Methodology

How we measure Polymarket signal accuracy

Plain-language definitions for signal, follow-up, accuracy, cohort, and sample size. Honest limitations included.

Definitions

Signal
A timestamped alert emitted when a wallet's BUY activity in a single Polymarket market matches a public detection rule. Each signal records wallet, market, side, detected size, entry price, and emission time.
Resolved signal accuracy
Share of original signals proven correct after the underlying market resolves. Each original signal counts once. Unresolved markets are excluded from numerator and denominator.
Follow-up quality
Whether and how a wallet later updated the position the signal was based on. Reported separately from accuracy and never re-scores the original signal.
Cohort
A fixed slice of resolved signals reported with the same definition every period: all resolved signals; signals above USD 5,000; signals above USD 10,000.
Sample size
Number of resolved signals available for a cohort. Every published accuracy figure should be read together with its sample size and date window.

Detection rules

We run two BUY-only rules. SELL events are rejected before evaluation.

  • whale_accumulation - sustained directional buying by a single wallet on one market over a short window above a size threshold.
  • large_directional_entry - a single large BUY entry by a wallet on one market above a size threshold.

A signal is emitted only when one of these rules matches. Matching events are recorded with a stable signal identifier and timestamp at emission. Rule definitions and thresholds are versioned; changes are noted in the methodology changelog.

Signal accuracy

The original signal is counted once. Resolution comes from the underlying Polymarket market closing on a Yes/No outcome.

  • A signal whose direction matches the resolved outcome counts as correct.
  • A signal whose direction does not match the resolved outcome counts as incorrect.
  • A signal whose market is unresolved counts in neither numerator nor denominator.

Follow-up behavior is analyzed separately from this resolved outcome score. That keeps the track record readable: one original alert, one final market outcome, plus lifecycle context beside it.

Follow-up quality

Follow-ups describe what the same wallet did on the same market after the original signal. They are lifecycle context, not new signals.

Lifecycle event types

  • Added conviction - additional BUY on the same side.
  • Partial exit - SELL of part of the position.
  • Full exit - SELL down to zero on that side.
  • Take-profit - exit while in profit relative to entry. Inferred from price.
  • Stop-loss - exit while in loss relative to entry. Inferred from price.
  • Hedge - opens the opposite outcome without closing the original.
  • Reversal - exits original side and opens the opposite side of the same market.

What follow-ups never do

  • Become new signals.
  • Change the original signal's win or loss.
  • Combine into a single accuracy number.
  • Imply trader intent. We label observable trades and prices, not motive.

Cohorts and sample size

We publish three cohorts side by side, with the same definitions every period.

All resolved

Record building

Every signal that has reached market resolution.

Above USD 5k

Record building

Resolved signals with detected size above USD 5,000.

Above USD 10k

Record building

Resolved signals with detected size above USD 10,000.

The track record shows "Record building" while a cohort is still too small to be useful. The operational floor is 30 resolved signals, with 50 preferred for public display. Once a number is shown, it appears with the cohort label, sample size, and date window.

Unresolved markets

Unresolved markets are excluded from resolved accuracy by definition. The track record will list them as unresolved until the underlying market closes; at that point they enter the cohort and are counted once.

Honest limitations

These are real and we do not hide them.

  • Whale-sized flow is not insider information. Large directional entries can come from informed traders, uninformed conviction, market making, or hedging unrelated exposure. We detect size and direction, not motive.
  • On-chain visibility is not motive visibility. Lifecycle labels like take-profit and stop-loss are inferred from observed price relative to entry. The wallet does not declare intent.
  • Catalyst checks are limited. Internal news context is based on RSS headline matching. It is not proof that a market has or has not moved on public information, and we do not sell it as a product guarantee.
  • Execution and slippage are not modelled. Entry price is the recorded price at detection time. We do not simulate fills, slippage, or trading costs.
  • Premium does not promise higher accuracy. Premium gives fuller and faster lifecycle visibility around the same signals. The original signal's win or loss is identical for free and premium readers.

Disclaimers

WhaleFlow Alerts is independent and not affiliated with Polymarket. Content is informational only and not financial advice. Prediction markets carry risk and historical accuracy does not guarantee future results. See the full disclaimer.