[{"data":1,"prerenderedAt":4},["ShallowReactive",2],{"article-body-the-league-table-rescored-daily":3},"\nAsk a vendor why their model is right for your problem and you will\nreceive an architecture story: transformer this, foundation that,\nproprietary the other. Architecture stories are opinions. Which model\nbest predicts a given target, on given data, at a given horizon, is an\nempirical fact, and an unstable one. It changes as the data changes,\nsometimes within a quarter.\n\nThe forecasting literature has made this point for decades, most\nfamously through the M-competitions, where simple methods repeatedly\nembarrassed sophisticated ones on real series, and where no single\nmethod dominated across question types. A vendor whose product is one\nmodel, however capable, has answered an empirical question with a\ncommitment made before your data was seen.\n\n## How the bake-off works\n\nFor every prediction question, Prophesee's Foresight engine runs a\nstanding competition rather than a coronation:\n\n- **62 models from 13 families** compete: statistical methods, tree\n  ensembles, neural approaches, and hybrids, spanning the five question\n  types an operating business actually asks. How many? Which ones? How\n  risky? When? Why?\n- **Scoring happens on unseen data**, in rolling backtests, with the\n  placebo harness behind it so a lucky fit cannot take a seat it did\n  not earn.\n- **The table is re-scored daily** as fresh outcomes arrive. A champion\n  that drifts loses its seat to the contender that has not, without a\n  meeting, without a migration project, without anyone having to\n  defend last year's architecture choice.\n- **Every question gets its own champion.** The model that wins weekly\n  demand for fast movers is routinely the wrong model for intermittent\n  spares, for regulatory case timing, or for attributing a variance.\n  One question, one table, one current champion.\n\n> Nobody has to believe in a model family. The table settles it, and\n> keeps settling it.\n\n## The contender that keeps everyone honest\n\nEvery table contains one entrant that does not care about elegance: the\nnaive baseline. Last year plus ten percent. Same as last period.\nWhatever a sensible person would guess without a model.\n\nThe rule is absolute. A champion must beat naive on the question's\nagreed error metric, or there is no champion. This rule does real work.\nResearch on forecast value repeatedly finds that a large share of\nsophisticated forecasts fail to beat naive methods, a result most\norganisations have never tested against their own numbers. When nothing\nbeats naive, the honest output is not the least-bad model. It is the\nsentence: this target is not predictable yet with this data, and here\nis what would change that.\n\nThat sentence prevents the quiet catastrophe of enterprise forecasting,\nwhich is not bad models but confident automation of guesswork:\ninfrastructure, review meetings and decisions built on numbers that a\ncopied-forward spreadsheet cell would have matched.\n\n## What this replaces\n\nThe league table replaces two familiar failure modes. The first is the\ndata science queue, where each new question waits months for a\nhand-built model, which then ossifies because nobody has time to\nrevisit it. The second is the platform monoculture, where every\nquestion is answered by the vendor's one architecture, at whatever\nquality that architecture happens to achieve on it.\n\nAgainst both, the competitive mechanism is boring, continuous and\nauditable: every model's history is on the table, every substitution\nhas a scoring reason, and the answer to \"why this model?\" is never a\nbelief. It is a row.\n\nThe league table runs as standard inside Prophesee's Foresight engine.\n62 contenders, re-scored daily, naive baseline enforced. To see who\nwins on your data, [start here](/contact).\n",1786799035557]