OOS and OOT investigation that starts with the raw data already checked
Every out-of-limit result starts with its raw data already checked
See one case, screen by screen ↓Software that takes an out-of-specification or out-of-trend result from the LIMS to a signed conclusion: the Phase I laboratory investigation, any hypothesis testing, Phase II when no lab cause is found, and the report. OOS & OOT Investigation adds seven agents that gather and check the evidence and draft the plan and the report; the QC supervisor decides and QA approves.
How to choose AI solutions for life sciences →
Updated
When a QC result falls outside its specification, the evidence for the Phase I laboratory investigation sits in five places: the LIMS, the chromatography data system's run and audit trail, the instrument logbook, the standards and reagent logs, and the training records. The QC supervisor pulls it together by hand — most of a working day — while the lot sits on hold and the preparations count down to the end of their stability.
By then a cause has often been guessed. Hypotheses are written before the raw data is read, and an invalidation rests on a story rather than a documented test event — the kind of weak investigation FDA cites under 21 CFR 211.192. Out-of-trend stability results are found late too, because few labs fit each new result to the trend of earlier lots as it arrives.
Estimated: the QC lab's own estimate of supervisor hours per Phase I, by hand and with the agents, as shown on the solution's dashboard.
Seven agents check the raw data, rank the hypotheses, draft the report and watch the trends; the QC supervisor decides and QA approves.
Wednesday, October 7, at the Westbrook and Leiden QC laboratories. A new impurity result lands above its limit at 08:02, and yesterday's Corventa assay OOS is ready for its lab check. Here is what happens next, screen by screen, in the working solution.
Lena's lab bench places all 2,914 reportable results from the last 30 days inside one specification band: out of specification in red, out of trend in amber, closed investigations hollow. Below it run the investigation clocks — 30 days from result to closure, with each Phase I due date marked — and the repeated patterns the trend watcher has found.
“Hypotheses wait for the raw data. No cause is proposed until every Phase Ia check has a verdict and a source.”
OOS-26-0152 arrives from the LIMS: Corventa 20 mg tablets, lot CV26-0415, related substances — an unspecified impurity at RRT 1.32 at 0.24% against NMT 0.20%, on HPLC-09, analyst Marco Ruiz. Lena clicks the red dot. The Phase Ia lab check opens on the chromatogram, the sample against the diluent blank, and the agents check the raw data: the diluent blank shows a 0.09% peak at RRT 1.31, and the diluent was made from a new methanol lot on its first use.
“Now the hypotheses. 2 ranked from the 2 findings — most likely: diluent contamination from the new methanol lot (71 %).”
Seven investigations are open. OOS-26-0152 has its checklist ready and the plan to send to QA. The Talmirex dissolution OOS — a Stage 3 average of 79.8% against Q — has plan HTP-0149-1 with Omar Haddad for approval. The Veltrimab SEC result, HMW 2.4% against NMT 2.0%, is in Phase II with Jonas Weber in Leiden, and two out-of-trend stability results sit with Dr. Maya Chen.
OOS-26-0147: lot CV26-0412 assayed at 93.1% against 95.0–105.0% on HPLC-07, analyst Sam Patel. Lena runs the lab check. The raw-data reviewer reads run CV-ASY-1007 and its 64 audit trail entries, the lab-conditions checker reads HPLC-07, the standards and the training records, and the similar-event finder finds 4 pump events on HPLC-07 since June and no earlier assay OOS in 312 results.
Every step is shown as it runs, with what it found — read-only on the LIMS and CDS.
The supervisor's seven steps from the FDA OOS guidance, filled in: 11 checks, 8 pass, 3 findings. Sam Patel was trained on TM-CV-021 v6, the processing method was locked with no manual integration, and the calculation re-performs to 93.04% and 93.18%. The findings: the bracketing standard, the pump events in the audit trail, and a check-valve service due Sep 30 that was not done. Lena accepts the verdicts one by one or all at once, and each acceptance is recorded in the audit trail.
The bracketing standard after the Corventa preparations recovered 97.1%; the method needs 98.0–102.0% and says bracketed samples are not valid. The CDS audit trail shows pump A pressure ripple of 4.2% at 10:41 (injection 12) and 3.8% at 10:47 (injection 14), against an alert at 2.0%. Each finding opens the record at the cited line.
A transient pump malfunction on HPLC-07 leads at 86%, with the ripple alarms, the failing bracket and the overdue service as evidence for, and none against. A sample preparation or dilution error is 8% — weights and dilutions recalculate exactly and the two preparations agree at 93.0% and 93.2%. A reference standard error is 4%, and a true product result 2%.
The composer never proposes a cause before every check has a verdict, and never proposes averaging, unplanned retests or outlier tests for a chemical assay.
Plan HTP-0147-1 re-injects the retained preparations 1 and 2 on HPLC-04 with fresh bracketing standards, two injections each. Lena sends it to Omar, who approves it with his e-signature. The brackets pass on HPLC-04; preparation 1 gives 99.4% and preparation 2 gives 99.7%, agreeing within 0.3% against a criterion of 2.0%.
“Re-injection results are investigational: they are never reported and never averaged with the original 93.1 %.”
The justification is drafted from the evidence and Lena edits it freely. She signs “Invalidate — assignable laboratory cause (instrument)” with the meaning “proposed as QC supervisor”, and Omar Haddad approves for QA. The 93.1% is not used for the batch: Aisha Bello retests the original sample twice on Oct 8 under QC-SOP-114 §7.2, and those retests decide lot CV26-0412.
“This decision is yours. The agents propose; you sign with your password, and Omar Haddad approves.”
The report writer drafts the laboratory investigation report on form QC-FRM-114-2 in six sections, every statement with its citation. It names the other results the failing bracket touches — injections 15–17, lots CV26-0409 and CV26-0410, marked for re-injection — and proposes the CAPA: service the pump A check-valves before next use and shorten the service interval from 120 to 60 days on all quaternary pumps.
The same investigation, seen by the four people who carry it — what their week looked like, and what it looks like now.
Seven agents check the raw data, rank the hypotheses, draft the report and watch the trends; the QC supervisor decides and QA approves.
Opens an investigation from each OOS or OOT result in the LIMS — product, lot, test, method, specification, analyst, instrument, run — and starts the clocks.
Reads the chromatography run: system suitability, bracketing standards, integration events, processing-method version and the audit trail in the run window.
Checks the instrument (calibration, maintenance, events), the standards and reagents (lot, expiry, standardisation) and the analyst (training on the current method version).
Finds earlier OOS, OOT and lab deviations on the same product, method, instrument or analyst, and the result history.
Fills the seven-step Phase I checklist with verdicts and citations, then ranks hypotheses with evidence for and against and drafts the hypothesis-test plan.
Drafts the Phase I or Phase II report section by section with citations, the CAPA proposal and the impact on other results in the run.
Every night fits the stability results to the trend of earlier lots and flags out-of-trend results; finds repeated patterns and pushes them to the team that owns the cause.
Decides; QA approves. The agents propose; a named person decides.
Ask about any investigation, result or trend in plain words — or tell it what to change. Every answer points to its record.
What happened on OOS-26-0147?
Corventa 20 mg lot CV26-0412 assayed at 93.1% against 95.0–105.0% on HPLC-07. The evidence points at the instrument, not the tablets: pump A pressure ripple of 4.2% at 10:41 and 3.8% at 10:47, a bracketing standard right after the samples at 97.1% where the method needs 98.0–102.0%, and a check-valve service due Sep 30 that was not done. Calculations, standards and Sam Patel's training are all fine.
Can we average the retests?
No. The original result is never averaged with retests — it hides variability. When a method itself averages replicate injections into one reportable result, that average is one test, and that is allowed. The setting is locked in Settings.
Who can invalidate a result?
Only people. The QC supervisor proposes with an e-signature, and QA lab oversight approves — the agents can draft, never decide. Invalidation needs a documented test event that reasonably caused the result; for OOS-26-0147 that is the pump ripple and the failing bracket.
Which out-of-trend results need the stability team?
Three are open: Corventa impurity B at 12 months, 0.31% against a trend limit of 0.24%; Veltrimab acidic variants at 6 months, 24.8% against 23.9%; and Talmirex assay at 18 months, 96.2% against 96.8%. The Corventa one matters most: it is the only lot made with route B API, and at that slope it would reach 0.50% around 21 months.
13 screens from the working solution, on its sample data. Pick one to see it large.
Clocks, approvers, testing rules and the out-of-trend method are settings — every change versioned and recorded in the audit trail.
| Setting | Default | Choose from |
|---|---|---|
| Phase I due | 5 working days from the result | 3 · 5 · 10 working days |
| Investigation closes within | 30 calendar days | 20 · 30 · 45 calendar days; an extension needs QA approval |
| Hypothesis-test plans approved by | Omar Haddad | Omar Haddad · Dana Okafor |
| Retests after an invalidation | 2 retests, by a second analyst | 2 · 3 retests, number fixed before testing |
| Average retests with the original result | Never — hides variability | Locked |
| Outlier tests to invalidate a chemical assay | Not allowed | Locked |
| Bracketing standard drift alert | 2.0% from 100% | 2.0% · 1.5% · 1.0% |
| Out-of-trend method | Regression control chart, 95% prediction interval | Regression control chart · By time point · Slope control chart; 95% or 99% interval |
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 laboratory data · “target” = the design goal, measured in the live solution · “estimated” = the lab's own estimate, shown in the working solution · “typical” = published figures (FDA FY2020 drug inspection observations — 21 CFR 211.192 the most cited; FDA FY2023 inspection-based warning letters — 211.192 cited 30 times; Phase I and closure timelines from an industry OOS investigation guide; FDA, Investigating Out-of-Specification Test Results for Pharmaceutical Production, May 2022). People, products, lots and laboratories named on this page are fictional — characters and sample data in the working solution.
Software that takes an out-of-specification or out-of-trend result from the LIMS to a signed conclusion: the Phase I laboratory investigation, any hypothesis testing, Phase II when no lab cause is found, and the report. OOS & OOT Investigation adds seven agents that gather and check the evidence and draft the plan and the report; the QC supervisor decides and QA approves.
When a result lands outside its limits, the agents read the chromatography run, its audit trail, the instrument records, the standards and reagents, and the analyst's training, and fill the checklist with a verdict and a citation for each check. Hypotheses stay locked until every check has its verdict and source; only then are they ranked and a hypothesis-test plan drafted.
Every night the trend watcher fits each stability study to the trend of earlier lots and flags results outside the prediction interval — a regression control chart by default, with by-time-point and slope control chart methods and a 95% or 99% interval in Settings. It also finds repeated patterns, such as several pump events on one instrument, and pushes them to the team that owns the cause. The stability lead confirms or closes each flag.
No. Invalidation is a human decision: the QC supervisor proposes it with an e-signature and QA approves it. A result is invalidated only on a documented test event that reasonably caused it, and the agents never edit a LIMS result or reprocess data.
No. Retests are never averaged with the original result, and outlier tests are not used to invalidate a chemical assay; both rules are locked. Retests after an invalidation are done by a second analyst, with the number fixed before testing.
The LIMS, the chromatography data system, the QMS, the instrument logbook, the learning system and the ERP — all read-only for the agents. Each verdict links to the record and passage it came from.
Yes. The QC supervisor accepts every verdict, sends the plan and signs the conclusion; QA approves every hypothesis-test plan before testing and every invalidation; the stability lead confirms or closes out-of-trend flags. Every step and every change to a rule is recorded in the audit trail.
The Agentic Solution Engine builds and deploys it from your requirements and documents — your investigation SOP, test methods, specifications and a sample of past investigations — and it goes live once every quality gate has passed. We will walk you through it on your own results 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 OOS & OOT Investigation on a sample of your own past investigations.