What the model calculates, what AI drafts and what a person checks before anything is published.
Touchline Atlas uses software at every step, and we want you to know exactly where. The probabilities come from a statistical model, AI tools gather sourced news and draft text, and an editor checks every claim before it goes live. This page sets out those rules and how we correct mistakes.
Our statistical model (Elo and Dixon-Coles), run by code with fixed rules
Automatic checks (probabilities add up, data is complete), then locked before kickoff, normally 24 hours ahead. Nobody edits numbers by hand.
Team news dossier
AI tools (Claude, Codex) collecting facts from dated sources, each with a link
Automatic source checks; items without a dated source are dropped. Matches without verified news are labelled “limited context”.
Match analysis text
Drafted by AI from the locked numbers and the dossier
An editor checks every claim against its source before publishing. The draft may not add facts that are not in the data.
AI forecasts
Claude and Codex, each giving its own probabilities
Graded in shadow; they do not change the published numbers until they pass the rules on the methodology page.
Guides, answers, reports
Drafted with AI assistance from our method and data
An editor fact-checks each one; the reviewer and date appear on the article. Changing numbers are filled in from live data, not typed.
Grading and track record
Code, from the locked forecast and the final score
A score counts only when two independent sources agree. Every change to a graded result is logged in public.
1
Sources
We record facts, such as a score, a kickoff time or an injury, with a link to where they came from. We don’t copy other sites’ text, we don’t use anonymous rumours, and team news needs a dated source.
2
Language
We write probabilities, not promises. No “sure bets”, “bankers”, “tips” or “fixed” matches, no betting calls to action, and confidence labels always mean a stated probability range.
3
Independence
No money from bookmakers, no affiliate links and no sponsored forecasts. We are not affiliated with any club, league or AI company, and nobody can pay to change a forecast.
4
AI disclosure
We say which AI tools we use and what they do. AI-generated images are labelled as such, and AI text is never published without an editor’s check.
5
Images
Photos are used under their licence and credited; images made with AI are labelled. We do not use club crests or competition logos: each club appears with a plain shield badge drawn for this site.
6
Prompts and content
Text inside sources is treated as data, never as instructions to our AI tools, and every AI step has a written prompt that we version and review.
How we correct mistakes
Data errors (a wrong score, time or name) are fixed in the next batch.
Graded results are re-graded only when the official result changes or our grading was wrong. Each change is listed under Record changes, with the date and reason.
Locked forecasts are never edited. If one was published in error, it stays visible with a note, and its hash still matches the ledger.
Articles are corrected in place with a dated note at the end describing what changed.
Who is responsible
The site’s operator, The Touchline Atlas, is responsible for everything published. Match analysis and articles are reviewed before they go live by Mr. 07, the pen name of the site’s editor; the reviewer and date appear on each article and analysis.
There is no second reviewer. When the editor cannot review a match in time, its page shows the statistical forecast without written analysis and says so; the forecast itself is still locked and graded as usual.
Can we count your visit?With your permission we use Google Analytics, which sets cookies, to see which pages people read. No ads. The site works the same if you say no.