Trust the measurement. Trace the impact.

1A calibration system files the certificate. Seal follows the measurement into the work.

A traditional calibration management system is a system of record for the instrument. It holds schedules, procedures, results and certificates, while the measurements that relied on the instrument sit in batch records, laboratory results and logbooks. It can show that an instrument failed its as-found check. On its own, it cannot show which batches, tests and qualification runs used the instrument, or stop the next use.

Seal records the calibration on the same platform as the work that relies on the instrument. Each point keeps its as-found and as-left observations, recorder, procedure and reference standard, and the event links to the recorded uses of the instrument. In Figure 1, TT-204’s −40°C point read −40.8°C against limits of ±0.5°C. After adjustment it read −40.1°C. The original stays on record, use remains restricted while review is open, and OOT-061 lists the two freeze-dryer cycles and the qualification run measured since the last acceptable check on 3 February.

Calibration belongs with the work that relies on it. A certificate describes an instrument at one point in time; the operating record also needs its intended use, range, procedure, current restrictions and the measurements it supported. With that history in one place, Neil can look across calibration results for drift and propose a changed interval or procedure, which your team verifies and releases under change control.

You can start with one critical instrument, beside the calibration system you already run. Agree which system owns instrument identity, schedules, results and current restrictions, and Seal uses that evidence where the work runs.

2Define the measurement hierarchy before the schedule.

Sites, systems, equipment, loops, instruments, channels and measurement points form a controlled hierarchy. Calibration attaches to the function that is measured, and each level keeps its own identity.

A transmitter can serve a control loop, an instrument can contain several ranges or channels, and a portable device can move between areas. Each keeps its own identity, tag, location, range, resolution and parent equipment, and replaced assets retain their lineage.

Criticality comes from intended use. The criticality assessment links the instrument to the process parameters, methods, alarms, interlocks, release tests and safety functions that depend on it, with the failure effect, detectability and independent checks. Criticality then drives calibration frequency, procedure, tolerances, review, overdue behaviour and impact assessment. A utility pressure gauge and a release-assay balance should not inherit the same controls because they share an asset class.

Range and tolerance are use-specific. The master record distinguishes manufacturer, calibrated and operating ranges, resolution, acceptance tolerance and the process or method requirement. Multi-range instruments retain separate points and criteria.

Configure the use workflow to check the relevant range and requirements. A device can be within its calibration acceptance yet inappropriate for the precision required by a particular method. Retain the reason for the eligibility decision with the work, including any missing prerequisite or restriction.

3Procedures are executable measurement designs.

An approved procedure defines prerequisites, environmental conditions, reference-standard requirements, calibration points, repetitions, calculations, uncertainty treatment, tolerances, adjustment rules and exception behaviour. For automatic, mechanical and electronic equipment in US drug manufacturing, routine calibration follows a written program.¹ An event instantiates the effective procedure against the exact asset, range, location, standards and due requirement.

3.1Keep the reference evidence with the event.

A reference standard needs to be suitable when it is used, not merely current when someone later opens its record. Calibrations use standards traceable to certified standards, where they exist.²

Reference and working standards retain their range, uncertainty, certificate, traceability source, drift, expiry, status and use history. Configure validity and suitability checks for the procedure, so range, uncertainty, validity and condition are checked at event time, and retain the certificate, corrections and standard version used.

A problem with the reference can reach beyond one instrument. If a standard later fails, its recorded uses provide the starting population for investigation: the events that used it, then the work performed with those instruments. PRT-009 leads to CAL-0882, and CAL-0882 to TT-204’s cycles and qualification run. Record what is known, what is missing and which uses need assessment; missing or unrecorded uses need to remain explicit.

3.2Schedule around the measurement and the work.

A planned date works only if the reference standard, the technician and the production window are ready on that day. Time-based, usage-based, campaign, before-use and condition-triggered strategies can coexist.

PRT-009’s certificate covers TT-204’s calibration on 7 September but expires before TT-205’s on 8 September, so the second job needs another standard or a recalibration first. Certificate validity is only one prerequisite; scope, uncertainty, condition, personnel and access also need review.

Due dates account for grace policy, production windows, access, cleanroom state and dependent maintenance, and forecasting groups work by location, standard set and technician qualification. Deferral is a governed decision with risk, duration, conditions, approval and revised due state, not a changed date.

4Keep the original result beside the final condition.

Preserve as-found data before adjustment, repair or configuration change. Attributable correction and version history retain what was recorded and why it changed, so a passing final result never overwrites the discovery state.³

The technician confirms asset identity, procedure, standards, conditions and configuration before collecting as-found data. Each point retains its reference value, observed value, error, tolerance, pass state and actual time, and opens as a record of its own.

4.1Adjustment creates a new dataset.

When permitted, adjustment records what was changed, why, by whom and under which authorisation. As-left testing demonstrates the final condition with a complete new data set.

Seal distinguishes calibration without adjustment, adjustment, repair, limited-use return, failure and inability to calibrate. A passing as-left result cannot hide a failing as-found condition.

In the TT-204 example, the indication at −40°C is −40.8°C at 09:10 UTC on 7 September. After adjustment, at 10:10 UTC, it is −40.1°C. The point record keeps both observations with their times and source references. The other four points read within ±0.3°C before adjustment and within ±0.1°C after it. The new results are within the example’s assigned limits, while the original failure still needs its own impact assessment.

4.2Record the uncertainty model and decision rule.

Where required, the event records uncertainty contributors, the calculation, coverage factor and decision rule, with versioned budgets and formulas.

Guard banding and statements of conformity are configured deliberately. Do not claim more precision than the model supports, or imply that all GMP calibrations require the same metrology model.

5Follow the failure into the work it may affect.

When a critical instrument fails its calibration, the work it measured since the last successful calibration needs assessment.² Where identity and time are captured with the use, Seal follows the instrument into that work.

An OOT calibration identifies the last known acceptable point, discovery time, potential failure start, maximum observed error, direction, affected range or channel, and immediate equipment state. The initial interval is conservative and can be narrowed only with evidence. For TT-204, the window runs from the last acceptable check on 3 February to the failure on 7 September; the earlier check is a review boundary, not proof of when the error began.

The investigation distinguishes standard failure, instrument drift, damage, environment, procedure, technician, data and use-related causes. Retesting or adjustment does not erase the discovery condition.

Use history identifies a population to assess. Follow the instrument into batches, process phases, samples, tests, environmental observations, alarms, qualification tests, cleaning records and decisions during the review window, and check completeness before treating that population as exhaustive. OOT-061 holds three recorded uses: the freeze-dryer cycles for batches B-071 and B-072, and temperature-mapping run OQ-118.

For each use, reviewers compare the affected range and observed error with process limits, independent measurements, control-system redundancy, product data, method tolerance and decision margin. Retain inclusion and exclusion rationale, unresolved evidence and the required decisions. Finding a linked batch is not proof that it failed; passing an as-left check is not proof that earlier batches were unaffected.

6Keep use permission separate from a passing check.

A passing as-left check shows the instrument’s final condition. Whether it may be used again, and for what, is a separate decision, and an instrument that does not meet its established specifications is not used.⁴

Manufacturing, laboratory, maintenance and qualification workflows can check eligibility before use, and the response gives the range or function, current status, restrictions and reason. In Figure 1, TT-204’s as-left check has passed, yet its use stays restricted while technical review is open. The equipment blueprint shows the same rule in release testing: after CAL-007 found HPLC-007’s pump flow at 0.92 mL/min against 0.98–1.02 mL/min, release testing on that instrument is blocked while authorised service work goes ahead.

A label can show tag, status, due date, restrictions and a verification code, and scanning it resolves the live record, which catches replaced, quarantined, overdue or range-restricted assets. A label does not override the system state. Continuity rules define what happens during an interface outage: cached status, independent checks, controlled override and later reconciliation.

6.1External calibration remains under site control.

A PDF certificate alone does not release the asset. Vendor capability, scope, accreditation where required, procedures and adjustment authority are assessed, and shipments preserve chain of custody.

Incoming review reconciles the serial number, ranges, points, as-found and as-left data, tolerances, standards and adjustments against the ordered scope and the actual instrument. Request what is missing, then reconcile the raw data, standards, decision rule, adjustments and required approvals before the site’s return-to-use decision.

6.2Change and replacement preserve measurement lineage.

A range change, move, firmware update, tolerance change or instrument swap identifies the affected schedules, open work, methods and validated uses. Replacement records predecessor, successor, equivalence assessment and baseline calibration, and historical use remains tied to the actual serial number.

7Neil prepares the work. Your team decides.

Neil is Seal’s AI agent. Bring it a failed standard, a calibration procedure or an instrument family’s drift history. It analyses the records the requester is permitted to see, follows their relationships and authors the assessment, workflow or controlled change needed for the next step. The useful output is linked work and inspectable configuration, not a summary to copy elsewhere.

To investigate, Neil assembles the standard’s history and a proposed review window with source references, groups the calibration events by instrument, range and procedure version, and lists the candidate batches, tests and qualification uses with missing links flagged. It ranks hypotheses with the evidence for and against them, and proposes follow-up checks. A candidate use is not a failed batch: your team assesses the evidence and records the disposition.

For a new procedure, Neil can author instrument, reference-standard and calibration-point templates, capture fields for separate original and post-adjustment datasets, versioned calculations and decision rules, and review states. It can prepare verification cases for an expired standard, a missing point, an out-of-range result, a changed formula or an unauthorised completion attempt. Where a script is needed, Neil can write and lint it for operator-run testing. Metrology specialists review the model and decision rule; authorised people execute verification, approve and publish the change.

To improve an interval or point set, Neil compares as-found trends, use intensity and failure history for the defined asset population, and proposes a revision with its rationale and uncertainty made explicit. The proposal lists the affected procedures, schedules, training and open work, with an effectiveness review defined by its recurrence measures, reporting window and source records. A favourable average does not extend an interval: review the technical justification and required verification before making the change effective.

Keep requirements, risk assessment, configuration versions and verification evidence with the change. Review the effect on open work, validated uses, training, labels and interfaces before cutover. Existing events retain their recorded procedure and source evidence.

In line with the EU’s draft GMP Annex 22 on AI, Neil is not used in GMP execution. It helps set up configuration, which your team verifies and releases under change control. See how Neil fits Annex 22 and more about Neil.

8Measure whether a change helped.

Interval optimisation uses stable history, risk, use, environment and failures under approved change, not automatic schedule extension. Completion counts alone can reward unnecessary work, so choose measures that show whether the measurement is under control.

Define the instrument population and reporting window before comparing a changed interval or procedure with the one it replaced, and keep the source records with the result so the team can challenge it.

9Start with one instrument. Prove the whole chain.

Take one critical instrument from master data and risk through procedure, standards, execution, an as-found failure, adjustment, as-left pass, quarantine, use-history extraction, batch and test impact, and interval review.

Include a multi-range instrument, overdue standard, external certificate mismatch, mobile asset location gap, interface outage and successor replacement. The implementation is ready when each measurement decision can show its traceability and suitability.

Keep an existing calibration system responsible for scheduling and certificates if it already serves that purpose. Agree which system owns instrument identity, schedules, results and current restrictions, which dates, results, approval states and certificate references the transfer includes, and how updates reach the execution workflow. Missing or stale evidence stays visible in the receiving check. Review anything Neil drafts from certificates and procedures against its source before relying on it.

Book a demo with one critical instrument, and we can scope the instrument model, use capture, eligibility checks and review responsibilities around that real operation.