How Oraca works — and why you can trust it.
Oraca isn't one model. It's more than a dozen systems that feed one another — measuring how good a player is, projecting how he'd translate to your league, forecasting what he'll become, judging whether he fits the way you play, and mapping how you'd actually sign him.
These pages are written for two readers: the director of football who wants the thinking, and the analyst who wants to pressure-test it. The intuition is in plain English; the published research each model is built on is cited at the foot of every page; and where a model has limits, we name them rather than hide them.
“Why should I trust this?”
Because every claim is tested the hard way. Signals are validated out-of-sample — trained on older seasons, judged on later ones the model never saw — against what players actually went on to do. We surface a confidence level when the data is thin, and we publish the academic research each method rests on. A model that tells you its own limits is worth more than one that claims to see everything.
“Couldn't I build this myself?”
The map below is the honest answer. It isn't a single rating in a spreadsheet — it's a dozen models built on years of reconciled history, each one a research problem in its own right, feeding the next, rerun and re-checked every night. The unglamorous parts — making one player one record across every source, work-permit eligibility — are where most of the work, and most of the years, actually go.
The whole machine, on one page.
Each layer feeds the one below it. A player's rating informs his projection; his projection and his forecast inform his fit; his fit and your network decide whether he's a signing you can make.
Recomputed and checked every night — the whole pipeline reruns daily, and refuses to publish a day's numbers if a correctness check fails, so a broken night is loud, not silent.
Highlighted systems have a deep dive below. The rest are part of the same machine — the full picture is more than a dozen models feeding one another, not a single score.
How we keep ourselves honest.
The difference between a model that looks good and one that is good is the test you put it through. Four principles run under everything here.
Out-of-sample, always
A model is judged on data it was never trained on. We train on older seasons and test on later ones, predicting what a player actually went on to do — not how well the numbers fit the past.
Leak-free by construction
Every forecast is anchored to a fixed point in time and is forbidden from peeking at the future. A signal that secretly sees the answer always looks brilliant and is always worthless.
Honest about uncertainty
When a result rests on thin data, the product says so — a confidence level and a sample size, surfaced rather than smoothed over. Stated uncertainty is a feature, not an apology.
We name the limits
Every page ends with what its model can't do. Production isn't tracking; coverage is the leagues we hold; some work is still research. We'd rather tell you the edges than pretend there are none.
Past model performance is not a guarantee of future results. Oraca narrows the field; your scouts do the part no model can — watching a player live.
The deep dives.
One page per system — the thinking, the method in plain English, how it's validated, what it can't do, and the research it's built on.
Measuring players
Player Rating
A cross-league, position-aware ability score built from three independent signals — opponent-adjusted production, impact on the scoreline, and ceiling games — plus a separate undervaluation signal that flags players whose ability runs ahead of their club, age and division.
Club & League Strength
How every club is placed on one global strength scale across divisions and borders — a margin-aware base rating with explicit draws and home advantage, calibrated to an independent reference at the top, and a cross-league exchange rate learned from the players who move between divisions.
Market Values
How each player's value is sourced — the observed market consensus rather than a guess, resolved primary-then-secondary-then-last-known by a single writer over never-mutated source histories, matched precision-first so a price lands on the right player.
Projecting players
Cross-League Projection
How a player's performance profile is translated from one league into another — an empirical translation learned from real movers, anchored on whole promoted and relegated squads to defeat selection bias, with confidence levels surfaced wherever data is thin.
Trajectory & Ceiling Forecasting
How Oraca forecasts a player's future, not just his present — a causal model that reads an entire career match by match, a multi-horizon forecast of where he'll be in one to five years, and a separate model for the right-tail question of how high he could ever peak.
Career Outlook
A career foundation model — the same family of AI behind modern language models — trained on two decades of month-by-month professional careers. It simulates hundreds of plausible futures per player and publishes calibrated probabilities, validated by rewinding to 2021 and 2017 and grading its picks against what really happened, then forward-logged monthly from June 2026.
Fitting your club
Recommendations
How the club-conditioned shortlist is built — a transparent weighted sum of need, upgrade, model consensus and affordability (minus a work-permit penalty), scored against your own squad and league, laid out like a team-sheet with a plain-English reason, and wrapped in a board / assign / recycle / discussion workflow.
Tactical & Realistic Fit
How Oraca models playing style and fit — eight style dimensions that resolve into recognisable team archetypes, a per-player fit estimate by formation, and a signing shortlist that keeps level and fit separate while folding in tier realism, age and signability.
Signing players
Connection Degree
How the network “degrees of separation” feature works — a career-overlap graph where a link means the same club at the same time, computing first- and second-degree paths from your club to any player through teammate and coach bridges, every path fully attributed.
Work-Permit Eligibility
How UK work-permit eligibility is computed — the FA's GBE points tables transcribed verbatim (auto-pass at 15), the four-route ESC fallback, UK/Irish exemptions and FIFA bands — a faithful regulatory calculator whose real difficulty is assembling complete, dated international appearances for every nation.
The foundations
Identity Resolution
How the same player and club are recognised as one record across independent data sources that share no common ID — record linkage with date-of-birth blocking, a nickname-aware name signature, a human review queue for ambiguous matches, and nightly fail-closed checks that one player stays one record.
How the Data Stays Correct
How every model is rebuilt from raw data on a fixed nightly schedule in a verified dependency order, with a fatal core and tolerant edges, ending in fail-closed checks — no duplicates, no broken references, no lost coverage — so a broken night is loud, not silent.
See it on your own shortlist.
Oraca is in private beta with a small number of clubs. Tell us where you work and we'll be in touch about getting you in.