Methodology · v1.5

The ANTE Score

One verified number, 0–100, that says how much a creator's public trade calls have actually worked out. It is computed only by the platform from the resolved-call log — never anything a creator can set, edit, or cherry-pick. The method is public and versioned so the rating carries authority, not vibes.

How it's computed

• A win is a kept promise, not drift. A call counts as a win only if its stated target was hit before its stated stop, checked against the actual price tape. Stopped out first = loss. No target/stop hit by the horizon = resolved by realized P&L at that point.
• Only falsifiable calls score. A call must state a price level (target or stop). Vague "watching $X" posts aren't scored — they count against coverage, not the grade.
• Small samples shrink. The score is the Wilson lower bound on the win rate over resolved calls, so a 5-for-5 record doesn't masquerade as proven. Under 20 resolved calls we show "Building", not a number.
• Coverage gate, corroborated (v1.1). When we can verify fewer than 80% of a creator's detected calls, their unverified calls become the tiebreaker: if the 1-Week Read on those vague calls doesn't contradict the verified record (≥ −2% vs the market, over ≥ 10 calls), the honest score stands. If their stated calls "win" while their vague calls lose money — the cherry-picking signature — the score is capped at 60 and flagged. Missing data keeps the cap. A partial coverage badge is always shown either way.
• Default horizon. When a call states no timeframe, it is evaluated over a disclosed 60-day window.
• Chart-read levels are labelled. Levels read off a chart image (vision OCR) are validated against the real entry and carry a provenance flag; an unreadable chart is skipped, never guessed.
• Always shown with its evidence. The score never appears as a bare number — it's shown with the decided-call count and a 95% confidence interval, recomputed on every new resolution.

Bands

The 1-Week Read (beside the score, never in it)

Most public calls state no falsifiable level, so they can't be scored. For those we show a separate 1-Week Read: the average direction-adjusted price move over the 7 days after each call, from the entry at post time, minus the same-window market return (SPY for stocks, BTC for crypto) so a trending tape doesn't flatter everyone. It is momentum, not a kept promise — it is labelled unverified everywhere it appears and is never blended into the ANTE Score.

v1.2 — while a creator is Building (fewer than 20 decided calls) the profile and leaderboard show this read as a provisional signal once it covers at least 10 calls, together with the progress toward the gate and an estimated date. It is styled as a read, never as a score.

Horizons

A stated timeframe always wins. When a call states none, the evaluation horizon is a per-creator default by archetype (v1.2): about a week for scalpers and options-flow callers, 30 days for swing and chart traders, and the disclosed 60-day default for everyone else. The archetype is a property of the creator, never of the call, so it cannot be gamed call by call; changing a default only affects calls still open at the next check.

Record Shape — the five-axis radar

The ANTE Score is a verdict; the radar is the anatomy. Each spoke is arithmetic over the same resolved-call log the score used — nothing on it is a model's or an editor's opinion. Two creators with the same score can be opposite traders; the shape shows the trade-off the single number compresses. A radar is drawn only for creators with 20+ resolved calls — below that, any shape would be noise wearing precision.

• Accuracy — “How often are they right?” The raw win rate over resolved calls, using the same win definition as the score: stated target hit before the stated stop, checked against the tape. Alone it is the most misleading statistic in trading — which is why the next spoke exists.
• Expectancy — “If you took every call, did it make money per trade?” The average P&L per call, wins and losses netted, each trade capped at ±100% so one lottery-ticket option can't perfume the mean. 50 means breakeven; above it, following made money per trade, below it lost. The scale is stretched around breakeven (±5%/trade reads as ≈38/62, ±10% as ≈27/73, saturating for extreme records) so realistic differences are visible. High Accuracy with low Expectancy is the hidden leak: wins often, loses money, because the average loser dwarfs the average winner.
• Falsifiability — “How much of their feed can even be checked?” Coverage: the share of detected calls that stated a checkable price level, making them gradeable at all. The same coverage number that gates the score. Low falsifiability doesn't prove bad faith — it means most of the feed is unaccountable and the verified record rests on a thin slice.
• Discipline — “Do they define their risk before the trade?” The share of calls stating both a target and a stop — a complete, falsifiable risk plan. Across the cohort this is the spoke that separates “stops honored, losses small” from “no stops, averages down.”
• Track record — “How much evidence backs this shape?” Depth of the resolved sample, log-scaled from the 20-call gate (20 ≈ 20, ~60 ≈ 52, 300+ = 100). It measures confidence, not quality — a creator can max this spoke while every other sits at zero, which simply means the record is very well measured and bad.

Reading the shape: don't look for the biggest blob. A tall-Accuracy / short-Expectancy kite wins often and loses money; its mirror image is the classic trend-follower — wrong often, but winners dwarf losers. Tall Falsifiability and Discipline on a modest record is an honest, measurable trader worth watching mature. The shape is the trader's fingerprint; the score can't show it.

Changelog

v1.5 (2026-09) — a bar spanning both the target and the stop is declined as not adjudicable instead of resolving stop-first; stock minute bars are regular-session only; every verdict now records whether minute or daily bars decided it. v1.4 (2026-09) — options are judged at the exit the creator stated rather than held to expiry (see above). Every settled option had previously been graded as if the position were carried to expiration, which recorded near-total losses on trades whose stated target had been reached. v1.3 (2026-09) — spot FX and metals are no longer graded (see above); 641 previously booked FX verdicts were withdrawn and the affected records recomputed. The corpus hit rate moved from 24.9% to 30.2% once those calls left the denominator. v1.2 (2026-09) — archetype default horizons for calls without a stated timeframe, and the 1-Week Read shown as a labelled provisional signal while Building. No scored value changes on its own; open calls resolve under the new default at their next check. v1.1 (2026-07) — one rule changed: the blanket partial-coverage cap became the corroborated cap above. For records we scrape exhaustively, the platform — not the creator — controls what's counted, so the honest Wilson score stands whenever the unverified tail agrees with the verified one. Every prior value is preserved in the public score history. v1 (2026-06) — initial release.

Options: judged at the stated exit

Most option calls name a level on the underlying — “NVDA 190c, target 195, stop 180”. We adjudicate that level on the underlying's own tape (minute bars, so the entry day counts only from the moment the call was posted), then price the contract from its own premium on the day the level was reached. The verdict follows the creator's stated exit; the percentage follows the contract — so a kept promise can still carry a small loss, and a broken one can still show a gain. Both are shown.

Holding to expiry is used only where the creator named no exit. Where the stated levels are incoherent for the trade (a target on the wrong side of spot, or a number that is really the strike restated), or where one day's range spans both the target and the stop so first touch cannot be established, the call is not graded on those levels at all.

What we decline to grade

A verdict is only published when the price data is fine-grained enough to support it. Where it isn't, the call is marked not adjudicable and leaves the score and the coverage denominator entirely — it is never booked as a loss. Three cases today: option contracts with no tradeable bars; spot FX and metals, where we have no intraday feed (judging a stop set 0.26% from entry against a daily bar that ranges 1.82% would make the data's verdict, not the trader's); and any call, in any instrument, where the bar that first touched a level spans both the target and the stop. We drill minute bars into such days; when none are available, we no longer assume the stop came first — the call is declined. Stocks are judged on regular-session prints only (09:30–16:00 ET) — the same session the daily bars cover — so a stop that only printed pre-market does not count. Every verdict records which tape decided it (minute or daily), shown on the audit page. Those calls are listed on every audit page under “Not adjudicable”.

Limitations (disclosed)

Seeded scores are computed from public posts, so a pre-scrape deleted call can't be recovered; futures and options legs that can't be priced are excluded from the gradeable set. Latency-honesty, recency weighting and a variance penalty are planned for a calibrated v2 and are not yet in v1. Every value is immutable and auditable; the version stamp moves to v2 when the method changes.

Not investment advice and never a recommendation to follow anyone. Records are computed from public posts — see methodology. Full disclaimer · Corrections & removal