A plain-language description of BOTBOT's methodology, published for scrutiny. See also the code of ethics.
BOTBOT processes court hearing lists that chambers lawfully receive (by email distribution today, and via the HMCTS Courts and Tribunals Hearings service under licence once granted), together with the subscribing chambers’ own records: its diary, cases and membership. No other data sources are used.
Lists arrive as documents in a small number of known formats. Deterministic parsers convert them to structured entries (court, courtroom, time, case reference, parties as published, hearing type, annotations). Parsing runs behind quality gates: output that loses material content is rejected and routed to fallback handling or manual review rather than used. The original document is retained unaltered and remains the authoritative record.
Each parsed entry is matched to the chambers’ own records using, in order of reliability: the unique case reference number, the court case reference, and finally surname within the same court centre and date. Matching links a listing to a case the chambers is already instructed in; it draws no inference about anyone. Entries that do not concern the chambers are not analysed further.
Where a matched listing differs from the chambers diary (courtroom, time, judge, hearing type), the difference is flagged for a clerk to review. BOTBOT never silently corrects either side. Time differences are only treated as meaningful in the window when courts finalise times, reflecting how listing actually works.
When a hearing needs cover, BOTBOT shortlists members of chambers using objective professional criteria: availability in the diary, practice area, seniority, panel accreditation, current workload, and realistic travel between court centres from a stated reference table. Artificial intelligence is used to weigh these stated facts and produce a ranked shortlist with written reasons. Identifying details of defendants are excluded from this processing, candidate barristers are represented by neutral tokens, and judicial identity is not supplied. Suggestions are advisory; a clerk makes every decision.
Listing data is retained for its operational currency and removed automatically on expiry or on instruction from the publishing service. Superseded versions of a list are replaced, with changes recorded in an audit trail.
No listing data is published, republished, sold or shared outside the subscribing chambers. No profiles, ratings, predictions or histories are built about any defendant, witness, judge or other person. No analysis is performed on individuals or groups; names are used only as matching keys. There is no public access, no search engine indexing, and no cross-chambers visibility.
Published hearing lists are not a complete or uniform record: not every hearing, court centre or list type is published through every channel, formats differ, and lists change up to and beyond publication. BOTBOT therefore treats the absence of a listing as meaning only that no listing has been received, never as evidence that a hearing is not taking place or has been vacated. The service shows which lists it has received for each court and date, so users can see what has and has not been checked, and the chambers diary is never altered on the strength of something missing from a list. BOTBOT produces no statistics or findings from listing data, so the incompleteness of the collection cannot distort any published conclusion.
Parsers are validated against real list formats per court centre. Suggestion criteria and model instructions are version controlled; changes are re-validated before deployment. Automated outputs are logged with their reasoning alongside the human decision, and the logs are reviewed periodically for systematic error or skew, as described in our code of ethics.
Questions: john@botbot.uk.