01
A prediction enters once
A contract enters its country cohort at the first monthly snapshot where its estimated renewal date is 6 to 18 months away. The buyer identity, title, category, estimated date and estimation basis are frozen at entry.
Prediction ledger v0
Renewal estimates need a public scorecard.
Facts come from official notices. Renewal timing is our estimate. This is the protocol we freeze before checking whether those estimates were right.
Results pendingProtocol frozen July 2026
The earliest July 2026 cohort entries can mature no sooner than April 2027. We will not publish a headline precision result before at least 100 predictions have matured in that country.
What counts
02
The whole window must pass
We look for a qualifying successor notice from three calendar months before the estimate through three calendar months after it. A prediction does not mature until that seven-month window has fully elapsed.
03
The match must be reproducible
A match requires the same frozen buyer identity plus deterministic title and procurement-category similarity. Matching rules and aliases are locked before outcomes are inspected. There is no manual rescue for a near miss.
04
Every eligible result stays in
Later source corrections are reported as a tagged subset, not silently removed. Only duplicates, pre-snapshot withdrawals and identity failures discovered before outcome review can be excluded.
The primary result
Prospective forecast precision is the share of matured eligible predictions with exactly one qualifying successor notice. A contract notice, concession notice or competition-opening prior information notice can qualify. Awards, modifications, corrigenda, cancellations, direct-award notices and market consultations do not.
Results will include a 95% Wilson confidence interval. We will report both the numerator and denominator, never a bare accuracy percentage.
Country cohorts stay separate
- Netherlands
- Prospective cohort collection is under way. Results remain pending until the maturity and publication thresholds above are met.
- Germany
- The German evaluation will be pre-registered with its own frozen inputs, aliases and hashes before any outcomes are read and before any DACH accuracy claim is made.
Netherlands results will never be presented as Germany or DACH accuracy. Each country has its own cohort, denominator and result.
Corrections and material changes
We correct factual errors promptly. Send a written correction request to ravid@tendervane.com with the page URL, the disputed passage, and a supporting source. Material corrections are logged below as: date; page; corrected statement; reason; whether the headline, chart, or CSV changed. Typographical fixes that do not alter meaning are not logged. A report keeps its original as-of date unless its data are refreshed; a refresh receives a new as-of date and a dated log entry.
Published notices can arrive on TED after our measurement date. A re-run of the same query on the same as-of date can therefore return slightly different totals later. We treat that as a property of the source, not an error: reports keep the numbers that were true at measurement, and material drift is logged here.
Log
2026-07-18: /rapporten/stand-van-concurrentie-overheids-it-2026; late-arriving TED notices shifted a same-as-of re-run from 1,935 to 1,938 notices and the headline from 35.7% (640 of 1,795 lots) to 35.6% (640 of 1,797 lots). The published report keeps its 15 July measurement; headline, charts, and CSV unchanged.
No other material corrections logged.