Documenting an investigation

01Documentation is what makes a finding hold

An investigation is contested after publication, not before. The question then is not whether the reporter believed the finding but whether the finding can be shown: which record establishes it, when it was obtained, what was asked of whom, and what was decided at each fork.

Documentation is therefore not administration around the journalism; it is the part that survives. It also has an internal function, because a documented investigation can be handed over, resumed after months, reviewed by an editor and defended by a lawyer, while an undocumented one exists only in one person's memory.

02One structure per investigation

Set up the structure before the material arrives, because retrofitting it never happens. A workable arrangement separates raw sources exactly as obtained, processed working files, correspondence, the request log, the decision register, drafts, and the material intended for publication.

Naming should encode date, source and version, so that a file identifies itself without being opened. The test of the structure is simple: can a colleague, given the folder and nothing else, find the record behind any sentence in the draft within a minute?

03The source register

Every item of evidence gets an entry: what it is, where it came from, the date obtained, the licence or access conditions, who else knows it exists, and which claims in the draft rest on it. For datasets, add the download date and a stored copy of the original file; for documents, the provenance; for interviews, the date, the terms agreed and whether it is on the record.

The link from evidence to claim is the part most often missing and most often needed. When a body disputes a sentence, the answer must be a specific record rather than a search through a folder, and when a sentence has no record behind it the register makes that visible before publication rather than after.

04The request log

Access requests and information enquiries run for months and outlive anyone's memory of them. Log the date sent, the body, the exact wording, the reference number, the statutory deadline, all correspondence, the outcome, any exemption cited and the appeal position.

Across a series of requests the log becomes evidence in itself. It shows which bodies answered within the deadline, which applied exemptions expansively, and where records are held in a form that frustrates release. Each of those is a reportable observation about how public administration behaves.

05The decision register

Investigations are a sequence of judgements: which definition to use, which records to exclude and why, whether two entries describe one person, whether to name someone, what to redact, when to approach a subject. Each should be recorded when it is made, with the reasoning, because reconstructed reasoning is unreliable and looks like justification.

Two categories deserve particular care. Analytical choices that were tested and dropped should be recorded, so that the published specification cannot be a quiet selection of the most striking one. And ethical decisions — a withheld name, a cropped image, a delayed publication — should carry the reason, because those are the choices a reader or a regulator is most likely to question.

06Security of the material

Research files should be encrypted at rest and in transit, access-limited to those who need them and separated from general working documents. Devices need full-disk encryption, backups need the protection of originals, and shared-drive permissions should be set deliberately rather than inherited.

Source protection is an operational matter as much as an ethical one: which channel is used, what metadata it leaves, what remains on a device that could be lost or examined, and whether a source could be identified by elimination from details in the published piece. Minimising what is held reduces the harm of any single failure.

Compiling information about identifiable people is processing personal data, and proportionality, minimisation and a retention decision apply even where journalistic exemptions are engaged. This is a description of practice, not legal advice, and specific situations warrant a lawyer.

07Approaching subjects, on the record

People and organisations criticised should receive the substance of what will be published and a reasonable, stated deadline to respond. The approach should be specific enough to allow a meaningful answer, which also means specific enough to expose a misreading of a document before it appears in print.

Record what was sent, when, to whom, what deadline was given and what came back. A refusal to comment is information, a partial answer is information, and an unanswered approach is information — but only if it can be shown to have been made.

08Annotating the draft against the record

The most effective single review technique is also the least sophisticated: go through the draft sentence by sentence and mark each factual assertion with the identifier of the record in the source register that establishes it. Working copies of a draft carry those markers inline and they are stripped before publication, but while they are there they turn a vague sense that the piece is well sourced into a count.

What the exercise finds is consistent across investigations. A handful of sentences have no record behind them at all, usually connective material written to make a transition read smoothly and quietly asserting something nobody checked. Several sentences rest on a record that supports a narrower claim than the sentence makes: the document shows one authority, the sentence says the sector. And one or two sentences depend on a chain in which an intermediate step was assumed.

The count itself is worth keeping as a review measure. A draft that goes into legal or editorial review with every assertion mapped to a record is a different document from one that goes in with twelve unmapped sentences, and the difference is visible before anyone reads for meaning.

09Handover, absence and the state-of-play note

Investigations outlast the availability of the people running them. Someone goes on leave, moves desk or is pulled onto a breaking story, and an undocumented investigation stops at that moment rather than continuing. The remedy is a short state-of-play note, kept current rather than written when it is needed.

One page is enough if it contains the right things: what the current finding is in one sentence, what is established and by which records, what is asserted but not yet established, which approaches have been made and to whom, which requests are outstanding with their deadlines, what the next three actions are, and which decisions were taken that a successor should not silently reverse. The last item is the one most often omitted and the most expensive to omit, because a successor who re-opens a settled question wastes the work and may reach a different answer for no documented reason.

The note is also the test of whether the rest of the documentation works. If a colleague can read one page and then find every record it refers to, the structure is sound. If they cannot, the gap has been found before it mattered rather than during a dispute.

10Pre-publication review

Before publication, every factual claim in the draft should be traced to a record in the source register by someone other than the author. Numerical findings should be re-derived independently from the sources rather than checked against the author's working file, because a shared file reproduces a shared error.

Legal review, where the material warrants it, examines the strength of the evidence for each imputation, the fairness of the approach to subjects, the treatment of personal data and the language used. The most common outcome is not removal but precision: a verb changed, an inference marked as an inference, a claim attributed to the document that supports it.

11The published methodology

Readers should be able to see how the work was done: the sources and their access dates, the period, the analytical steps and formulas, the cleaning and matching rules, the requests made and refused, the limitations, and the difficult decisions with their reasoning. Where licences and data protection allow, publishing the processed data and the code is stronger again.

This is also where the boundary of the finding is set in public. A note that states what the records establish, and what they do not, prevents the story from being read as claiming more than the evidence supports — including by other outlets summarising it.

12A documentation timeline

The order reflects a single principle: each record is created at the moment it is cheap to create. A decision written down on the day costs a minute; reconstructed six weeks later it costs an afternoon and is less reliable.

• At the outsetCreate the structure, the source register, the request log and the decision register.
• On every acquisitionLog the item, its origin, date and access conditions; store the original untouched.
• On every judgementRecord the decision and the reasoning on the day it is made.
• Mid-investigationReview the register for claims with no record behind them.
• Before approachesPrepare specific questions; log what was sent and the deadline.
• Before publicationIndependent tracing of every claim; independent re-derivation of figures.
• At publicationPublish the method note and, where possible, the data.
• AfterwardsKeep a correction route; log and date every correction made.

13Corrections and retention

A correction route should be published and answered. When a correction is made, it should say what was wrong, what is correct, when it changed and how the finding was affected, and it should remain attached to the article. A documented investigation makes this precise, because the error can be located in a specific record or step.

Retention needs an explicit decision: what is kept to defend the published work, for how long, in what protected form, and what is deleted, including copies on personal devices and in message threads. Keeping everything indefinitely is not caution; it is a standing risk to the people the material describes.