The Inspection-Ready eTMF Playbook

It is six weeks to your inspection, and the file still isn't ready

You know the feeling. The inspection date is confirmed, and the real work begins: a frantic reconciliation of what should be in the Trial Master File against what actually is. Emails go out asking sites for the delegation logs that were never uploaded. Someone discovers three months of monitoring reports sitting in a shared drive that no procedure ever named as part of the TMF. A version of the protocol is filed twice, and nobody is certain which one is current. The team works late, not to improve the trial, but to make its record survive scrutiny.

This scramble is avoidable, and the evidence says it should be. Regulators have been remarkably consistent about what goes wrong. In the MHRA's GCP inspection metrics, record keeping and essential documents is one of the largest single categories of findings, and TMF completeness issues recur year after year — a signal that organizations are not acting on prior observations. Inspectors have also drawn a hard line at the extreme: where a TMF is so incomplete that it cannot form the basis of an inspection, that alone can rise to a critical finding, because it obstructs the inspector's ability to verify how the trial was conducted. The problem is rarely a lack of effort. It is that readiness is treated as an event at the end of the trial instead of a property maintained throughout it, so the same collection, chasing, and reconciliation happens under pressure every time.

What inspection-readiness actually means

Inspection-readiness rests on three qualities, and all three have to be true at once.

• Completeness. Every essential record that should exist for this trial does exist and is filed where an inspector can find it. Findings here almost always come from two sources: documents that were never filed, and content living in ancillary systems that no procedure ever declared to be part of the TMF.
• Timeliness. Records are filed contemporaneously — close to when the underlying event happened. An eTMF stamps every upload, so a reviewer can see at a glance whether a document was filed in the normal course of the trial or backfilled in a hurry. Large gaps between an event and its filing are a well-known red flag.
• Quality. The right version is filed, it is legible and complete, signatures are present, and metadata matches content. ICH E6(R3) reframes the appendix around records rather than documents and asks sponsors to apply risk-proportionate, quality-by-design thinking to the outputs that matter most — the TMF among them. A complete file of the wrong versions is not a ready file.

The sponsor is accountable for all three, even where day-to-day filing is delegated to a CRO or sites. That accountability is why a shared structure and measurable completeness matter so much.

The structure that makes completeness measurable: the TMF Reference Model

You cannot measure completeness without first agreeing on what complete looks like. The CDISC TMF Reference Model — a DIA initiative now maintained under CDISC — provides that agreement. It organizes trial documentation into a standard taxonomy so that sponsors, CROs, and sites classify records the same way.

The eleven zones cover the whole conduct of a trial — Trial Management, Central Trial Documents, Regulatory, IRB/IEC and other approvals, Site Management, Investigational Product, Safety Reporting, Centralized Testing, Third Parties, Data Management, and Statistics. Within each zone, sections group related artifacts, and each artifact is a defined document type with an identifier. Crucially, every artifact is filed at a level — trial, country, or site — because a single artifact type, such as an ethics approval, exists many times over across the countries and sites of a multiregional study. The Reference Model is a starting template, not a mandate: you tailor it to your protocol, removing artifacts that do not apply and adding those you need.

Driving completeness from an Expected Document List

The Reference Model tells you what could exist. An Expected Document List (EDL) tells you what should exist for this study. An EDL is a study-specific checklist derived from the model that names the artifacts expected — and, just as importantly, those specifically not expected — for the trial and for each country, site, product, and organization within it.

The EDL is where completeness stops being a matter of opinion. Because each expected artifact is tied to the trial's design and its milestones, the system can continuously compare what has been filed against what should be there and express completeness as a live percentage rather than a number someone assembles by hand the week before an audit. Milestones — first site initiated, last patient last visit, database lock — set the timing expectations, so an overdue artifact surfaces the moment its window passes. The EDL also has to evolve: when a country is added, a site closes, or a protocol amendment changes what is required, the list of expected records changes with it. Keeping the EDL current is itself part of the discipline, because completeness measured against a stale expectation gives false comfort.

Contemporaneous filing, version control, and Part 11 signatures

A complete list of the right documents still has to be trustworthy as a record. Three controls carry that trust.

Contemporaneous filing means the document enters the TMF close to when it was created or executed. This is not administrative tidiness; it is evidence that the record reflects the trial as it actually happened.

Version control ensures that superseded documents are retained but clearly distinguished from the current effective version, so no one has to guess which protocol amendment was in force.

21 CFR Part 11 governs electronic records and signatures. It requires secure, computer-generated, time-stamped audit trails that record who did what and when, and that never obscure previously recorded values; audit-trail data must be retained at least as long as the records themselves. It requires authority checks so only authorized users can sign, and signatures that are permanently bound to their records. The FDA's 2024 guidance on electronic systems, records, and signatures in clinical investigations updates how these expectations apply to modern platforms and service providers.

Where AI changes the work

Every step in that lifecycle has historically been manual, and the manual points are exactly where completeness, timeliness, and quality erode. This is where AI classification meaningfully shifts the workload — not by replacing the reviewer, but by moving effort upstream to the moment a document arrives.

• Classification. Instead of a coordinator choosing a zone, section, artifact, and level from a long list, a model reads the document and proposes the classification, extracting metadata such as dates, site, and version at the same time. The reviewer confirms or corrects rather than starts from scratch.
• Auto-filing. Once classified, the record is placed against the EDL automatically, so the completeness picture updates the instant a document lands rather than after a batch is processed.
• Quality checks. Before filing, the system can check the obvious failure modes — missing signature, wrong version, a filing window already breached, metadata that disagrees with the content — and route anything that fails to a human instead of letting it slip into the file unexamined.
• Gap detection. Against a live EDL, the system knows what is missing or overdue and can say so continuously, turning the pre-inspection reconciliation into a standing report.

The discipline that keeps this credible is human-in-the-loop review. AI proposes; a qualified person disposes. The regulations do not care whether a classification was suggested by a model or a person — they care that the record is accurate, controlled, and attributable.

A continuous-readiness operating model

Put together, these pieces describe a way of working, not a one-time cleanup. Readiness becomes a loop the team runs continuously, so the file is inspection-ready every week of the trial rather than for one tense fortnight before an inspector arrives.

The payoff is not only a calmer inspection. A TMF that is complete, timely, and quality-controlled throughout the trial gives the whole team a reliable operational picture — oversight of sites, evidence for sponsor decisions, and a defensible record if a finding is ever raised. The scramble disappears because there is nothing left to scramble for: the file has been ready all along.

If you want to see what continuous readiness looks like against your own study structure, we are happy to walk through it — including where AI helps and where it does not — with no obligation and honest answers on the limits.