Batch record review by exception — QA opens only the entries that break a rule, each linked to its page
QA opens only the entries that break a rule, each linked to its page
See one case, screen by screen ↓Batch record review by exception means QA reviews only the entries of an executed batch record that break a rule, instead of every line. In Batch Record Review by Exception, six agents read every page against the master batch record and the GDP rules, and the reviewer opens only the marked pages, decides each exception and signs the review.
How to choose AI solutions for life sciences →
Updated
Before a batch can go to disposition, QA reviews its executed batch record against the master batch record and the site's good documentation practice rules: 200 to 500 pages of printed steps, handwritten values, initials, dates, attached balance printouts and eBR exports. Done line by line, a 300-page paper record costs a reviewer about 30 hours, and the batch waits for release while it is read.
The findings that matter are the hardest to see. A hold time that only exists between two pages 23 pages apart, a purified water quantity that disagrees with the reconciliation, coating started before compression ended, a missing second-person check on a weighing. Errors found late send the record back to production, and missed ones can become observations under 21 CFR 211.192.
Estimated split of one reviewer's hours on a paper record, line by line and by exception. The totals match the 31 h line-by-line baseline and the 10.8 h median target shown on the working solution's dashboard.
Six agents read every page, check every entry and write each exception; the QA reviewer decides and signs.
Northwind Bio, Plant 2 Westbrook. CR26-1109 — Corventa 20 mg film-coated tablets, 600,000 tablets — arrives as a 304-page paper record on Wednesday, October 7. Here is what happens next, screen by screen, in the working solution.
Dana's home screen is her review bench: every executed record drawn as a strip of its pages, the coloured ticks marking the only pages she opens. CR26-1109 has just arrived from the Westbrook scan station against master record MBR-CRV-020 v7.2. Six records are already in review, with 15 exceptions open for her, worst first.
“Six agents read every page against the master record and the documentation rules. You open only the pages they mark.”
The record intake and page reader confirms 304 of 304 pages, 41 of them handwritten, 0 illegible boxes. The entry extractor turns them into 4,270 entries, 612 signatures and initials, and 1,388 dates and times. The master record comparator covers 188 steps and finds one critical process parameter outside its range and one allowed re-dry; the cross-page reconciler checks 14 hold times.
The page reader never guesses an illegible entry — it flags it — and never alters the scan.
The strip fills and nine ticks remain: 9 exceptions on 9 pages, 295 pages clean, 6 of them drawn for her spot-check. One critical, three major, one check and four GDP. The new critical exception joins the list to decide, worst first.
304 pages → 4,270 entries checked → 9 exceptions on 9 pages → 295 clean.
Dana opens the first exception and lands on page 101 with the entry highlighted. Step 7.12 records the final blend in bin blender BB-03 from 13:05 to 13:17 — 12 minutes — against 15 ± 1 minutes in the master record. The rule is quoted beside the page, with the agent's confidence of 0.98 and a suggested decision: raise a deviation.
Master record, step 7.12: “Blend 15 ± 1 minutes at 8 rpm after adding magnesium stearate. Blend time is a critical process parameter.”
She confirms the suggestion. The exception writer has drafted DEV-26-0631 — batch, title, evidence (p.101, master record step 7.12), classified critical, owned by Sam Patel in production with Dana for QA, and the batch on hold until the deviation is closed or assessed. She checks it and raises it into the quality system, linked to the batch.
Because the exception is critical, Omar Haddad second-signs DEV-26-0631, and the bench moves Dana to the next one: binder preparation on page 49, where 41.4 kg of purified water was weighed by J.M. with no second-person verification. The suggestion this time is a correction to production, added to one list for Sam Patel.
“Omar Haddad second-signed DEV-26-0631.”
On CR26-1108, the record before, granulation ended on October 1 at 21:40 (page 58) and drying started on October 2 at 23:55 (page 81). The reconciler pairs the two steps and shows both pages with the calculation: 26 h 15 min against a 24-hour limit. The drying page records “N/A” for a deviation, so none was raised before drying.
Master record, step 6.4: “Hold time is a critical process parameter. A longer hold needs a deviation before drying starts.”
The 295 clean pages are not taken on trust. Dana opens a random 2 % sample — 6 pages on CR26-1109 — each one with every entry already checked by the agents. If she finds something, it is raised as an exception and logged as a miss against validation report VR-RBE-0412, and Lena Ortiz, the validation lead, is notified.
“Page 1 of 6 in the 2 % sample · picked at random · every entry checked by the agents.”
VL26-0428, a 478-page eBR record for Veltrimab drug substance from Leiden, came through with nothing raised. Dana spot-checks 10 clean pages, then signs with the meaning “Production record review complete — 21 CFR 211.192”. The signed review goes to disposition with Dr. Maya Chen, Qualified Person, with any open deviations listed beside it.
“Your name, the time and the meaning are bound to this record and shown on every copy. Agents cannot sign.”
Across Westbrook, Leiden and Ashford the dashboard shows the median review at 10.8 hours a record against 31 hours line by line, 4.2 % of pages opened by a person, and 91 % of the agents' suggestions kept as decided. 298 clean pages were spot-checked in the last month, with 0 misses.
The same record, seen by the five people who carry it from the scan station to disposition — what their week looked like, and what it looks like now.
Six agents read every page, check every entry and write each exception; the QA reviewer decides and signs.
Receives scanned, hybrid and eBR records, checks the page count against the issued copy, classifies every page against the master record sections and reads print and handwriting.
Turns every box on every page into a typed entry — value, unit, initials, signature, date and time — with its page and position, and marks overwrites, strike-throughs, blanks and N/A.
Checks each entry against the master batch record and specifications — limits, critical process parameters, required steps and second-person checks — and follows the retest and rework paths the master record allows before raising an exception.
Applies the good documentation practice rules to every entry: signatures, second-person checks, corrections, blanks, late entries, date formats, and initials against the signature roster and training records.
Builds the batch timeline and material balance from all entries: hold times between paired steps, step order, every yield and reconciliation recomputed, and each quantity compared with every other mention of it.
Merges duplicate findings and writes one plain exception per problem, with page citations, the rule, severity, confidence and a suggested decision: raise a deviation, send a correction, or accept with a reason.
Decides and signs. The agents propose; a named person decides.
Ask about a record, an exception or a rule in plain words — or tell it what to change. Every answer points to its page.
How was the hold time worked out?
No page states the hold — the cross-page reconciler built it. It paired the granulation end, Oct 1 21:40 on p.58, with the drying start, Oct 2 23:55 on p.81, and compared the 26 h 15 min with step 6.4: no longer than 24 hours, a critical process parameter. It checks 14 hold times like this on every Corventa record.
How do I know the clean pages are clean?
Every record gets a random spot-check of 2 % of its clean pages. Last month reviewers opened 298 clean pages and found 0 misses, and you keep 91 % of the agents' suggestions. The one miss this quarter, an unmarked late entry, became a rule in v3.1.
Where do GDP errors come from?
Last 30 days: 128 GDP errors on 46 records. The biggest pattern is corrections without initials, date or reason at compression — 17, mostly shift B — like the overwrite on CR26-1108 p.133. You can send the pattern to training from the GDP log.
Flag any value within 5 % of its limit
Done — added as a check-only rule. On the open records it finds 1: tablet hardness 8.3 kP at 07:00 on Oct 3, CR26-1108 p.147, inside 8.0–12.0 kP; the press was adjusted and the next check was 9.6 kP. Saved as rule set v3.2. Check-only rules add exceptions and never hide one; Lena Ortiz signs the validation addendum.
13 screens from the working solution, on its sample data. Pick one to see it large.
What reaches a person, who signs and what the signature means are settings — each change a new rule-set version on the audit trail.
| Setting | Default | Choose from |
|---|---|---|
| Steps with critical process parameters | Always shown to the reviewer | On · Off |
| Values close to a limit | Off | Raise a “check” when a value is within 5 % of its limit · Off |
| Late entries | On | Raise an exception when an entry date is later than the activity and not marked “late entry” · Off |
| Spot-check of clean pages | 2 % of clean pages | 2 % · 5 % · 10 % of clean pages |
| Record types in scope | Paper, hybrid and electronic | Paper · Hybrid · Electronic |
| Second signature on critical exceptions | On — the QA operations lead signs after the reviewer | On · Off |
| Deviation owner | Production supervisor of the plant | Production supervisor of the plant · QA operations lead |
| Signature meaning | Production record review complete — 21 CFR 211.192 | Production record review complete — 21 CFR 211.192 · Batch record reviewed — EU GMP Chapter 4 |
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 batch records · “target” = the design goal, measured in the live solution · “estimated” = our estimate · “typical” = published figures (Pharmaceutical Online, “2024 Trends in FDA Observations for Sterile Drug Manufacturers” — production record review, 21 CFR 211.192, second most cited section, +171 %; PharmTech, “Review by Exception: Connecting the Dots” — paper-based release 10–40 days). People, companies, products and batches named on this page are fictional — characters and sample data in the working solution.
Batch record review by exception means QA reviews only the entries of an executed batch record that break a rule, instead of every line. In Batch Record Review by Exception, six agents read every page against the master batch record and the GDP rules, and the reviewer opens only the marked pages, decides each exception and signs the review.
Yes. The same agents read paper scans with print and handwriting, hybrid bundles and eBR exports. On the sample data a 304-page paper record and a 478-page electronic record go through the same review bench, and each record type can be switched in or out of scope in Settings.
The cross-page reconciler builds the batch timeline and material balance from every entry. It computes hold times between paired steps, checks step order, recomputes yields and reconciliations, and compares each quantity with every other mention of it — always showing both pages and the calculation.
A random share of clean pages — 2 % by default, or 5 % or 10 % — is spot-checked on every record before it can be signed. Anything found there is raised as an exception, logged as a miss against the validation report and the validation lead is notified. Spot-check misses and reviewer agreement are shown on the dashboard.
No. Steps with critical process parameters are always shown to the reviewer, even when every entry is inside its limit. The master record comparator follows a retest or rework path only where the master record allows one.
Yes. The agents suggest a deviation, a correction or acceptance; a named QA reviewer decides each exception and signs the review with its meaning. Critical exceptions need a second signature from the QA operations lead, and agents cannot sign.
Yes. Critical-parameter steps, the near-limit check, late entries, the spot-check share, record types in scope, second signatures, the deviation owner, who receives correction lists for each plant and the signature meaning are all settings. Every change is a new rule-set version on the audit trail.
The Agentic Solution Engine builds and deploys it from your requirements and documents — your master batch records, GDP SOP and a sample of executed records — and it goes live once every quality gate has passed. We will walk you through it on your own records first.
Any batch-record question answered in seconds, cited to the page and step.
1-min demoValidation Document AutomationProtocols drafted in minutes, every value traced to its source.
1-min demoManufacturing Batch AnalyticsBatch analysis that took three months now takes hours, with the query behind every answer.
1-min demoWe'll run Batch Record Review by Exception on a sample of your own executed batch records.