PunditScore
MethodologyEvidence-backed · No black box

How PunditScore works

Transparency over black-box scoring. Evidence over reputation. Comparability over anecdote.

Every PunditScore traces back to a dated claim, a real-world outcome, and the reasoning in between. This page describes exactly how the number is produced — and what it does and doesn’t claim.

01Four stages, kept separate

A pundit is assessed as three separable questions — did they say it, did it happen, how good was the call — so each can be checked on its own. The cardinal rule across all of them: retrieve first, reason second. The model never recalls a claim or an outcome from memory; it only works from retrieved sources.

01 · RetrieveExa

Search the open web for who the person is and the predictions they've publicly made — pulling the actual source documents, never the model's memory.

02 · ExtractOpenAI

Read only those retrieved sources and pull out dated, attributable, falsifiable claims — each with a verbatim quote and its source.

03 · GroundExa + LLM

For each claim, search for what actually happened and classify the real-world outcome against that evidence.

04 · ScoreFormula

The model judges each prediction against its resolved outcome; the KPIs and the composite are then computed by a fixed formula — not emitted by the model.

02The four KPIs

Every score decomposes into four measures. Directional, Timing, and Revision are the average of the per-prediction judgments across an analyst’s resolved predictions; Market Impact is judged at the analyst level.

Directional AccuracyDid the predicted direction of change actually occur? (magnitude/timing aside)
Timing AccuracyDid it happen within the stated or implied time horizon?
Revision QualityAre updates transparent and evidence-driven? Honest revision is rewarded; waffling is penalized.
Market ImpactHow widely cited and consequential was the call? For operator-advocates this reads as bias risk.

03The composite is a formula, not a guess

The headline PunditScore is computed from the four KPIs by a fixed, versioned weighting — it is not a number the model emits. That means anyone can re-derive it by hand, and the same inputs always give the same score.

COMPOSITE · CONFIG v0.1
0.35 · Directional  +  0.20 · Timing   +  0.20 · Revision  +  0.25 · Market Impact

Weights are normalized by analyst type. For Macro & structural analysts, whose calls run over long horizons, the Timing weight is lowered to 0.10 (from 0.20) and shifted to Directional — a decade-out structural call isn’t judged on timing like a near-term operational one.

04Score bands

85100
Exceptional
7584
Above Average
6074
Mixed
4559
Below Average
044
Unreliable

05How each prediction resolves

correctThe predicted direction/outcome clearly occurred within a reasonable reading of the horizon.
partialThe direction was right but timing or magnitude was off, or it only partly happened.
wrongReality contradicted the prediction.
pendingNot yet resolvable — the stated horizon has not yet arrived.
low-falsifiabilityToo vague to score cleanly, or the horizon has passed with no resolvable evidence.

06Where we say “we don’t know”

A score is only as strong as the resolved evidence behind it. When fewer than 3 of an analyst’s predictions have actually resolved, the scorecard is marked Provisional rather than presented as firm — many pundits (especially futurists) make calls that simply haven’t come due yet, and we say so instead of inventing confidence. Likewise, a prediction whose horizon has already passed can never sit as “pending”: it must resolve or be flagged unresolvable.

Scoring is pinned for reproducibility — a fixed model version, zero-temperature sampling, and a versioned weighting config — so the same analyst returns the same score for the same reasons. The archetypes we normalize by: Operational & industry · Macro & structural · Financial & Wall Street · Institutional & advocacy · Investor-banker & operator-prognosticator.

07What a PunditScore is not

It is not an absolute verdict on a person. It reflects the public, checkable record we could find — paywalled research, private notes, and off-the-record calls are not scored, so analysts whose work is mostly private will look thinner than their reputation. Scores are relative and evidence-bounded, calibrated against each other rather than against absolute truth. A score you can’t interrogate would be a failure of the design — so every number here opens to its underlying quote and outcome.

Explore analysts →Compare side by side