Publishing standards

How we publish, review, and correct our work

CFB Power Board publishes original model calculations with visible sources, reproducible methods, and an explicit distinction between automated output and reviewed analysis.

Sources and original work

Scores, schedules, venues, team identities, and available conference information come from College Football Data. These source facts are separate from our own ratings, schedule-strength calculations, résumé metrics, weekly analysis, and research comparisons. Team logos identify schools; they do not imply endorsement.

Methods describe the data sample and settings used. Synthetic worked examples are labeled illustrative. A research rerun of known results is described as retrospective sensitivity, not as a successful forecast. Coverage and source corrections can change a result.

Automation and AI assistance

Software calculates the ratings and builds data-based reports and research commentary from structured templates. No reporter is implied to have watched the games. AI assistance is used in software and content development; it is not a source of scores or evidence of human editorial review. The published calculations use the model's game dataset.

Automation can check whether a calculation runs and whether required data is present. It cannot establish that an explanation is complete, that a source feed has every game, or that an article has received a person's sign-off.

Review states

Automated analysis · human review pending: Computed output and template-based interpretation. No personal review is claimed.
AI-assisted article · human review pending: Written or assisted explanatory copy without a recorded personal sign-off.
Human review recorded: A named reviewer and actual review date are recorded for the displayed content revision. Changing that revision requires a new review before the reviewed state can apply again.

A personal author byline requires an explicit authorship record. Creating the model does not make its creator the personal author of every generated article. The model creator profile identifies Andrew Lemons's established role.

What a review should verify

Match prominent scores, records, ranks, sample counts, and team links to the underlying publication.
Reproduce the central calculation and check the comparison population, settings, and selection of examples.
Verify the information cutoff for ratings and probability calibration. Separate walk-forward reconstructions, current retrospective résumé calculations, and forecasts with recorded pregame publication evidence.
Check that the interpretation follows from the evidence, includes material uncertainty, and does not confuse mathematical uniqueness with measured cross-division strength.
Record the actual reviewer, time, and content revision. Automated tests or a data refresh do not substitute for that record.

Timing and archive limits

Data refreshes and model builds run automatically when source results are available. Written analysis has no guaranteed daily or weekly human-review cadence. A reporting week identifies the games being discussed; its last kickoff is not an article's publication time. Rebuilding a page is not a new publication or a new review.

Weekly ratings are reconstructed using games through that week. They can change if source data or model code changes. Archived reports are not claimed to be immutable copies of what visitors saw at a historical moment. Research pages describe their analysis specification and data revision; the downloadable dataset is the current publication.

Corrections and model changes

Report a problem through the contact page or hello@cfbpowerboard.com. Include the URL, season, relevant team or game, the incorrect value or claim, and a source or reproducible calculation. A missing source game, a code defect, and a disagreement with a modeling assumption need different fixes.

For a substantive correction, the record should identify what changed, the affected page or dataset, the corrected revision, and the actual correction date. Model-setting changes should explain the old and new settings and their effects. Project updates distinguish implemented changes from work still planned. We do not backfill unverified publication dates or reviewer names.

Publisher and advertising

Lemony Click, LLC, a software studio, publishes CFB Power Board. The site may display advertising. Advertisers have no influence on model results. Advertising eligibility for reports and research requires a recorded review of the current revision. Advertisers do not set ratings, select outcomes, or purchase a higher rank.

Our privacy policy explains the site's data and advertising practices. See About for the project's scope and independence.