GxP audit trail review on every row, each finding cited to its log rows and decided by a person
Every audit-trail row screened on its risk schedule, each finding explained with its log rows cited.
See one case, screen by screen ↓Every GxP system writes an audit trail: who changed what, when, and why. The review procedure says to check it — for critical data before the result is approved, for everything else on a risk-based schedule. In one week a chromatography data system at one plant writes 148,220 rows. A LIMS shared across sites writes 186,410.
So reviewers sample. They export each system’s trail, work out which account belongs to which analyst and which shift, and scroll for the things that matter — an injection deleted, a peak reintegrated three times, a clock set back, a result approved by the person who entered it. What falls outside the sample goes unread, and every review period still needs a record that says what was screened, what was found and who signed.
Estimated hours per week across nine systems, sampling by hand and with the solution — the same estimate the solution’s dashboard shows (about 45 hours against about 12).
Five agents collect, screen and explain every audit trail, two more draft deviations and the review record; a data integrity reviewer decides every finding and signs.
Week 40 at two plants, a development lab and a clinical study. Here is how the data integrity lead works through it, screen by screen, in the working solution.
Priya opens the solution to nine systems under review: 100% of rows screened, 9 of 9 systems on schedule, 8 findings to decide — 4 of them high — and 490,574 rows screened this week. The activity radar plots 1,946 events of interest by time of day, with the 20:00–06:00 window shaded.
“Every GxP audit trail, screened on its risk schedule · 9 systems at 2 plants, 1 lab and study VLT-302”
The Waterford chromatography trail for week 40 is collected: 121,906 rows, exported at 06:10 with a verified checksum. She presses Run the review. The log collector confirms no gaps in sequence numbers, the normaliser resolves 38 accounts to people, the rule screener raises 290 events of interest and the pattern finder spots one late-evening cluster. Two findings come out: six impurity peaks integrated by hand with the same copied reason, and an aborted sequence rerun with its reason recorded.
“Record drafted · waiting for your decisions”
F-112, CDS Leiden: the automatic integration gave an assay of 94.1% against a 95.0–105.0% specification. Over 21 minutes the analyst moved the main-peak baseline three times; the third version reported 95.3% and was signed three minutes later. Two of the three manual integrations have no reason, a test injection ran before the official sequence and was excluded the next morning, and no OOS investigation is linked to the lot.
Every statement carries a number that opens its row in the export, shown exactly as exported.
The chromatogram for injection 4 shows the baseline moved to 5.30–7.21 min, under the signal. The version table lists every integration: automatic at 22:31, 94.1%; manual #1 at 22:41, 94.6%, no reason; manual #2 at 22:49, 94.9%, no reason; manual #3 at 23:02, 95.3%, reason “baseline drift”.
Repeated reprocessing, manual integration without a reason and trial injection before a sequence all fired on the same result. The pattern finder adds that 22:41–23:02 is outside the analyst’s rostered shift of 07:00–15:30, and that his 90-day rate of manual integrations is 0.4 per sequence. The suggested decision is to open a deviation — a possible unreported OOS.
The finding writer suggests one of three decisions — Justified, Ask the person or Open a deviation. It never decides.
Priya records her decision with a note: possible OOS not investigated. The deviation drafter fills DEV-2026-0612 from 9 cited rows — event date, system, description, lot CV50-2609-14 on hold plus three other lots the analyst tested after 20:00 in September, immediate actions and a proposed class of Major. She sends it to Omar Haddad, the site quality head, who approves it and it is raised in the QMS.
“A person approves before it reaches the QMS.”
Back on the CDS Leiden week 40 record, two findings remain: a processing method edited after the run (suggested: ask the person) and a sequence aborted for pump pressure with a reason recorded (suggested: justified — no impact). She accepts both suggestions in one step. The question on the method change goes to Anouk Smit, and her answer is attached to the record when it arrives.
“High findings always need their own decision.”
148,220 rows, 412 events of interest, 3 findings decided. Priya chooses the meaning “Reviewed — all findings decided”, enters her password and signs. The reviewer on the record is shown as independent of the records under review.
“Your signature, its meaning and the time are linked to this record, which is then locked.”
The signed record lists scope, export checksum, rule set v4, every finding with its decision and decider, and the deviation number. Each of the 14 active rules and pattern checks states its result — the rules that raised nothing say “no anomalies”. The record is locked; any change needs a new version with its own signature.
The dashboard shows 100% of audit-trail rows screened against about 20% sampled by hand, 2.39M rows and 31 findings in the last four weeks, and a median 1.6 days to a signed record against a target of 3. Findings are broken down by type, by system and by decision, and every reviewer’s records are shown with on-time rate.
The same review week, seen by four people who carry it — what their work looked like, and what it looks like now.
Five agents collect, screen and explain every audit trail, two more draft deviations and the review record; a data integrity reviewer decides every finding and signs.
Pulls audit-trail exports from the chromatography data systems, LIMS, MES, environmental monitoring, EDC, QMS and ELN on each system’s schedule, with the user and role lists and the shift roster, and checks each export is complete.
Maps each system’s columns to one event model — who, what, object, old and new value, reason, workstation, server and workstation time — and resolves account ids to people, roles and shifts.
Applies the review SOP’s rules — deletions, repeated reprocessing, manual integration without a reason, trial injections, method changes after a run, clock changes, shared accounts, self-approval and elevated rights — to every event in the period.
Compares each person, instrument and site with its own 90-day history — late-night clusters, bulk back-dating, unusual rerun rates — and explains why an event stands out.
Groups related events into one finding with a plain story in time order, the result before and after against its specification, a severity, a suggested decision and the exact log rows cited.
Turns a finding the reviewer sends to deviation into a QMS-ready draft: what happened, scope, affected lots, immediate actions, a proposed classification and the cited rows.
Assembles the review record for each system and period: scope, export checksum, rule set version, counts per rule, every finding with its decision and decider, and “no anomalies found” where that is true.
Decides and signs; QA approves deviations. The agents propose; a named person decides.
It answers from the audit trails of all nine systems, the findings and the review records, citing the log rows behind its answers — and it can add a rule or draft a deviation for approval.
What needs me before Friday?
Env. monitoring Leiden week 40 is due first, on Oct 8, with one finding — a clock change. CDS Leiden (3 findings), LIMS (2 findings) and CDS Waterford (not screened yet) are due Oct 9; EDC VLT-302 for September is due Oct 10. Start with F-112: a Corventa assay moved from 94.1% to 95.3% by three manual reintegrations late on Oct 1.
Any shared accounts?
One: qc_shift_b in the LIMS signed in from three workstations within 7 minutes on Sep 30 and entered 41 results. Three people were on shift B, so none of the 41 can be attributed. It was a label-printer account given result-entry rights in 2024. Suggested: open a deviation and disable the account.
Who changed a system clock?
On Oct 2 at 03:15 the environmental-monitoring workstation was set back 2 h 10 min, then the particle alert limit for Grade B room 2.14 was raised from 2,000 to 3,500. Separately, EDC site 104 entered 37 visit dates 3–6 weeks after the visits, against a usual lag of 3 days.
Add a rule: flag sequences rerun more than twice in 24 hours
Done. Over the last 90 days it finds 3 sequences: SEQ-24-1187 at Leiden, already F-112, and two at Waterford that were reruns after documented instrument faults. The rule is live for the next screening and sits in rule set v5 for Omar Haddad’s approval.
13 screens from the working solution, on its sample data. Pick one to see it large.
Review frequency, the out-of-hours window and who approves come from your risk assessment and SOP, and every change is versioned.
| Setting | Default | Choose from |
|---|---|---|
| Critical systems (CDS, LIMS, MES) | Weekly, plus before each release | Daily · Weekly · Every 2 weeks |
| High-risk systems (environmental monitoring, EDC) | Weekly | Daily · Weekly · Every 2 weeks · Monthly |
| Medium-risk systems (QMS, ELN) | Monthly | Weekly · Monthly · Quarterly |
| Out-of-hours window | 20:00 – 06:00 | 19:00 – 07:00 · 20:00 – 06:00 · 21:00 – 05:00 · 22:00 – 05:00 |
| Reviewer must be independent | On | On · Off |
| Second approval for high-severity findings | Omar Haddad | Omar Haddad · Priya Raman |
| Who approves deviation drafts | Omar Haddad | Omar Haddad · Lena Ortiz |
| Sign records that have no findings | On | On · Off |
Every figure is labelled: a target the solution is built to, an estimate, a typical published result, or a proven one.
“demo” = seen in the working solution, on its sample data · “target” = the design goal, measured in the live solution · “estimated” = the solution’s own estimate · Regulations named as context: 21 CFR 11.10(e), FDA data integrity guidance (2018), EU GMP Annex 11 draft (2025). People, companies and products named on this page are fictional — characters and sample data in the working solution.
Software that collects the audit trails of your GxP systems, screens them against your review procedure, and records who reviewed what, when and with which result. The Audit Trail Review Agent screens every row instead of a sample, explains each finding with the exact log rows, and produces a signed review record for each system and period.
Deletions, repeated reprocessing, manual integration without a reason, trial injections before a sequence, aborted runs, method changes after a run, clock changes, shared or generic accounts, results approved by their originator, elevated rights used on quality records, audit-trail setting changes and failed sign-ins — plus out-of-hours edits and unusual rates against each person’s own 90-day history.
No. Exports are read-only and kept unchanged with their checksum, and rows are shown exactly as exported. Original rows are kept beside the mapped ones.
Yes. The agents suggest a decision; a reviewer independent of the period decides every finding. High-severity findings need a second approval, and deviation drafts reach the QMS only after QA approves them.
The record still states what was screened and that each rule found no anomalies, and it is still signed by a person.
Yes. Review frequency by risk tier, the out-of-hours window, approvers and the rules themselves are settings. You can describe a new rule in plain words; it is tested on the last 90 days and goes into the next rule set version for approval.
Chromatography data systems, LIMS, MES, environmental monitoring, EDC, QMS, ELN and document management, with your identity feed and shift roster so every account resolves to a named person and shift.
The Agentic Solution Engine builds and deploys it from your requirements — your review SOP, risk assessment and sample audit-trail exports — and it goes live once every quality gate has passed.
Every system's validation pack composed from its requirements and supplier evidence, with testing sized to the risk of each function.
1-min filmAI Use-Case IntakeEvery new AI idea classified against the rules in minutes, each conclusion tied to its clause.
1-min filmVendor Cyber RiskEvery vendor’s security evidence read and mapped to your controls, with a cited risk rating ready to sign.
1-min filmWe’ll run Audit Trail Review on a sample of your own audit-trail exports.