Out-of-Specification Investigation AI Agent

What this agent does

Assembles the file, never sets a result aside

A result comes back outside specification, and the raw data, the sequence around it and the method file are pulled in.

A result is normalised across the laboratory, chromatography and quality systems, each fact carried with its source.

Reason

A record is built against the steps FDA sets out for the supervisor assessment in its May 2022 OOS guidance.

A retained sample is checked for quantity against the retest count your own written procedure fixed in advance.

A retest is proposed only where the plan naming it is written and dated before the confirmatory run.

Decide

A prior-occurrence search runs on method, analyst, instrument, column and reagent lot, not on product alone.

A distributed batch is flagged on day zero, because 21 CFR 314.81(b)(1) runs a three-working-day alert clock.

Out

A result is set against the trend rule your firm wrote; we located no primary FDA or EU definition of out-of-trend.

Writes run only inside the approval boundaries agreed at implementation, and never onto a signed investigation.

→ Product statement

The agent assembles and makes no GMP determination; a named analyst and supervisor assess, and QA signs the conclusion.

Example workflow

One result, failure to sign-off

AgentHuman

1Result raisedLIMS entry, chromatography data system, stability pull or contract-laboratory report

2Evidence assembledRaw data, the injection sequence, instrument logs, retained-sample quantity and prior occurrences

3File draftedAssessment checklist, evidence trail, hypothesis, trend picture and confidence

4Controls appliedCompleteness checks, retest pre-specification checks, clock checks and confidence threshold

No human action required

Stages 1 to 4 run unaided, and no result is set aside at any of them — the agent is assembling, and the assessors lane opens at the confidence gate.

5DecisionBranches at the confidence threshold

High confidence

Goes to the analyst and the supervisor.

Low confidence

Adds a second laboratory read first.

Assessment and sign-off

The file is held with its evidence, its open steps and the confidence.

Sign · Amend · Send for a second read

Signed — conclusion recorded▼

6Laboratory systems updatedOnly where write access and approval policy allow it

7Outcome evaluatedAssessment completeness, days to close, invalidations questioned and inspection findings

Amendments

A file reopened later is read by FDA, and what it attacks is the justification, not the number.

What should not run autonomously

Human approval stays in control

Outside the boundary — human approval required8 items

Invalidating a result, or setting one aside.

Assigning the cause of a result.

Concluding on batch impact or release.

Judging whether a laboratory error occurred.

Automation boundaryAgent acts unaided

✓Assemble the evidence against each step the FDA guidance enumerates.

✓Pull the instrument logs and the audit trail around it.

✓Search prior occurrences by method, instrument and lot.

✓Test the file for completeness and hold it for the assessors to read.

Writes stay inside the boundaries set at implementation, and never onto a signed investigation.

Averaging an original result with its retests.

Applying an outlier test to a suspect result.

Deciding whether a Field Alert report is due.

Changes to assessment, retest or trend rules.

Example output

One result, annotated

What the agent assembles is attached to the record it was drawn from.

OOS investigation output · single resultIllustrative example

Result

Assembled line

Rule cited

Time entered by

Confidence

Deadline

Assay below limit

Two reprocessings and one deleted injection sit in the sequence before this result

21 CFR 211.194(a)(4)

Analyst-entered

86%

Thirty business days, by your SOP

As receivedTaken from the chromatography, laboratory and instrument systems — nothing on this side is written by the agent.

Evidence used Audit trail of the run Instrument log Retained-sample amount

Why this ruleIt is the record rule the data has to satisfy and a person still decides.

ActionSignAmendSend for a second read

What the score decidesBelow the configured threshold the file picks up a second laboratory read before it goes up.

Value

Where AI adds value

The same four claims, placed at the point in the workflow where each one applies.

Where the value landsValue 01 – 04

Every resultFrom the laboratory record

03Assembly

Start where a number already exists

The sample-tracking page owns custody up to the number; this one starts when that number is out of specification.

01Approved path

Only evidence invalidates

No laboratory or calculation error identified, no scientific basis for setting the first result aside. FDA, May 2022.

02Human review

The clock nobody quotes right

Thirty business days from the occurrence comes from Barr, decided 1993 — not from FDA guidance, which sets no limit.

04Build an evidence trail

The result, the assessment it went through and the manager who signed stay on the file.

Integrations

Typical integrations

Five system groups connect to the same agent. Which of them are in scope is decided in discovery.

Laboratory systemsLabWare · STARLIMS
LabVantage · Sample Manager

Chromatography dataEmpower · OpenLab
Chromeleon · LabSolutions

Quality managementVeeva Vault QMS · TrackWise
MasterControl · ETQ

Agent

OOS & OOT investigation

Reads the result
Assembles the file
Holds for sign-off

Stability and trendingStability programme data
Trend and control charts

Observability & evaluationOpenTelemetry · Langfuse
Supported monitoring/evaluation sources

Integration availability depends on the client's existing systems and API access.

Agent controls

Six layers between the model and the file

Six gates in order, the widest first. What passes every one is set out in the map below.

L6 · Outermost — last line of defenceInward → L1 · closest to the model

L6Rollback / safe modePull the agent back to evidence tracking when evaluation or production signals degrade.Roll back

L5Version monitoringTrack model, prompt, checklist and trend-rule configuration changes.Track

L4Result traceabilityRecord the sources, the file, the amendments and the sign-off time.Record

L3Assessment sign-offHold files for the named assessors; the gate governs release, not whether the conclusion is sound.Gate

L2Policy guardrailsTest files against completeness, pre-specification and clock rules; a failure returns the file and names no cause.Restrict

L1Confidence thresholdsRoute low-confidence files to a second laboratory read before the assessors see them.Require review

Model coreFile assembled — assessment checklist, evidence trail, hypothesis, trend and confidence

L1 – L2Test whether a file may stand

L3Puts the conclusion in QA hands

L4 – L5Keep the result and the assessment behind it

L6Narrows to evidence tracking when signals degrade

How Nestack evaluates it

Evaluate the investigation workflow — not only the signed file.

Coverage runs the whole depth of the workflow, and every layer is cut by slice.

Surface — the file an inspector reads

Depth of coverage ▼

E1Final-output evaluationDid the file carry each enumerated step, its evidence and its date?

E2Step-level evaluationDid the agent use the right result, the right sequence and the current procedure?

E3Tool evaluationDid it read and write the correct result and the correct field?

E4Confidence calibrationDo low-confidence files actually attract more supervisor amendments?

E5Slice evaluationHow does performance change across specific method and test types?

E6Business outcomeHow many files were reopened, and how many invalidations were later questioned?

Floor — the outcome the laboratory answers for

Failure modes

Where each failure originates in the agent

Seven failure modes, set at the stage in the lifecycle each begins.

Agent lifecycleDirection of processing →

01 · Retrieval1 mode

ET-03

Sequence not pulled

The injections around the failing one never arrived.

Stage gathersRaw data, instrument logs, audit trail and prior results

02 · Reasoning2 modes

ET-04

Cause named too early

Probable laboratory error stands in for evidence.

ET-06

Trend limits widen

Limits recalculated from data that holds the excursions.

Stage proposesChecklist, evidence, hypothesis and confidence

03 · Tool / write2 modes

ET-02

Clock started at entry

Days counted from the LIMS record, not the occurrence.

ET-05

Retest count adjusted

A further run appears after the specified number.

Stage writesOnly where write access and approval policy allow it

04 · Output1 mode

ET-01

Prior failure missed

The same column chemistry failed three times before.

Stage returnsThe file the supervisor reads and QA signs

05 · Change / Version1 mode

ET-07

Silent checklist drift

A model or rule change drops an enumerated step.

Stage tracksModel, prompt, checklist and trend-rule config

Sev-1 · acts on a signed investigation Sev-2 · a result is set aside unsupported Sev-3 · evidence degrades, file holds

See the complete Biotechnology failure-mode catalogue →

Affected slices

The failures concentrate in a few methods

A laboratory-level assessment-completeness figure can read as settled while a few methods carry most of the amendments and the questioned invalidations. Nestack reports the amendment rate by method, not only in total.

Slice performance — reported separately, not only in aggregateIllustrative example

Slice	Failure rate	Lift	Status
Microbiological and sterility tests	9.7%	3.7×	Review
Stability-programme timepoints	7.1%	2.7×	Review
Dissolution and content uniformity	4.2%	1.6×	Watch
Routine HPLC assay tests	1.6%	0.6×	Normal

Bar: amendment-rate lift vs. the routine HPLC assay baseline · scale 0–4.0× · tick marks the 2.0× review threshold 2 of 4 slices over threshold

Evidence-linked improvement

A cycle ends where a new test starts

A cycle is done when the unsupported invalidation has become a case the next release must pass. That suite is what the next result worked is measured against.

Improvement cycle · five stagesSwitchback — the path turns at Improve and returns at Learn

01Detect

Amendment rate rises in one method.

02Diagnose

The result that failed once and was never allowed to fail again is read back until the cause narrows to one.

03Improve

Stamp the change; the results behind it are filed under that number.

04Verify

Each touched result case is run again, and one red stops the release.

05Learn

It stays a standing test, and the assessment rules move with it.

Learn → DetectThe return edge. The next detection runs against a suite one case longer.

Typical build scope

Twelve workstreams across six weeks

The build scope read against the delivery timeline. Week structure follows the six-week plan — discovery, sources, investigation workflow, evaluation, integration, then production validation and handover.

Workstream Week 1Week 2Week 3Week 4Week 5Week 6

01Investigation workflow discovery and boundary work.

02Laboratory and CDS source assessment.

03Assessment, retest and trend-rule procedure mapping.

04Result ingestion and evidence mapping.

05Assembly logic and source binding.

06Confidence scoring and exception routing.

07Analyst and supervisor sign-off workflow.

08Laboratory and CDS integration.

09Assessment and trend cases.

10Guardrails and invalidation controls.

11Result-trail instrumentation.

12Deployment, documentation and Agent Care handover.

12 workstreams · 6 weeks · bar shows the weeks a workstream is active — several run in parallel Final scope and sequence confirmed in discovery

Engagement tiers

What each tier includes

Rows are the capabilities named in each tier's scope. Higher tiers include everything below them.

Capability✓ in scope · — not at this tier PilotOne laboratory, one method ProductionProduction laboratory systems AdvancedMultiple sites / laboratories

Introduced at Pilot

Assembly to your result record✓✓✓

Assessment sign-off✓✓✓

Assessment-completeness baseline✓✓✓

Introduced at Production

Reporting by method—✓✓

Sign-off workflow in your systems—✓✓

Approved write-back—✓✓

Laboratory-system integration—✓✓

Introduced at Advanced

Complex trend-rule sets——✓

Multi-stage laboratory sign-off——✓

High result volume——✓

Multi-laboratory investigation controls——✓

Build price From $5,000 From $8,000 Custom quote

Final build priceConfirmed after discovery based on integrations, workflow complexity, transaction volume, approval controls and deployment requirements.

Separate from buildBuild pricing is separate from recurring Agent Care, which covers managed monitoring, evaluations, incidents and verified improvements after launch.

What we need from you

What you bring, and what we build with it

Each input maps to a piece of build scope and a week in the delivery timeline.

You bringWe build with it

01Your result record structure and procedures → Result ingestion and evidence mappingWeek 1

02Representative closed OOS files → Assembly baseline, checklist extraction and source bindingWeek 2

03Your assessment, retest and trend procedure → Procedure mapping and automation-boundary definitionWeek 1

04Access to relevant APIs, feeds or exports → Laboratory, CDS and quality assessment, then integration setupWeek 2

05Results you would not want re-tested → Hypothesis cases and the evaluation suiteWeek 4

06What no invalidation may rest on → Confidence scoring, exception routing, guardrails and sign-off controlsWeek 3

07Named analysts and supervisors to assess → Assessment sign-off workflow, then pilot and production validationWeeks 5–6

Nothing else is required Deployment, documentation and Agent Care handover are ours.

Delivery timeline

Four phases across six weeks

The widths follow the weeks the work actually occupies, which is the only reason week five overlaps.

Phase W1W2W3W4W5W6

Discovery W1

Build W2 – W3

Evaluate W4 – W5

Pilot & Launch W5 – W6

Week focus W1Investigation discovery, procedure mapping and the automation boundary W2Laboratory-system integration and the assembly baseline W3Assembly workflow, confidence logic and sign-off controls W4Evaluation suite, guardrails and failure-mode testing W5CDS and stability integration, pilot results and targeted corrections W6One investigation cycle run under the QA manager, then Agent Care handover

Reading the bandEach bar spans only the weeks its own work is named in. The week 5 overlap is two phases running, not padding.

At the end of W6Validation closes on live results, and Agent Care picks up monitoring.

DurationSix-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.

Next step · Biotechnology AI agent

Build an out-of-specification investigation agent around your laboratory.

Show us your OOS procedure, your chromatography system and who signs. Not an FDA rule. Thirty business days, from the occurrence of the problem, held in United States v. Barr Laboratories, 812 F. Supp. 458, decided in the District of New Jersey on 30 March 1993 — a district decision, not national precedent, and absent from FDA May 2022 OOS guidance, which sets no numeric limit at all. It binds a firm because the firm wrote it into its own procedure, and 21 CFR 211.22(d) requires that written procedures be followed. Invalidating a result, naming its cause and concluding on the batch stay with your analyst, your supervisor and your quality unit. Draft EU/PIC/S GMP Annex 22 — consultation closed 7 October 2025, no final text located as at 20 August 2026 — says generative models should not be used in critical GMP applications, and this agent is built so that it is not one. If your quality-event work starts at the deviation, this page is the laboratory result before it.

Discuss this agent →

Nestack Agents · Out-of-specification investigationAGT-BT-18 · Agent Care available after launch

Out-of-Specification & Out-of-Trend Investigation AI Agent