; current implementation floors negative alpha at 0 in the public 0–100 score.
Specificity	15	Entry, target, stop-loss, and timeframe can each add credit; current data is mostly target-driven.
Regime difficulty	10	Contrarian calls receive more credit through the stored regime_difficulty field.
Target hit	10	Stated target reached within 90d.
Total	100	Sum of components, no rescaling.

Tracked coins.

18 symbols · Binance OHLCV

Calls on these tickers are matched against minute-grained Binance candles. Anything outside this universe is logged but not scored.

BTCETHSOLBNBXRPDOGEADAAVAXDOTLINKTAORENDERFETNEARARINJSUIPENDLE

Creator rank contract.

official / provisional / watchlist / stale / excluded

Call Score ≠ Creator Rank Score

Call Score scores one matured prediction. Creator Rank Score ranks a creator only after official eligibility gates pass. The current writer stores average Call Score in creator_stats.alpha_score, then assigns rank with a sample-adjusted Creator Rank Score before raw score, win rate, sample size, and creator id tie-breakers. The legacy column name is misleading: it is not raw average alpha.

→ sample_adjusted_score DESC
→ alpha_score DESC
→ win_rate DESC
→ total_calls DESC
→ creator_id ASC

Official thresholds

All time	24+ official · 50+ certified
12m	12+ official · 25+ certified
90d	3+ official · 10+ certified
30d	disabled: PENDING_MATURITY

A serious creator making roughly one mature qualifying call per month can enter the 12m official leaderboard at 12 calls. Certified labels require larger samples. The 30d official leaderboard stays disabled as PENDING_MATURITY, not a fake zero-call ranking.

Creator states.

public-safe buckets

officialRankedRows

Official ranked

Passes official sample, rank, freshness, target-creator, and exclusion gates and may appear in the public leaderboard.

provisionalRows

Provisional

Has meaningful scored evidence but does not yet meet the official ranking contract.

watchlistRows

Watchlist

Tracked creator with insufficient mature public-scored evidence for an official or provisional signal.

staleRows

Stale

Historical evidence exists, but current coverage is too old or incomplete for official ranking.

excludedRows

Excluded

Non-target or policy-excluded source; permitted only in admin/audit contexts, not public rankings.

pendingMaturityRows

Pending maturity

Period or calls are waiting for outcome windows to mature; not eligible for official ranking.

Score lifecycle.

status is separate from value

A real scored call can eventually have a value of 0. The long-term methodology separates score lifecycle from score value. Today some writer/count paths still use stored score 0 as an unscored placeholder, so the read layer is conservative until a schema/recompute migration is approved.

state 01 · raw_candidate
Raw candidate
AI-extracted prediction candidate before ownership, confidence, market-data, and maturity checks.
state 02 · confidence_pass
Confidence pass
Candidate passes the public extraction floor and is eligible for price matching review.
state 03 · matched
Matched
Candidate has a valid supported asset and entry price anchor against market data.
state 04 · pending_maturity
Pending maturity
Candidate has not yet reached the outcome window required for public scoring.
state 05 · scored
Scored
Candidate has all required market outcomes and can receive a 0–100 Call Score.
state 06 · excluded
Excluded
Candidate or source is non-target, contaminated, too low-confidence, or policy-excluded.
state 07 · invalid
Invalid
Candidate failed extraction or validation and must not enter public scoring denominators.

Call-score bands.

S / A / B / C bands

Tier	Score range	Reading
S	70 – 100	Elite individual-call score, not a guarantee of creator rank.
A	55 – 69	Strong call-level outcome with directional edge.
B	40 – 54	Mixed call-level result; some edge but incomplete score components.
C	0 – 39	Weak call-level result or insufficient component evidence.

Sample adjustment.

implemented in source ranks

Official rankings use a sample-adjusted Creator Rank Score rather than a noisy raw average. This lets serious lower-frequency creators qualify while pulling 12-call samples toward the global baseline so variance alone cannot dominate proven larger samples.

(creator_raw_score * N + global_baseline_score * prior_N) / (N + prior_N)

45% · Sample-adjusted average Call Score
25% · Sample-adjusted BTC-relative alpha percentile
15% · Consistency
10% · Freshness
5% · Specificity quality

Audit me.

Reproducible · source available for audit on request

The recompute pipeline is designed to be reproducible. Every public score should trace to a transcript line, a Binance candle range, a lifecycle status, and a deterministic formula. There is no hand-tuned weighting per creator.

Disagree with a number? Request the audit trail, rerun the score against the documented formula, and tell us where it diverges.

Source repository currently privateflag a wrong score