Working specification · August 14, 2026

Every number
needs a receipt.

This is the operating logic behind Cowboys Review. Version 0.1 is explicit enough to test, but remains provisional until the historical backtest and 2026 holdout are complete.

01

Three separate systems

A credible reporter can be repeated by a misleading headline. A careful outlet can aggregate instead of break news. One number cannot describe all three.

Story

Evidence supporting one atomic claim at a recorded moment.

Reporter

Resolved qualifying original reports, scored by category.

Outlet

Attribution, linking, headlines, labeling, corrections and chains.

02

Rate the claim actually made

“Dallas is trading a second-round pick for Player X and plans a four-year extension” is not one claim. The trade, compensation and extension plan can each resolve differently.

Source sentenceDallas is trading a second-round pick for Player X and plans to sign him to a four-year extension.

Atomic claim ADallas will trade for Player X.
Atomic claim BDallas will send a second-round pick.
Atomic claim CDallas plans a four-year extension.

To stop one detailed report from dominating a profile, each claim in an n-claim bundle receives weight 1 / √n.

03

Unresolved is not wrong

Verified

Strong primary or convergent evidence establishes the proposition as stated.

Incorrect

Strong evidence establishes material falsity, or a dated predicted outcome fails.

Unresolved

Evidence cannot close a private-process or otherwise unknowable claim.

Superseded

Evidence shows it was accurate when reported, then circumstances changed.

A failed signing does not prove talks never happened. A player going elsewhere does not disprove team interest. The proposition—not the fan expectation—is what resolves.

04

Reporter reliability, by category

Trades, contracts, injuries, coaching and draft reporting depend on different access. Category scores are primary; an overall score is secondary and requires at least three publishable categories.

Adjusted category rate(12 × category baseline + Σ weighted outcomes) / (12 + Σ weights)

Claim weight uses a 24-month half-life and bundle dampening. Small samples are pulled toward the disclosed category baseline.

<10Provisional
10–24Limited
25–49Established
50+Strong sample

05

Story credibility formula

30%E + 25%R + 20%C + 10%D + 10%S + 5%F − contradiction
30%Evidence quality

Records, named confirmation and sourcing detail.

25%Reporter reliability

Adjusted history in the matching category.

20%Independent corroboration

Distinct roots, not the number of rewrites.

10%Source directness

Distance from the primary record or original root.

10%Specificity

Clear subject, action, scope and time window.

5%Freshness

Current within the claim’s real decision window.

Contradictory evidence is documented separately and can subtract 0–30 points. A team or agent denial is evidence, but its incentives and direct knowledge must be assessed.

06

Verdicts need gates

95–100Confirmed

Primary confirmation required

85–94Strongly reported

Independent support or exceptionally direct sourcing

70–84Credible report

Clear sourcing; no decisive contradiction

55–69Plausible

Meaningful support; important uncertainty

35–54Unverified

Not enough evidence to rely on

20–34Weak / speculative

Thin sourcing, derivative chain or contradiction

0–19Contradicted

Strong contrary evidence; “False” needs final proof

“False” is intentionally difficult: it requires a final Incorrect resolution supported by Tier A or B evidence. A low score alone is not enough.

07

Outlet integrity is not scoop volume

A careful aggregator can have high integrity. We display original-report rate separately, then evaluate how the outlet handles what it publishes.

25% Attribution15% Direct links25% Headline fidelity15% News/opinion clarity10% Corrections10% Chain disclosure
Headline fidelity starts at 100. Turning “could” into “expected,” “interest” into “negotiations,” or a proposal into team action produces explicit, visible deductions.

08

Corrections stay visible

Material before/after text, timestamps, reasons and score effects remain in an append-only record. Challenges receive a public reference, evidence review and written decision.

Cowboys Review will publish reviewer agreement, overturned challenges, missing-source rates and score-band calibration. The first 50 qualifying pilot claims are double-coded.

See how we’re testing it