One stream of the GradedLP record: Poolwatch's TOP picks vs a blind CONTROL on its EVM chains, graded daily. The figures below load from record.json (machine-readable, regenerated daily). Every other stream's verdicts are on the verification page.

Poolwatch Selection Record — TOP vs CONTROL

Were the calls right? Poolwatch's daily picks vs a blind control drawn from the same pools.

This page scores call accuracy for Poolwatch: a call is correct when the position did what its entry call said. The site's headline selection edge is a different measure — the share of positions that finished with positive net return (fees minus divergence loss), picks vs blind control, across Poolwatch's EVM chains — shown on the home page. Both come from the same graded positions. Counts here are the current score version (v5, graded since 2026-07-18); the home page also counts the few earlier-version calls. This is one stream, not the whole GradedLP record: every stream's verdicts are on the verification page.

All roles combined: —

The picks · TOP

pools the model flagged from the WATCH tier

—

gap

—

Baseline · CONTROL

blind selection — a randomly sampled control set

—

The picks

—

Baseline

—

All roles

—

50% · coin-flip

—

Breakdowns — every rate over its own denominator

By chain

By model version & role

The ledger — every cohort position, no dedup, hits and misses

Chain

Role

Outcome

Entry verdict

—

Opened	Entry call	Entry APY	Outcome	Realized net	Pool	Chain	Role	Graded

How a grade is decided

Each pool gets a verdict at entry — OK, CAUTION, or AVOID. Fourteen days later the position's realized net LP return (fees minus impermanent loss) is measured, and the call is scored by a single zero-crossing rule:

OK / CAUTION → correct if realized net ≥ 0 · AVOID → correct if realized net < 0

The grade is always scored against the verdict the pool had at entry — not its current rolling verdict. (The live board shows today's verdict; the grade judges the call that was actually made.)

CAUTION shares the OK pass-line

CAUTION is a tradeable-but-flagged call, not a "stay out." It is scored on the same "held at or above water" line as OK; it is not treated as an AVOID. This is a deliberate, frozen choice, not a per-case adjustment.

Void handling

A position with no usable price series (scorer abstained / EXCLUDE), or one that died before resolving (DEAD), or a resolved position whose entry snapshot is missing, is marked void with a stated reason. Void and pending rows are shown in full but are excluded from the accuracy denominator — they are never guessed into a hit or a miss.

A miss is a miss

There are no exemptions for external events — a depeg, an exploit, a market crash. If an AVOID pool went up, that is a miss; if an OK pool went down, that is a miss. Cause-of-miss may be added later as optional commentary, but commentary never changes the grade.

Verifiability

This record is append-only: resolved rows are published as they mature and are not edited afterward. It is not claimed to be immutable or tamper-proof — it is a public log kept in good faith, checkable against the methodology below.

Methodology: METHODOLOGY.md Scoring model: v5, frozen in code Public pre-registration: live

The full pre-registration record — historical freeze (2026-07-23) + ongoing tier-3 commits — is published at github.com/GradedLP/publiciq. Resolved tier-1 calls here are recorded results; tier-2 and tier-3 entries are forward-validated against that public record.

—