About MissedTakes
We keep score on the people keeping score. MissedTakes records checkable predictions made by named pundits, with a link to where they said it, and grades each one against what actually happened. The NFL was the proof of concept for the 2026 season; US elections, graded against called and certified results, is the second topic, and a pundit's record in one is never blended with the other.
Who runs it
MissedTakes is published by The GoodApps Company of New York, New York. It was built by Ben Garves, the company’s founder, who writes the site’s essays and white papers and reviews and approves every machine-drafted post before it publishes under his name (how our posts are made).
Why it exists
Punditry, in sports and in politics, publishes thousands of confident predictions a year and almost nobody checks them. Being wrong costs a pundit nothing, and being vague costs even less. We think the people who make their living forecasting football or elections should have a record, the way a weather service or a stock picker does — and that the record should be public, checkable, and computed the same way for everyone.
Two rules follow from that. We grade claims about outcomes, never claims about people: a pundit being wrong is not a character judgment, and the methodology is explicit that predicting football, or an election, is genuinely hard. And we do not rank on accuracy alone. The headline number is the MissedTakes Score, which asks how much better than a naive guess a pundit did on the same claims — because someone who only ever picks the home team can be right often and still tell you nothing.
How it is paid for
MissedTakes is built and paid for by The GoodApps Company. There is no advertising on the site and no paid placement: no pundit, publisher or league can pay to be included, excluded, or scored differently. Reading it needs no account. The optional member accounts (Picks) exist so readers can record their own calls, and are governed by their own privacy policy.
How the scoring is kept honest
Every number on the site is computed from the published methodology, which is shown in full and unedited. Every graded call shows its own arithmetic — what was claimed, what happened, what a naive guess would have scored — so a pundit who disagrees with a grade can see exactly what we compared their claim against. Amendments to the methodology are recorded in the document and trigger a regrade of the whole archive, so no two takes are ever scored under different rules.
Nothing reaches the public because a model extracted it. A person approves every pundit and every recorded piece before it is published, a pundit has no page until something on it has actually been graded, and a record needs ten graded articles before it is ranked at all. Below that bar the numbers are shown and labeled provisional.
How our posts are made
Most of what appears under Posts — the dailies, the weekly and monthly awards, the game digests, the season previews, the election briefs and the press releases — is drafted by a language model. It is given a packet of the site’s own data for the subject (the graded calls, the scores, the quotes and their sources) and a brief, and it writes from that packet alone: it is not asked to know anything about football or politics that the packet does not say. Every number in a post is a number on the site, and every quote links to the piece it came from.
A draft is then held until a person reads it. Two automatic checks run first: a wording check against a list of phrases we refuse to publish, and an output check for leaked markup, placeholders, a betting line, or a link that names the wrong person. Only after those pass does Ben read the draft and publish it, or send it back, and the post says so under its byline: “Drafted by a language model … then reviewed and approved by Ben Garves before publishing.” Essays and white papers he writes himself say “Written by Ben Garves.” We do not credit the model as an author, because the person who approved the words is the one accountable for them.
If a post gets something wrong, the corrections page is the way to tell us, and a correction is noted on the post.
Reaching us
If a grade is wrong, a quote is misattributed, or we misread a claim, the corrections page says how to tell us and what happens next; every graded call also carries a link to it. For anything else — press, a partnership, a question the methodology does not answer — use the contact form. Both land in the same queue, with a reference code you can hold us to, and a human reads every one.
New posts and the weekly awards are also published on the accounts below. They link back here; nothing is posted there that is not on this site first.
Using it from an AI assistant
Claude, ChatGPT and other assistants can read MissedTakes through a free, read-only connector: a pundit’s record, a game’s room, the calls on an election race, each with a link back to its page. How to add it.
What we collect about you
We count visits without cookies, and Google Analytics loads only if you say yes. The site notice is the full statement: how visits are counted, what Google Analytics receives if you allow it, and that there are no ad tags, pixels or social widgets anywhere.