OOS and OOT investigations. Follow the evidence before choosing the cause.

An out-of-specification result is not a failed number to be routed around. It is evidence that must be preserved, assessed promptly, and resolved through a scientifically justified investigation. Out-of-trend and atypical results require the same discipline even when they remain within specification.

Seal connects the original observation, raw data, sample, method, instrument, analyst, preparation, specification, batch, manufacturing history, hypotheses, retests, resamples, impact decisions, CAPA, and disposition. The result is a single investigation record whose conclusion can be reconstructed without rebuilding the case from laboratory, quality, and manufacturing systems.

The effective specification determines OOS. A governed statistical or scientific rule can identify OOT. An analyst or reviewer can identify an atypical observation that deserves assessment before a formal limit is crossed. Invalid describes a test conclusion supported by evidence—not an inconvenient result.

Seal preserves the initial classification, triggering rule, specification and method versions, detection time, reporter, immediate actions, and later reclassification with rationale. The workflow can differ by state without collapsing them into a generic deviation.

The system freezes the first reportable value, every underlying injection or observation, calculations, audit trail, instrument file, sample preparation, analyst entries, environmental conditions, and review state. Corrections and reprocessing create traceable versions.

No later passing result replaces the original failure. Reviewers see which result was generated first, what changed, who authorized further work, and how all valid results contributed to the final conclusion.

The initial assessment checks obvious and assignable laboratory conditions: sample identity and custody, preparation, calculations, standards and reagents, system suitability, instrument state, method execution, analyst observations, chromatograms or source files, and contemporaneous anomalies.

Tasks are specific and evidence-bearing. “Check instrument” becomes a review of qualification, calibration, maintenance, alarms, sequence events, acquisition method, processing method, and relevant system logs. A checklist cannot be closed without the evidence or an explicit not-applicable rationale.

The FDA's OOS guidance distinguishes laboratory assessment from a full-scale investigation. Seal supports that operating model while the manufacturer's procedures and scientific judgment remain controlling.

A hypothesis states the suspected mechanism, evidence that led to it, the experiment, expected discriminating outcome, sample or solution to be used, method, replicate plan, acceptance logic, and authorization.

Seal prevents testing into compliance by requiring an approved protocol before execution and by reconciling planned versus actual injections, preparations, and results. Exploratory injections remain visible and cannot silently become reportable or disappear from the sequence.

A retest examines another portion of the original homogeneous sample or preparation under an approved plan. A resample obtains new material from the batch and can introduce sampling uncertainty. The system records which occurred, why, who authorized it, the population represented, and how all results are evaluated.

Retest plans define number of analysts, preparations, replicates, instruments, methods, and stopping rules before results are known. Resampling requires documented evidence that the original sample may not represent the batch.

When no conclusive laboratory cause is demonstrated, the case expands into manufacturing and product history. Seal carries the laboratory evidence forward and opens a governed scope across batch execution, materials, equipment, environment, process data, deviations, changes, cleaning, personnel, and comparable lots.

The scope records potentially affected batches and products before root cause is known. Later narrowing requires evidence and approval; it does not rewrite the initial risk assessment.

Batch phase, recipe version, material lots, equipment state, calibration, alarms, holds, process values, yields, samples, interventions, and deviations attach through shared identifiers and time windows. Reviewers do not have to export and manually align histories.

Seal distinguishes direct evidence from inference. A historian interval can support a process observation; a correlation across lots can guide a hypothesis; neither becomes a confirmed cause without the approved reasoning and review.

The conclusion can be confirmed laboratory error, confirmed manufacturing cause, probable cause, no assignable cause, sampling cause, method issue, material issue, or another governed category. Each classification requires evidence and can carry confidence, limitations, dissenting review, and residual risk.

“Human error” is not accepted as the final mechanism without examining the conditions that made the action possible: instruction, interface, workload, training, equipment, environment, and control design.

Investigation status and batch disposition are related but separate. The impact assessment considers the original result, valid additional results, product and process knowledge, other attributes, stability, distributed lots, related batches, and uncertainty.

Seal records which lots are held, rejected, released, recalled, or require further action, who decided, what evidence was available, and any conditions or follow-up. A closed laboratory task cannot accidentally release material.

Trend limits, regression models, stability expectations, control-chart rules, prior-lot comparisons, and analyst review can generate OOT signals. Each rule has product, test, method, strength, timepoint or stage, population, calculation, minimum data, effective dates, and owner.

Method, specification, site, process, or instrument changes create visible population boundaries. A signal is not allowed to arise from mixing incompatible cohorts without an approved comparison basis.

Structured factors—product, method, analyte, instrument, column, analyst, material, supplier, equipment, phase, root-cause family, and failure mechanism—support cross-case analysis. Text search retains the nuance of narratives while structured relationships expose repeated patterns.

Related cases can be linked without prematurely declaring a common cause. A pattern review records its population, inclusion logic, evidence, conclusion, and actions.

An action starts from the confirmed or probable mechanism, affected control, risk, owner, due date, and effectiveness measure. Method revision, instrument remediation, training, supplier action, process change, specification change, or software correction follow their own governed workflows.

Effectiveness criteria are defined before implementation and use a bounded future population. “No repeat observed” is not sufficient unless exposure and observation opportunity are demonstrated.

Analyst notification, supervisor assessment, QA involvement, laboratory investigation, manufacturing response, impact review, and disposition each have accountable roles and time expectations. Paused clocks preserve reason; overdue states escalate without changing the original due history.

The person who performed a test can provide evidence without being forced to approve the conclusion. Conflicts of interest and independent review remain explicit.

Useful measures include time to containment, Phase I completeness, hypothesis yield, recurrence, aging by state, overdue impact decisions, invalidation rate by method, retest frequency, no-cause rate, CAPA effectiveness, and open product exposure.

Dashboards retain denominators and links to cases. A declining OOS count is not interpreted as improvement if testing volume, reporting behavior, or method sensitivity changed.

From an OOS result, an inspector can navigate to raw data, method, sample, specification, Phase I tasks, hypotheses, additional testing, manufacturing scope, root-cause rationale, impact, actions, and signatures. From a batch, the reviewer sees every open and closed laboratory signal affecting disposition.

Approved exports freeze the evidence and versions reviewed at decision time. Live records can continue with later effectiveness or trend information without rewriting the historical disposition.

The first implementation should include an original failing result, inconclusive Phase I, an approved hypothesis that does not explain the failure, a bounded retest plan, manufacturing expansion, an affected-lot population, a no-assignable-cause conclusion, batch disposition, CAPA, and trend review.

Also rehearse a clear calculation error, an OOT stability result, instrument data-transfer failure, sample mix-up concern, invalid system suitability, and a passing retest after OOS. The workflow is ready when every path preserves the original evidence and reaches the right accountable decision.