How accurate is this?
Every figure on this site is read by software from a decision the tribunal published on GOV.UK, and every page links to the original. Software misreads things. This page says how often, how we know, and what we do about it. Figures here are counted from the database and the correction lists on 20 Oct 2026.
Hand checks
Each kind of figure was checked by a person reading the decisions, on a sample drawn before the software was tuned, and published only once the first pass met the bar (at least 90% right, nothing confidently wrong).
| What | Sample | Result |
|---|---|---|
| Rent figures and dates in Section 13 decisions (current, proposed and decided rent, effective date, notice date) | 110 decisions: the 50 most recently published, 35 from 2019 to 2025, 25 more read after the software was frozen | No rent or effective date wrong; the regexes alone read 85% of decisions fully right, the model cross-check fills the rest, and a value is kept only if it appears in the decision's text |
| Rent repayment orders: the amount ordered and the maximum | 60 decisions: the 40 most recent and 20 at random from 2019 to 2023, read before the software | Amount ordered and maximum right in over 90%; the amount counts in the statistics only where two independent readings agree |
| Application dates (for waiting times), Section 13 | 30 decisions at random | 29 of 29 dates given right; 1 left blank where the decision gives two dates |
| Application dates, rent repayment orders | 25 decisions at random | 19 of 19 dates given right; 6 blank, 1 stated date missed |
| The tribunal's good-condition rent and deductions | 30 decisions at random | Read and scored before use (details on the about page) |
Errors found and fixed
Some decisions carry a typo in the published PDF or the GOV.UK listing; some the software misread. Each class found so
far, the count, and where the fix lives (every correction is a line in pipeline/corrections.py in the
code, with the reason, and is re-applied after every run):
| Class | Found | How |
|---|---|---|
| Rent figures read wrong (weekly or quarterly rents stored as monthly, typing slips in the decision such as "£1ooo", a rent read from the wrong sentence) | 64 corrected; 57 read and kept | A check that lists rents under £300 a month for a whole home, over £15,000, rises over 150%, decided rents far from the proposed one, and whole-home rents under 40% of the local median; each one read |
| Decision dates (the GOV.UK listing's date, typed by hand, differs from the decision's own) | 133 corrected; 136 read and kept | A check against the publication date, the notice date, the case reference's year and the date in the decision; all 269 flagged read |
| Notice dates (a year typo in the decision, or another document's date stored as the landlord's notice) | 26 corrected; 26 read and kept | A check against the proposed start date, the decision date and the case reference; every decision counted under the new rules read as well |
| Effective dates (a tenancy start or an earlier decision's date stored instead) | 7 corrected | Seen while reading the above |
| Housing association decisions shown as a rent rise | 1 class, 431 decisions | Now labelled "open-market rent" with a note; never in the statistics |
Nightly checks
- Every new decision is read twice: by the regexes and by a language model that has to quote the passage holding each figure word for word; a figure the two don't agree on, or that isn't in the text, goes to review.
- Rents between £100 and £20,000 a month; a decided rent not above the proposed one under the new rules; dates that parse and agree with each other; a postcode that exists.
- The rent, date and notice-date checks above run every night; anything new they flag goes to review until a person reads it.
- A decided rent whose decision mentions a withdrawal, a strike-out, a dismissal, a lack of jurisdiction or "no determination" goes to review until a person confirms the tribunal really set a rent. Every shown decision was checked this way on 20 October 2026: 131 mention one of those and all 131 are real determinations (the words refer to an aside, an earlier application or a comparable); the eight that weren't were already in review.
- Decisions in review are shown with "figures not yet checked" and are left out of every rate and median.
In review today: 181 Section 13 decisions and 321 rent repayment order decisions (119 orders whose amount is unchecked), out of 3,375 and 1,629 shown. The queues are worked through by hand, London and the last two years first; each decision read is recorded as corrected or kept.
What is left out of the statistics, and why
- Housing association and affordable rents: the tribunal sets an open-market rent the landlord isn't obliged to charge, so the figures aren't comparable with private rents. Shown, labelled, not counted.
- Rooms in shared houses and rents that include bills: not like-for-like with whole homes. Shown, flagged, not counted (rooms are compared with rooms in the challenge pack).
- Register-only decisions (only the one-page notice was published): they have a decided rent but no proposed rent, so they count in totals, not in rates.
- Scanned PDFs we can't read: shown as "figures not available", counted in totals only.
- Fair-rent and section 22 cases mis-filed in the same GOV.UK category: different law, kept in the dataset, never shown.
- Rent repayment orders: the amount counts only where two readings agree; the maximum only where the decision states it as one figure; the share of the maximum only where both are clean.
- Waiting times: only decisions that state the application date (about 70% of rent decisions, 72% of rent repayment orders); cases still waiting, settled or withdrawn can't be in them, so recent figures are grouped by decision date.
Figures from fewer than 5 decisions are shown as "too few". Every figure carries its n. If you find an error, write to hello@rentrisecheck.uk: it will be checked against the original and corrected in the open. More on how the data is made.