Court case verification methodology
Some of our legality pages carry a short, anonymized account of a real prosecution, so a reader can see what the law described above it actually did to somebody. This page explains what has to be true before one of those appears, and what we do when it is not.
The most important thing on this page
Our engine currently publishes no autonomously extracted records. Every published record to date was authored by hand and then verified against its source document by the automated checks described below. We have built the machinery to extract records from judgments automatically. Nothing it has produced has ever been good enough to publish, so nothing it has produced has been published.
We would rather say that plainly than imply a scale we do not have. The number of verified records we hold is small, and it is small because the bar is where it is.
Withhold by default
A record has to pass every check. There is no score, no majority and no threshold. A record that fails one check of twenty-one is withheld exactly as firmly as one that fails all twenty-one.
- Withholding is the default outcome, not the exception. Any path that is not an explicit pass on every check produces a withheld record, including an error inside the checking code itself.
- Every check runs. The process does not stop at the first failure, so a withheld record comes back with a complete account of what was wrong rather than one symptom.
- A withheld record is logged with its reasons. It is not deleted and it is not a permanent ban: withholding is a statement about the evidence we hold today, so a better source and a fresh run can still clear it.
The twenty-one checks
These are the checks the engine actually runs, grouped by what they test. Each one is verified against the live document at the moment of checking, not against a stored copy.
Source resolution
Whether the cited document is actually there, and whether it is actually about this case.
- Source resolves. Every citable URL is fetched and read at verification time. A dead link, a paywall or an unreadable scan fails.
- Source content matches the case. The fetched text has to carry the substance, the year, an outcome and the location the record claims.
Source tier
Whether the publisher is one we are willing to cite at all.
- Tier qualifies. One government, court or official source is enough. Two independent established news outlets that agree are enough. Everything else is a lead we may follow, never a citation.
Does the source support each claimed fact
Taken fact by fact, not record by record. A record can be right about the outcome and wrong about the weight.
- Outcome explicitly supported. The stated result has to appear in a source that is asserted to back it.
- Charge supported. Every section number in the cited provision has to appear in the live document.
- Quantity supported. The figure, its unit and its hedge all have to match. A source saying not less than 367.2g does not support a record saying 367.2g.
- Statutory limit supported. Where a record says a sentence sat at a legal floor or ceiling, that has to be a quotation from the document. We never work it out ourselves.
- Global sentence not understated. Where a document states a total sentence larger than the single charge we are describing, the record has to carry the total. Leaving it out would be true and flattering, which is the failure this check exists to stop.
Cross source agreement
Whether the sources we hold contradict each other.
- Cross source agreement. Conflicting outcomes, years or dispositions between sources fail the record rather than being reconciled by us.
Absence of personal names
Checked twice, by two separate pieces of code, on purpose.
- No named individual. A name anywhere in the record fails it. The publishing step then runs an independent second name scan over the field that renders as the who, because that field is the one where a miss is both irreversible once indexed and invisible in a diff full of legal prose.
Required fields
Whether the record is complete enough to mean anything.
- Required fields present. A missing field holds the record. It is never filled in with a plausible value.
- Format and style.
Duplication
Whether we already hold this prosecution under another write up.
- Not a duplicate. The test follows the correction chain, so a corrected record is recognised as the same case rather than as a second one.
Jurisdiction and scope
Whether this case belongs on this site at all.
- Jurisdiction matches. The country has to be one we publish a legality page for.
- Substance in scope. The drug has to be one of our substances.
- Recency in range. Cases before 2000, and cases older than twenty five years, are out of range because the law described no longer resembles the law in force.
Outcome attribution
Whether the sentence we are about to print belongs to the person and to the drug we are about to attach it to. This is the largest group and it is the one that holds the most records.
- Outcome bound to one defendant. In a judgment covering several people, the outcome has to bind to exactly one of them.
- Disposition region names one sentenced person. The part of the document that carries the sentence has to name a single sentenced person.
- Substance is the case's drug. A drug named in passing, in a cited precedent or in a statutory schedule, is not this case's drug.
- Charge's own drug, not an incidental item. The same test again, applied per charge rather than per record.
- Every charged drug resolves. A drug the document charges that we cannot resolve to a known substance holds the record rather than being dropped from it.
Checks are re-run before anything is published
A record that passed on the day it was written is not a record that passes today. Links rot and documents are replaced. Every record is re-fetched from its live sources and put through all twenty-one checks again at the moment of publication, and nothing reaches the site on the strength of a past result.
The audit log
Every decision the engine makes, whether it published a record or withheld it, is written to an append-only log with the reasons.
- Nothing is ever rewritten. A correction is a new entry that points back at the entry it replaces. The original stays in the log and stays readable.
- Because of that, a record has a history rather than a state. We can show what we said, when we said it, and what changed our mind.
- A record that cleared every check and is nevertheless wrong is retracted by a person, and the retraction is an entry like any other. Passing twenty-one checks is not proof that a record is true, and we do not treat it as one.
The collection is incomplete
This is not a comprehensive database of drug prosecutions and it does not attempt to be one. It is a small set of individually verified records, and its coverage is uneven by construction.
- For the United Kingdom, note that Crown Court judgments are not routinely received by Find Case Law. Most drug prosecutions in England and Wales are tried in the Crown Court, so the great majority of them will not appear in the archive we search, and cannot appear here either.
- The absence of a case on one of our pages means only that we have not verified one. It never means none exists. That is why a page with no verified case shows nothing at all rather than an empty state.
Corrections
If a record here is wrong, tell us and we will fix it or remove it. Email corrections@2cb.com with the page URL and what is wrong. See our editorial standards for what happens next.