A guide to reading the ledger

Separate the estimate from the evidence.

A probability, a price, a source, and a performance score answer different questions. This is how to keep them separate.

Submit a corporate earnings-call mention event on the home page to request a forecast for its active word contracts. The report compares each word, its captured Kalshi probability and our estimate. Other event families are not currently supported.

Submitting an event requests a forecast; opening a report does not start new research or place a trade.

A probability is an estimate

A mention market asks whether a designated speaker will say a word or phrase during a specified event. Its exact contract rules determine which statements and events qualify; a word appearing in a transcript is not, by itself, a settlement decision.

A 70% model estimate means an estimated 0.70 chance that the contract resolves YES under its exact rules. It is not a guarantee, an accuracy score, or a return.

Synthetic teaching example · not a forecast

70%

Across ten genuinely comparable events, seven YES outcomes would be an expectation under these probabilities—not a promised observed count. Even that long-run interpretation needs calibration evidence.

Current forecasts focus on corporate earnings calls. Where comparable transcripts are available, they provide additional context. Otherwise, the model uses captured contract context and general company/word priors, not invented recent research. Unfamiliar or unlikely words can still receive uncertain estimates; genuinely uninterpretable targets remain explicit abstentions.

A supported issuer-history method can instead estimate directly from complete prior same-company calls. Its Beta(1,1) frequency estimate is (matching calls + 1) / (complete calls + 2), using the contract’s representative/operator scope and excluding analyst questions. This is a disclosed historical-frequency prior, not a new LLM response or a proven calibrated probability. Partial calls and unresolved speaker attribution cannot establish absence.

Contracts from the same physical event are dependent; ten words in one call are not ten independent events. These estimates are not established as calibrated or better than the market.

Freeze what was known

Current earnings reports are as-of forecasts. The information cutoff identifies the captured input snapshot; the generation time identifies when the estimate was produced. Each comparison shows when its Kalshi quotes were captured. A later quote refresh does not change the original model probabilities or first selection snapshots.

A separate fixed-horizon research benchmark

The research benchmark, not the current as-of report, uses a corroborated scheduled start, S. Its information cutoff C is 65 minutes before S and publication deadline D is 60 minutes before S. Operational reports are not relabeled as that benchmark.

At or before C
Capture admitted source versions and baseline quotes. An old page date cannot excuse a later capture.
Strictly before D
Observe the original forecast’s actual public bytes. A timestamp equal to the deadline is too late.
After publication
Keep the original probability, inputs and baseline. Later quotes and page edits do not replace them.

Scheduled start is not actual-start evidence. If the event starts earlier, append a timing qualification correction; never backdate a forecast. Unknown timing is a visible limitation.

On an event page, Open captured artifact points to the retained bytes. Original URL · may have changed opens today’s source page. They are not interchangeable. A retained excerpt is not necessarily a full source archive.

Compare captured probabilities

The Kalshi probability in a current earnings report is the captured midpoint of valid, non-crossed YES bid and ask quotes. A missing, invalid or crossed book has no comparison probability and cannot create a selection. This midpoint is not an executable offer or a promise of liquidity.

Synthetic teaching example · not market data

Model estimate68%
Captured Kalshi probability52%
Difference+16 pp

The difference in this snapshot is 16 percentage points, not 16% profit or an executable return. A later price of 64¢ does not enter this comparison for the original snapshot. A refreshed report may show a newer quote, but it never rewrites the first logged selection.

Source quality is not a reason to buy

Synthetic teaching example: a 55% model estimate with a YES ask of 56¢ (bid 48¢) does not recommend buying YES—even if the evidence quality is high. The model’s estimate is below the executable ask before fees. Evidence quality describes support, not price advantage.

The ledger does not establish order-book depth, fills, fees, execution delays, or profitable trading. Current executable prices must not be inferred from a frozen quote.

The separate research benchmark freezes its baseline at or before C, with a five-minute freshness limit and a maximum $0.10 spread. Those benchmark eligibility rules are not a claim that every current-report midpoint meets the same conditions.

State the proof you actually have

Only claim the evidence actually retained. A hash identifies bytes; it does not, by itself, prove when they existed publicly. The research record distinguishes these tiers.

  1. 1

    Internally recorded

    internal_recorded

    Self-recorded metadata describes claimed execution order. It does not independently prove timing or public availability.

  2. 2

    Platform reader receipt

    platform_reader_receipt

    The operator’s anonymous reader observed matching public bytes. Retained response headers and observation times are not independent clock proof.

  3. 3

    External pre-event witness

    external_pre_event_witness

    An actual independent receipt must bind the exact core hash before D. A self-authored date, a generic repository timestamp or an after-event archive is insufficient.

Selections and probability scores differ

The earnings selection ledger logs a YES selection only when our forecast is strictly higher than its captured Kalshi probability. A 30% estimate versus a 20% market probability qualifies, even though the estimate is below 50%. Equality, lower estimates and missing probabilities do not qualify.

Each word has one immutable first positive-gap snapshot for this policy. Repeated submissions do not duplicate it or replace its original probability and quote. Later reports may differ; the running record keeps the first selection.

A finalized YES is a win and a finalized NO is a loss. Win rate is wins divided by settled selections; it is unavailable before any selection settles. Unresolved, disputed, amended or changed-rule results remain pending rather than receiving invented labels. Corrections preserve earlier observations.

This is a conditional word-selection win rate, not calibration, independent-event accuracy, proven market advantage or profit.

Separate fixed-horizon research scores

Finalized binary exchange results are the scoring labels: YES = 1, NO = 0. For each contract, Brier error is (p − y)². Lower is better.

Average contract errors within each physical event, then weight events equally. Compare paired model-minus-baseline event errors on the same declared cases. A negative paired difference means lower model error on that cohort—not a proven trading edge.

  • Pending: no finalized label, so no invented result. Primary scores wait for complete supported, on-time issued event outcomes.
  • Abstained or missed: retain coverage failures. An unpublished fallback does not become a headline forecast.
  • Missing baseline: disclose the unmatched case. Keep it out of the paired comparison, not out of the opportunity ledger.
  • Excluded: apply predeclared support and timing rules, not retrospective removal of bad predictions.
  • Synthetic: exercise arithmetic and record handling; never count as real performance evidence.

Calibration, probability skill, arithmetic reproduction, public timing and trading value are different claims. Reliable model promotion needs a registered forward comparison with enough independent events for the declared effect and uncertainty rule. A small sample may remain inconclusive.

Read coverage and measured results

Correct the outcome, not the forecast

Append exchange corrections and score revisions. Keep the old observation and original model probability. Transcript review can explain support or ambiguity; it cannot silently remove an unfavorable finalized label.

Synthetic arithmetic · excluded from production totals

Original model probability: 70%. Frozen market baseline: 60%.

Initial outcome: YES
Model Brier 0.09; market Brier 0.16. Model − market = −0.07.
Corrected outcome: NO
Model Brier 0.49; market Brier 0.36. Model − market = +0.13.

The 70% forecast does not change. Both outcome observations and the earlier report remain part of the history.

Inspect the files, not just the screen

The earnings selection record exposes original probabilities, captured quotes, source digests and outcome corrections. These allow the selection rule and running win rate to be checked without regenerating an LLM response.

Full publisher transcripts are not redistributed. Issuer-history audits expose source URLs, digests, counts and limited attributed excerpts; reproducing their lexical observations requires matching original publisher documents. Private retention is not a claim of a public full-source archive.

The separate research reproduction package contains fixed-horizon opportunities and dispositions, append-only outcome observations, the protocol and artifact digests. Its scores are not blended with the operational selection win rate.

Retained artifacts are necessary to inspect exact source bytes. A dead link, unretained source or missing version limits reproduction. Identical arithmetic does not establish calibration, forecasting skill, independent timing, or successful unattended operation.

Comprehension verification

These explanations have not yet passed the five intended-user comprehension tasks. Automated tests and visual checks are not evidence that readers understand the distinctions.