OOS and OOT investigation, explained.

What an OOS investigation is: finding out whether the test or the batch went wrong

One failed assay, from the lab bench to the batch decision

1. The test runs and a result falls outThe analyst sees the number and keeps every solution. In the example, 93.1% against a limit starting at 95.0%.
2. The lab checks the test before it tests anything elseThe supervisor reads the raw data, checks calculations, instrument, standards and method, and interviews the analyst. This is Phase I, the step this guide centers on.
3. Any further testing follows an approved planA hypothesis is written down and tested, usually on the kept solutions. Running them through the instrument again, called re-injection, can point at the instrument.2
4. If the lab is cleared, the search moves to productionPhase II brings in the quality unit, the independent group that approves or rejects batches, along with production and engineering, and looks at other batches too.2
5. The quality unit decides and the record stays wholeRelease or reject, with every result kept, passing and failing, and a corrective action where a cause was found.1,2

What step 2 finds decides whether the case stays in the lab or moves to production. The people, company and figures in this example are fictional.

Why a passing retest proves little, and three other things newcomers get wrong about OOS

Assignable cause, reportable result, retest, resample: the words OOS work runs on

Fourteen terms are defined in the summary page.

OOS, OOT and atypical results: three signals that carry three different duties

In shortAn OOS result is outside a written limit. An OOT result is inside the limit but off its expected path, usually on stability. An atypical result is inside the limit and looks odd, such as an unexpected peak on the instrument's trace. Only the first has a direct line to rejection and to a report to FDA on a fixed clock, and all three can be the early sign of the same problem.

Reading the citations. 21 CFR is Title 21 of the US Code of Federal Regulations. EU GMP is the EU's guide to good manufacturing practice, the rules for making medicines. CGMP is FDA's term for current good manufacturing practice. ICH is the International Council for Harmonisation, where regulators and industry agree shared guidelines. An ICH text at Step 2 is a draft out for public consultation, and Step 4 is the final text.

FDA's guidance covers chemistry-based testing of drugs under its drug center, CDER, including chemistry tests on biotech products, and says it is "not intended to address biological assays" beyond a short discussion of chapter 111 of the US Pharmacopeia (USP), the US book of drug standards.2 MHRA's flowchart is wider: it adds microbiological investigations and batches for clinical trials.5 It also excludes in-process tests used to steer a process to its end point.5

21 CFR 211.192, FDA's OOS guidance of 2006 and 2022, MHRA's flowchart: what binds and what advises

In shortIn the US the binding text is short: 21 CFR 211.192 requires the investigation and 211.165(f) the rejection. FDA's OOS guidance, issued in October 2006 and revised in May 2022, explains how FDA reads them, and it is nonbinding. The UK's MHRA publishes a slide flowchart dated October 2017. EU GMP Chapter 6 names both OOS and OOT in text in operation since October 1, 2014.

What FDA's 2022 revision of the OOS guidance changed from 2006

FDA calls May 2022 a revision and publishes no list of changes that we could find, so we compared the two PDFs. Most of the text is the same. These are the differences a lab should read.2,9

Other national texts follow the same lines. This guide states FDA, MHRA and EU text. Check your own market's inspectorate for local detail.

The Phase I laboratory investigation: FDA's seven supervisor steps, and MHRA's Phase Ia and Ib

In shortFDA asks for a prompt, objective first look at the test, "before test preparations" are discarded where possible, with "no preconceived assumptions" about the cause. It lists seven things the supervisor should do. MHRA splits the same ground into Phase Ia, for obvious errors, and Phase Ib, for the checklist with the analyst and supervisor. FDA's text has no Phase Ia or Ib, so check which one your procedure means.

How FDA's Phase I maps to MHRA's Phase Ia and Phase Ib

MHRA's checklist is longer than FDA's seven steps and worth borrowing whichever regulator you answer to. It adds sample labels and chain of custody, storage before use, glassware and volumes, whether the correct specification was applied, visual appearance, other activity at the time, temperature and humidity in the room, other batches in the same analysis set, and any other OOS results on the same batch.5

Read the evidence before you name a cause: why FDA cares about the order of a Phase I

FDA asks for "no preconceived assumptions as to the cause".2 In practice the order is easy to lose. A supervisor who hears "the pump was acting up" on day one tends to read the logbook looking for the pump. Writing down each check, with its verdict and its record, before any hypothesis exists is a simple defense against that. It also protects analysts: a cause reached by evidence rarely ends with an unfair note saying "analyst error".

Re-injection, retest and resample: what each OOS test can prove, and how testing into compliance happens

In shortRe-injecting a kept solution tests the instrument. Retesting takes a fresh portion of the same sample and tests the sample's handling or preparation. Resampling goes back to the batch and is rare. Each has its own rule. FDA wants the maximum number of retests fixed in writing beforehand, because a batch below its limit can still throw a passing number if you test often enough.

How many retests? FDA fixes the principle, and the number is yours to justify

FDA sets no number. It says the maximum "should be specified in advance", varies with the method's variability, "should be based on scientifically sound principles", and "should not be adjusted depending on the results obtained".2 MHRA says "papers have suggested 5, 7, or 9".5 Whatever you choose, the procedure must also say when testing stops and the batch gets judged.

The chart below shows why. Take a batch whose true assay is 94.6%, below a 95.0% limit, on a method with a standard deviation of 0.6%. One test has about a one-in-four chance of reading 95.0% or more by chance alone. Keep testing, and a pass becomes likely.

One low assay written up two ways: the file an inspector questions and the file that holds up

Both versions start from one low assay on Talvoren 10 mg tablets at Cobbleford Pharma, a fictional company. Every value is invented, and no real firm or case is described.

The first file would fail on FDA's own words: "a strategy of repeated testing until a passing result is obtained, then disregarding the OOS results without scientific justification".2 FDA's data integrity Q&A adds that all data, "including obvious errors and failing, passing, and suspect data", must be in the CGMP records.11

Can you average OOS results or use an outlier test? When FDA says yes and when it says no

In shortA method may define replicates as one reportable result, and then the average is the result. Averaging an original OOS with its retests "is not appropriate". An outlier test can tell you how far a value sits from the rest, and should not invalidate a chemical assay. It has no place at all in content uniformity (how evenly the drug is spread across single units), dissolution or release rate. A 1993 court opinion, known as Barr, still stands behind these lines.

The Barr ruling in one paragraph, and why a lab still meets it in 2026

In United States v. Barr Laboratories, 812 F. Supp. 458, a federal court in New Jersey heard FDA's case on how a generic manufacturer handled failing tests. The reporter dates the opinion February 5, 1993, amended March 30, 1993, and FDA's footnote cites it as February 4.

The court wrote that after a test failure traced to the process, "additional testing alone cannot infuse the product with quality", that averages "hide the variability among individual test results", and that the USP's silence on outlier tests for chemical assays "is prohibitory".16

FDA's 1993 inspection guide tells investigators what came of it: retesting "must be done on the same, not a different sample", a passing resample alone cannot invalidate an initial OOS result, and a batch released for content uniformity on averaged results had to be recalled. The same guide says the ruling "should not be considered as law, regulation or binding legal precedent".10 FDA's 2022 footnote still cites the opinion for the rule on outlier tests.2

The same objections still appear. Silgan Dispensing, October 5, 2026: "you invalidated the OOS results by averaging".22 ABS Corporation, July 23, 2026, told FDA it was "discontinuing the acceptance of averaged analytical results".23 ECA's 2026 survey summary reports greater variability on repeats and replicates, averaging and the reportable result.15

Invalidating an OOS result: the documented test event FDA asks for, and the three ways an investigation ends

In shortFDA allows invalidation "only upon the observation and documentation of a test event that can reasonably be determined to have caused the OOS result". Passing retests do not qualify. An investigation ends in one of three places: invalidated, confirmed, or inconclusive. In the third the OOS result keeps "full consideration" in the batch decision, and the quality unit decides.

FDA gives one worked case for releasing a batch despite an OOS result that was never invalidated. Look at how much it takes: a clean Phase I, a clean production review, a process with a good history, seven retests all within method variability, and consistent results from other tests. Only then might a quality unit "conclude that the initial OOS result did not reflect the true quality of the batch".2

Five questions a QA reviewer asks before approving an invalidation

Phase II of an OOS investigation: production at the table, other batches in scope and a three-day report

In shortWhen Phase I finds no clear lab error, FDA asks for a full-scale investigation, conducted by the quality unit, with production, process development, maintenance and engineering. It covers batches already shipped. No regulation we read sets a day count to finish the investigation. Two report clocks are law: 3 working days for a Field Alert Report on a distributed batch under an approved drug application, and up to 45 calendar days for a deviation report on a licensed biologic.

Source: FDA guidance, section IV.A.2 FDA ties corrective and preventive action to the principles in ICH Q10.25 The method behind root cause work and corrective and preventive action (CAPA) has its own guide on this site, on deviation and CAPA management.

How a Field Alert Report meets an OOS result on a distributed batch

For a product under an approved application, the applicant reports within 3 working days any failure of a distributed batch to meet its application specification.8 FDA's guidance says "unless the OOS result on the distributed batch is found to be invalid within 3 days, an initial FAR should be submitted", with a follow-up when the investigation ends.2

FDA's Q&A of July 2021 adds the stability case: results "scientifically invalidated (e.g., an analytical laboratory error is confirmed) within 3 working days do not require a FAR". For a batch that has not been distributed, no report is required.12

That clock has a practical effect on Phase I. A stability failure on a marketed lot gives the lab three working days to find clear evidence, so the records need to come together on day one. In the EU, Chapter 6 asks that a confirmed OOS or significant negative trend affecting batches on the market be reported to the competent authorities.4 MHRA says they "will require notification within a short time point of discovery due to recall potential".5

A licensed biologic has its own rule. 21 CFR 314.81 covers new drug and abbreviated applications, and 21 CFR 600.14 asks the license holder to report a deviation on a distributed biological product as soon as possible, "not to exceed 45-calendar days".8,20 The dates for finishing an investigation come from softer places. The Barr court wrote "within thirty business days of the problem's occurrence", and FDA's inspection guide of July 1993 wrote 20 business days.16,10

What we read for this. 21 CFR Part 211, 314.81 and 600.14; FDA's OOS guidance of 2006 and 2022; FDA's 1993 guide; the Field Alert Q&A; MHRA's flowchart; EU GMP Chapters 6 and 8, which say "in a timely manner" and give no number; ICH Q7; and the Barr opinion, on November 3, 2026. We did not read every national rule, so check your own market.1,8,2,9,10,12,5,4,7,16,20,26

Spotting an out-of-trend stability result: regression control chart, by-time-point and slope methods worked through

In shortA stability study tests stored samples at set times, such as 3, 6, 9 and 12 months. An OOT result is a point that leaves the path earlier lots took, and it can warn of a failure before shelf life ends. No regulator text we read prescribes a statistic. The three methods trade articles cite come from a 2003 PhRMA industry paper, and a lab must choose one, write it down and justify it.

The US rule, 21 CFR 211.166, requires a written stability testing program.27 FDA's guidance touches OOT only in a footnote: "much of the guidance may be useful for examining results that are out of trend".2 EU GMP goes further, asking that results be recorded in a way that permits trend evaluation, and that on-going stability data be enough "to allow for trend analysis".4

Trade authors note that FDA's guidance does not specify how to identify OOT results, and that regulatory guidelines do not fully address them.28,29 MHRA's flowchart calls "objective (often statistical) methods" advantageous and names none.5

What we read for this. We looked for a prescribed OOT statistic on November 3, 2026, and found none in 21 CFR Part 211, FDA's OOS guidance, MHRA's flowchart, EU GMP Chapter 6, ICH Q1E, the ICH Q1 draft, ICH Q7 or WHO's good practices for QC laboratories. We did not read every inspectorate's text. The method names below rest on the PhRMA team's 2005 follow-up, read in full, and two later reviews. We read only a listing of the 2003 paper, not its text.27,2,5,4,7,28,29,30,31,32,33

A worked regression control chart on one impurity

Take a fictional stability study of Dravenzo 20 mg tablets at 25 °C and 60% relative humidity. Earlier lots suggest impurity B starts at 0.05% and grows 0.0095 percentage points a month. At 12 months the line predicts 0.05 + 12 × 0.0095 = 0.164%. The band around it widens with time, by 0.05 + 12 × 0.0022 = 0.076 points. So the upper trend limit at 12 months is about 0.24%.

Lot CV25-0918 reads 0.31% at 12 months. The specification is 0.50%, so it passes. It is still well above 0.24%, so it is out of trend. A re-injection gives 0.30% and 0.31%, which clears the instrument. Its own slope, (0.31 − 0.05) ÷ 12 = 0.022 points a month, would reach 0.50% at about 21 months, on a product with a 36-month shelf life. The batch record then shows what differs: this is the only lot made with active ingredient from a second supplier route.

What an out-of-trend flag asks of you: MHRA's three alert levels

MHRA's flowchart sorts OOT stability results into analytical, process control and compliance alerts, and says that "as the alert level increases from analytical to process control to compliance alert, the depth of investigation should increase".

An analytical alert is a single aberrant result inside limits. A compliance alert suggests "the potential or likelihood for OOS results to occur before the expiration date".5 It also says that if a result predicts a failure before the next test point, schedule additional testing before that point.5 The same three alert levels appear in the PhRMA team's 2005 paper, an industry text.30

What FDA writes about OOS results: our reading of 514 warning letters since 2021

In shortA warning letter is FDA's published letter telling a firm what violations it found. About one in five recent drug and active ingredient warning letters carries a finding about how an OOS result was investigated: 110 of 514, on our count. On the same count, 54 fault a firm for setting a failing result aside. 21 CFR 211.192 was the second most cited section on FDA's file of Form 483s, its inspection observations, for fiscal 2025, by our sum.

How we counted, read November 3, 2026. The population is FDA's table of CGMP warning letters whose subject names finished pharmaceuticals or active ingredients: 516 rows, 514 letters once two duplicates go.

The 110. A word search for OOS wording, leaving out FDA's standard remediation requests, found 99 letters. We read a random 30 in full context and the matching sentence of the other 69, and dropped 6 that only mention OOS in passing. We then read the OOS sentences in the other 122 letters that use the term, which added 17.

The 54. We read the sentences on invalidating, retesting, averaging, discarding and leaving out results, and counted lab results only, so media fills and production leak or filter tests are left out. A person reading these can miss a letter or judge one differently, so treat both as our count and as about.34

What FDA objected to in OOS files, quoted from its letters

Each line below comes from FDA's copy of the letter and states FDA's position on its date. Firms respond, and some close out. The dates run from 2022 to 2026.

21 CFR 211.192 on Form 483s: about one citation for every three drug forms

A Form 483 lists what an FDA investigator observed at the end of an inspection. FDA publishes the citations by section each fiscal year. On our sum for fiscal 2025, 211.192 carries 236 citations, second to 211.22 at 309, across 713 drug Form 483s.44

Most of the 236 sit under "Investigations of discrepancies, failures" (164). Written records incomplete (32) and no written record (22) come next. The section covers all investigations, so OOS is only a part of it. FDA's fiscal 2025 quality report lists inadequate investigations among three top citations that resulted in facility withholds.45

MHRA inspections show the same subject in EU GMP Chapter 6 references. In MHRA's 2019 deficiency data, the newest GMP data on its page, C6.9 appears 31 times (12 major, 19 other) and C6.35 19 times (1 critical, 6 major, 12 other), out of 5,292 rows from 258 inspections, by our count.46

When a failed test reaches the market: drug recalls by reason, and why nobody publishes an invalidation rate

In shortIn FDA's enforcement data for 2021 to 2025, about 17% of drug recall records name a failed test or specification in their reason, by our grouping. Nobody publishes a trusted benchmark for how many OOS results a lab should see, how many it should invalidate, or what one investigation costs. FDA proposed reporting an invalidated OOS rate and has not made it a requirement.

The numbers nobody publishes, and what to measure in their place

FDA's quality metrics. In its drafts, FDA proposed a metric for invalidated OOS results. Its page, content current December 12, 2025, says FDA "is proposing to develop and implement a Quality Metrics Reporting Program", and describes a voluntary phase in the November 2016 draft and a March 2022 request for comment. So no required national number exists.48 A high rate may mean weak methods or weak investigations. A low rate may mean strong labs, or labs that do not write things down. Read the rate with its reasons.

How labs handle it now. ECA's Analytical Quality Control Group, an industry body, closed a survey in April 2026 with 169 complete responses to inform an update of its OOS guidance, whose SOP dates from August 9, 2013. Its first summary says many fundamentals are standardized: clear definitions, structured Phase I checklists and second-person review. Practice varies more on repeats and replicates, averaging, the reportable result, and how formal hypothesis testing is. The detailed percentages are for members, and we did not read them.49,15

AI in an OOS investigation: what draft EU GMP Annex 22, FDA's draft and Part 11 leave to the quality unit

In shortSoftware can gather records, read a run against its method, find similar past events, draft a report and re-fit a stability trend each night. The quality unit's decision does not move. Draft EU GMP Annex 22 says generative AI and large language models "should not be used in critical GMP applications", and it is still a draft. We found no published study of AI on OOS investigations, so ask any vendor for evidence on your own files.

What FDA has said to one firm about AI and the quality unit

FDA's letter to Purolea Cosmetics Lab, April 2, 2026, says: "If you use AI as an aid in document creation, you must review the AI generated documents". It places that duty under the quality unit's responsibilities, 21 CFR 211.22(c).51 It concerns one firm with other findings, so it sets no general policy. It does show how an investigator may read a machine-written report that nobody checked.

Four places software can mislead an OOS investigator

Who runs an OOS investigation today, how the records move, and where a supervisor's day goes

In shortAn analyst notices, a QC supervisor investigates, QA approves, and a stability lead owns trends. The evidence lives in five or more systems, and the supervisor usually pulls it by hand while the lot waits. No public figure says how long that takes, so the hours below are one estimate to replace with your own. Labs run this on paper and spreadsheets, on a quality system module, or with specialist investigation software.

The analyst, supervisor, QA reviewer and stability lead around one result

Paper, a quality system module or investigation software: three ways labs run OOS today

The lab bench at 8:02 am: every result against its limit, and an impurity OOS with its raw data read first

In shortLena Ortiz, head of QC laboratories, starts on one chart of 2,914 reportable results from the last 30 days. A new impurity result lands above its limit. She clicks it, and the panel reads the run before it offers any cause: a blank run of the diluent, the liquid that dilutes the sample, shows a peak in almost the same place, and the methanol in that diluent is a new lot on its first use.

1

Results from every test, drawn against one limit band

Each test's limits are stretched to the same band, so an assay, an impurity and a dissolution result can share a chart. Red is out of specification, amber is out of trend, and hollow circles are closed investigations. The footer counts 2,914 results and 6 open flags, and the 8:02 am result makes it seven. Each gray dot stands for about 11 results.

Beside the chart, the panel holds yesterday's assay, 93.1% against 95.0–105.0%. The note at its foot reads "Hypotheses wait for the raw data".

2

An unknown peak, a diluent blank and a new methanol lot

OOS-26-0152 arrives at 8:02 am: Dravenzo 20 mg tablets, lot CV26-0415, an unspecified impurity at 0.24% against a limit of 0.20%, at relative retention time (RRT) 1.32, its position on the instrument's trace. The instrument is HPLC-09, a liquid chromatograph, and the analyst is Marco Reyval. The panel draws the sample against the diluent blank.

Four checks follow. One passes. Two are findings: a 0.09% peak in the blank at RRT 1.31, and a diluent made from methanol lot MX-26-0712 on its first use. The peak itself stays open: not seen in the last 40 lots.

Only then does the panel name a likely cause, contamination from the new methanol, and it drafts a test using the previous lot. The screen's 71% is a ranking for the supervisor to weigh, with no stated probability model behind it. The blank peak is also smaller than the sample peak, which the card lists as evidence against.

A 93.1% assay at Cobbleford Pharma, from the first check to the signed invalidation and the CAPA

In shortLot CV26-0412 of Dravenzo 20 mg tablets read 93.1% on HPLC-07. Eleven checks with sources found three problems: a failing bracketing standard, pump ripple in the audit trail, and an overdue pump service. QA approved a re-injection plan. On a second instrument the kept solutions read 99.4% and 99.7%. The supervisor proposed invalidation, QA approved it for the quality unit, and two retests decide the lot.

1

Ten records named before a single verdict

The investigation opens on the evidence the checks will read: the CDS audit trail, the injection sequence, the instrument log, Sam Patel's training record, the interview note, the preparation worksheet, reagent and standard logs, the test method, 12 months of history and the investigation SOP. On the right, the audit trail is already open with key lines marked.

The diagram below it shows who does what, and it ends in a box marked "You decide".

2

Each step of the check, as it runs

Lena presses "Run lab check". The screen shows each step with what it read and found. The raw-data reviewer reads the run and its 64 audit trail entries. The lab-conditions checker reads the instrument, standards and training. The similar-event finder returns 4 pump events on HPLC-07 since June and no earlier assay OOS in 312 results.

3

Eleven checks in FDA's seven-step order, each with its source

Eight pass and three are findings. Passes include Sam Patel's training on the current method version, no manual integration or reprocessing, and a calculation re-performed to 93.04% and 93.18% from the areas, weights and standard purity. A number beside each sentence opens the record at that line.

Lena accepts each verdict or changes it, and each acceptance goes in the audit trail. The checklist's headings follow FDA's seven supervisor steps. The wording in each row is the sample's.2

4

A bracket that failed and a pump that rippled

The bracketing standard is a known solution injected around the samples to prove the system held steady. Here it drops to 97.1%, below the method's 98.0%, and the method says samples bracketed by a failing standard are not valid. The pressure trace shows the two spikes the audit trail records.

FDA's guidance speaks to this directly: if standard responses show the system is not working, "all of the data collected during the suspect time period should be properly identified and should not be used".2

5

Four hypotheses, with the evidence for and against each

The hypotheses open only after every check has a verdict. A transient pump malfunction leads, then a preparation or dilution error, a reference standard error, and a true product result. Weights and dilutions recalculate exactly, and the same standard weighing passed system suitability. Blend uniformity was 99.6% and tablet weights normal.

Each card names its test. A true product result would send the case to Phase II if the re-injection confirmed 93%.

6

QA approves the test before it runs, and the results stay investigational

Plan HTP-0147-1 states what, why and the criterion: both preparations agree within 2.0%, and all bracketing standards pass. Lena sends it, and Omar Haddad, QA lab oversight, approves before any testing.

On HPLC-04 the brackets read 99.8%, 100.1% and 99.7%. Preparation 1 gives 99.3% and 99.5%, preparation 2 gives 99.6% and 99.8%. These numbers support a cause, the way FDA describes re-injection.2 They are never reported or averaged with the original.

7

The supervisor proposes, and QA approves for the quality unit

The justification is drafted from the evidence, and Lena can edit it. She signs "Invalidate — assignable laboratory cause (instrument)" with the meaning "proposed as QC supervisor". Omar Haddad approves with the meaning "approved for QA". The file keeps the 93.1%.

The lot stays on hold. Aisha Brandt, a second analyst, retests the original sample twice on October 8, a number the SOP fixes in advance. FDA says that after a clearly identified lab error, the retest "would substitute for the original test result".2

8

A six-section report, with the run's other results and the CAPA

The report writer fills form QC-FRM-114-2, each statement cited. Section 5 follows the failing bracket: injections 15 to 17, lots CV26-0409 and CV26-0410, sat between the failing bracket and the next passing one, so they are marked for re-injection, while CV26-0408 was bracketed by passing standards and stands. That is the wider look that 21 CFR 211.192 asks for.1

The proposed CAPA services pump A before its next use, shortens the check-valve interval from 120 to 60 days on all four-solvent (quaternary) pumps, and adds a pump ripple flag.

When the retests pass and the OOS still stands, when Phase II opens, and when one vessel is the suspect

In shortAn instrument story is the tidy case. The same screens hold harder ones: a Talvoren impurity with no cause found, where the OOS result stood and the lot was rejected; a Veltrimab purity result reproduced on a second instrument, which moved to Phase II and a change in production; and a dissolution failure with three low units from one vessel, still waiting on QA.

1

Retests near the limit that do not rescue the lot

OOS-26-0141: impurity C at 0.21% against a limit of 0.20%, Talvoren 10 mg tablets, lot TX26-0219. No lab or process cause was found. Three retests by a second analyst, fixed in advance, read 0.19% to 0.20%. The summary says they "do not replace" the first result. QA closed the case as inconclusive and the lot was rejected on October 2.

This is FDA's third ending, where an OOS result keeps full consideration, and it is the quality unit's call.2

2

A result that reproduces, and a pump change upstream

OOS-26-0151: high-molecular-weight species (protein clumps) at 2.4% against a limit of 2.0%, Veltrimab drug substance (the active ingredient before it becomes a finished product), lot VD26-118. Two preparations both read 2.4%, and re-injection on a second instrument gave 2.4% and 2.3%. So the lab is cleared. The trend does the rest: lots before August 22 read 1.1% to 1.4%, then 1.8%, 1.9% and 2.4% after feed pump P-301 was replaced under change CC-26-1162.

Phase II runs with the process team. Three drug-product fills wait on it, and the options on screen extend to lots VD26-114 and VD26-116, as 211.192 asks.1

3

Three low units from one dissolution vessel

OOS-26-0149: Talvoren 10 mg tablets, lot TX26-0233. At Stage 3, the last round of testing, the average across 24 units is 79.8% against a Q of 80%, the share of the labeled dose that must dissolve. Three units sit below 65%, where no more than two are allowed. All three low units came from vessel 4, and its paddle height sits at 27 mm, the edge of the 25 ± 2 mm limit. The step that removes dissolved air from the test liquid, deaeration, was not recorded.

An outlier test has no place here: FDA says a value that looks extreme in dissolution "may in fact be an accurate result of a nonuniform product".2 So the plan with QA tests the apparatus directly, running kept tablets across vessels with freshly deaerated test liquid.

Nightly stability trending, investigation clocks, the lab dashboard and the testing rules your quality unit locks

In shortEvery night the trend watcher fits new stability results to earlier lots and flags those outside the band. Dr. Maya Chen, the stability lead, confirms or closes each flag. The rest of the product keeps the work in view: clocks for each case, repeated patterns sent to their owners, a dashboard of outcomes and causes, and settings where FDA's lines on averaging and outliers cannot be switched off.

The design targets for OOS and OOT Investigation, and the one estimate

These are goals, and no customer result is implied. Part C shows how to test them on your own files.

Seven OOS investigation agents, where each one stops, and what the quality unit keeps

In shortSeven agents open the record, read the run and the lab conditions, find similar events, write the checklist and the hypotheses, draft the report and watch stability trends. All of them read lab systems without changing them. People accept each verdict, the quality unit approves test plans and invalidations, and the stability lead decides each trend flag.

Who accepts, approves and signs an OOS conclusion

The lab records going in, the cited OOS file coming out

Works with the LIMS, CDS and logbooks a QC lab already runs

The solution reads what you run, by category: a LIMS for results and specifications, a chromatography data system for runs and audit trails, an instrument logbook, a learning system for training, a quality system for deviations, CAPA and change control, and an ERP for which materials went into which batch. Every verdict links to the record and the line behind it. It scales from one lab to a network of plants, with roles, maker-checker approval and an exportable audit trail.

Part 11 and Annex 11 controls on an OOS record, and the credentials we hold

The product gives you controls that support your compliance with 21 CFR Part 11 and EU Annex 11: audit trails, e-signatures with their meaning, and role-based access. Those rules apply to your own use of a system, and your validation answers for that use. No product can be certified for GxP.

Botminds is SOC 2 Type II audited and manages information security to ISO 27001. Draft EU GMP Annex 22 says generative AI and LLMs should not be used in critical GMP applications, and it is still a draft, so classify this use with your quality team and re-check when the text is final.21,50

Lena, Omar, Dr. Chen and Sam: the four people in the sample QC lab

What a late or shaky OOS file costs you, and how to count it from your last ten closed investigations

In shortYour supervisor's hours are the easy part of the bill. The bigger costs are lots on hold, solutions that expire before anyone tests them, and the file an inspector takes apart. We found no public benchmark for hours or cost per investigation, so count your own. A week with your last ten closed files gives you what the calculator needs.

Where OOS hours and hold days hide, and how to count them in your lab

The patient, the analyst and the next inspection: OOS costs no spreadsheet prices

Your return on OOS investigation software: three levers and a calculator with every assumption showing

In shortThree things can pay for this: supervisor hours given back, hold days taken out of investigated lots, and repeat investigations avoided because a pattern reached its owner. Hold days carry most of the value and rest on your softest number. Every input is an assumption until you replace it, and a small lab may get a negative answer.

What 60 OOS investigations a year could return after your yearly cost

Can your OOS process be automated yet? Score your SOP, your data access and how you invalidate

In shortSoftware can only apply a procedure you have written, to records it can reach, for a quality unit willing to look at its own numbers. Score each question from 0 to 2. The total places you in one of three bands, and each says what to do next.

What your OOS readiness score tells you to do next

What to ask an OOS software vendor about records, lab-error bias and signatures, then an eight-week replay

In shortAsk whether the tool can be wrong in ways you can see: does it cite the exact record, does it hold back a cause until the checks are done, can it mishandle a case with no lab cause, and does it change anything in your lab systems? Ask us too. Then test it on closed files whose answers you already know, including files where the lab got it wrong.

Where to start: the OOS steps that repay effort first

Eight weeks with your own closed OOS files

OOS and OOT questions people search: phases, retests, averaging, invalidation and software

In shortShort answers on OOS phases, retests, averaging, invalidation, Field Alert Reports and software.