# Methodology **TaxiTools UK Taxi and Private Hire Licence Fees, v1.0.0** How the figures were collected, what was verified by a person, and where the limits are. --- ## 1. What counts as a source Only the licensing authority's own material: - its own website - its own PDF fee schedule - its own committee papers Nothing else was used. In particular, three source types were tested during collection and rejected: **Search engine result summaries.** Barnsley showed £180 in a snippet against £198 on the live page. Flintshire showed £610 against £730.73. **Trade press.** Figures quoted in the trade press were checked against the authority and did not match. **Automated summarisation of a PDF.** A summarising tool reading a fee PDF produced a validity period that was not in the document and denied an exemption that was printed in bold on the page. Each of those failures happened during this work. They are why the collection method is what it is. ## 2. How each authority was read 1. The authority's licensing pages were fetched directly, with a normal browser user agent, from a domestic connection. Many authorities reject automated requests from data-centre addresses, so this cannot be done from a server. 2. A person read the fee schedule and wrote a concise factual summary. 3. The source domain, the exact source URL and the date of reading were recorded with it. 4. Where the authority published a figure in more than one place, the disagreement was recorded rather than resolved. The 315 published records were read between **31 July and 8 August 2026**, with individual records re-checked since. Every record carries its own date. ## 3. Why the fee fields are prose Authorities structure charges in genuinely different ways: one, two, three and five year durations; separate hackney, private hire and dual rates; instalment options; application packages that bundle a test or a DBS; fees that differ by vehicle age, fuel type or emissions. Flattening that into numeric columns would require choosing which charge is "the" fee, and that choice is not present in the source. The summaries keep the structure the authority actually publishes, and the record says what each figure covers. A buyer who needs numeric columns for a specific comparison can derive them, and should decide for themselves which components belong in the calculation. ## 4. How ambiguity is handled Where a source is unclear, the record says so and `ambiguity_flag` is set to `flagged`, with `ambiguity_types` naming the kind of doubt. **Every one of the 315 published records was read in full and judged individually.** This is a manual review, not a keyword scan, and the judgements are held in a separate file so each can be inspected and disputed. **122 of 315 published records carry a flag (39%).** The kinds: | | | | |---|---|---| | `duration_unclear` | 57 | Licence period not stated or not confirmed | | `schedule_lapsed` | 28 | Named schedule year has ended, no newer one found | | `figure_absent` | 20 | A fee that would normally exist is not published | | `sources_disagree` | 17 | The authority's own documents state different figures | | `source_quality` | 13 | No fees page, scanned image, archive, or figures from forms | | `not_adopted` | 11 | Proposal or draft not confirmed as ratified | | `figure_uncertain` | 2 | The record itself says the reading warrants checking | **What the review establishes, and what it does not.** It establishes that every record was read and judged, and that the judgement is recorded. It does not find doubt that was never written down: if an authority's page was unclear and the researcher did not note it, no reading of the record reveals that. **The review was itself wrong first time and was corrected.** A first pass was made against a truncated view of each record and missed 13 cases where phrases such as "duration not stated" sat at the end of a long fee string. The superseded automatic scan caught all 13. Every record was then re-read in full. The build still reports where the old scan and the review differ, because that comparison is what exposed the error. ## 5. What was deliberately not done - **No figure was estimated, inferred from a similar authority, or filled in to make a table look tidy.** Where a fee could not be found, the record is marked `not_published`, meaning no readable schedule was found during this research — not that none exists anywhere. - **No fee was invented for an abolished authority.** Eden District Council was abolished in 2023 and its successor has not merged the legacy schedules. The record stays, unpopulated, explaining that. - **The summaries are written by TaxiTools rather than copied.** An automated check rejects the release if any exported field contains a quoted span of forty characters or more. **That check detects quotation marks. It cannot detect unquoted copying or close paraphrase**, so it demonstrates the absence of long quoted spans and not the originality of the wording. Proving the latter would need each summary compared against its source or reviewed by hand, and neither has been done across the whole set. ## 6. Limits a buyer should plan around **Fees change without notice.** Authorities revise schedules at committee, usually but not reliably in line with the financial year. **Each record states what TaxiTools found published when the source was checked on its `checked_date`** — not that the figure was correct then, and not that it remains so now. There is no mechanism in this version that detects a change between checks. **Source URLs decay.** Authorities reorganise their websites and withdraw committee papers. Some links in this set will stop working. **Some authorities block automated access.** Around one in twenty-five rejected automated requests during collection and had to be read manually. This affects how quickly the set can be refreshed. **Coverage is England, Wales and Scotland.** Northern Ireland licenses taxis through the Driver and Vehicle Agency under separate legislation rather than through district councils, so it is not comparable and is not included. **Refresh cadence is not yet committed.** The full set was collected in five working days by one person. That establishes it is repeatable, but no refresh interval is promised in v1.0 and none should be assumed. ## 7. Verification available to a buyer Every published record carries the source domain, the exact URL and the date it was read. Any figure can be checked at its source. That is the intended way to evaluate this data, and a sample of ten authorities is supplied for exactly that purpose. ## 8. How the release checks are themselves checked Every automated check in this package carries a deliberately failing example proving it can detect the fault it claims to detect. Nine such self-tests run before any result is trusted, and if one fails the report states plainly that every result below it is untrustworthy. This exists because two checks in this package shipped containing a regular expression that could never match anything. Both reported success on every run, precisely because they were broken. A check that has never been observed to fail establishes nothing.