Methodology & accuracy. How Baseline scores, and how we test it.

21 companies, 97 hard factual claims, drawn from live rationale panels and verified individually against a primary or authoritative source. Coverage was deliberately split across the universe rather than concentrated in well-covered names: 9 marquee developed-market companies and 12 emerging-market and long-tail names, where public data is thinner and error risk is highest. Analyses carried generation dates between March and July 2026.

A hard claim is one that can be checked and can be wrong: a figure, a date, a named transaction, a stated market position. Judgements are not claims and are not counted.

Claims were pulled from live product output through the results API, not from a curated sample prepared for the test. Each was traced to a primary source where one exists, and to an authoritative secondary source where it does not, then graded against what that source states. Grading was done to the source, not to plausibility.

An earlier 18-claim pass from 3 July was retired rather than added to this corpus. It was not sourced per claim, and three of its names were re-verified into this corpus, so including it would double-count and import a shallower standard of checking. This corpus stands alone.

Every analysis carries a generation date, so live market data is read as of that date rather than as current. Excluding the 3 unverifiable claims and the 5 market-data claims graded dated on that basis, 89 durable checkable claims remain. Of those, 73 verified (82%), 12 were imprecise (13%) and 4 were contradicted (4%).

The finding that matters most is the absence of fabrication. Not one claim in this corpus was invented. The errors that exist are of a different and less damaging kind: real facts attached to the wrong entity, real figures given the wrong label, real ranges overstated at the edges. Every one is listed below.

The corpus is 97 claims across 21 companies out of a universe above 3,600. It is evidence, not proof, and it is not a statistical guarantee about any company we have not tested. The emerging-market long tail is where our confidence is lowest, which is why it is over-represented here rather than avoided.

Baseline surfaces and screens candidates on qualitative grounds. It does not predict returns, and it is not a substitute for an analyst's own modelling, valuation or final judgement. Every error listed above is tracked as an open engineering item with a named fix, and this document is reissued as the corpus grows.