As-Found / As-Left Calibration: Recording Both States

As-Found / As-Left Calibration is the practice of recording an instrument’s measured performance before adjustment or repair and its final measured performance at the end of the calibration event.

These are industry-practice labels, not statutory terminology. As-found calibration and as-left calibration describe measurement states within an event, not separate instruments or mandatory certificate formats. The jurisdiction-specific calibration and recordkeeping duties are addressed in the regulatory mapping below.

The distinction is temporal: as found data characterise the initial condition; as-left data characterise the final condition. That final condition may still be unacceptable. Neither label establishes suitability for the intended use or authorises return to service; those decisions depend on the applicable acceptance procedure and authorised review.

Calibration establishes a measurement relationship; verification checks specified requirements. Adjustment changes response, while repair restores function. One service visit may include all four activities, but their records answer different questions. For a balance, an eccentricity and linearity check describes particular performance characteristics, whereas minimum weight concerns suitability at the low end of use. Neither should be inferred merely from an “as-left” heading.

As-found and as-left identify measurement states, not two prescribed tolerance widths. In industry calibration practice, the same approved acceptance limits may apply to both states. Keep three separate criteria visible in the procedure:

• Calibration acceptance limits define acceptable instrument performance at the specified test points and conditions, including the relevant error and precision criteria.
• An adjustment target is an optional aim for setting the instrument within its acceptance limits. It is not automatically a second pass/fail specification.
• Process or product limits define acceptable manufacturing conditions or product attributes. A manufacturing operating range describes where the process may run, not how much measurement error is acceptable.

Instrument error is a difference between an indication and a reference value, expressed in measurement units such as grams or degrees Celsius. A process parameter tolerance band instead bounds the process variable. Even when both use the same units, they answer different questions; an instrument’s allowable error cannot simply be equated with the process operating range.

Decide whether adjustment adds value

A tighter adjustment target can be a documented engineering choice, for example to provide room for expected drift. It is not a universal requirement. Leaving an acceptable instrument unchanged can avoid unnecessary intervention, preserve a stable setup and retain continuity in its performance history. Adjustment may still be justified by a persistent bias, a drift trend or a changed measurement task, subject to the site procedure and authorised judgement.

Evaluate the result, not just the display

Near an acceptance limit, comparing displayed error with a threshold may be insufficient. Measurement uncertainty, resolution, repeatability and test-point coverage affect what the result establishes. Apply the agreed decision rule, including how uncertainty affects acceptance; do not assume a universal uncertainty ratio or guard band. For weighing applications, balance capability relative to tolerance helps frame that suitability question.

A calibration pass supports suitability for the assessed measurement task. It does not, by itself, establish the acceptability of every process measurement made with that instrument.

Use six record groups, not six isolated database fields. This is a recommended record design, not a prescribed statutory form. Each group can contain structured values, attachments and linked records; together they should let a reviewer reconstruct the event without guessing which method, instrument state or decision a result belongs to.

A traceability statement alone does not establish that the reference was suitable for this measurement. Retain enough certificate detail to assess suitability for the actual range, method and uncertainty needs—not merely a certificate number detached from the event.

Preserve original observations, intervention details and superseded versions through corrections and review. An amended certificate should remain distinguishable from the original, with the reason, author and time of amendment retained. Apply an audit trail review workflow to examine that history rather than silently replacing earlier evidence with the latest document.

The following sequence is recommended operational practice, not a workflow mandated identically by every jurisdiction. Its purpose is to preserve the relationship between the instrument’s initial performance, any intervention and its final condition.

1. Prepare for the actual task. Identify the instrument, its configuration and the measurement it performs in production or testing. Select the authorised procedure and a suitable reference. Confirm that test points cover the intended use and that relevant environmental and loading conditions are addressed. For weighing applications, consider the relationship between balance capability and tolerance.
2. Separate conditioning from intervention. Follow prescribed warm-up or stabilisation steps, but distinguish them from actions that alter the instrument response. Before testing, document any necessary handling that could influence performance, including cleaning, resetting or configuration changes. Do not describe a changed condition as untouched as-found evidence.
3. Capture before changing. Record the initial observations before adjustment or repair. Preserve unsuccessful readings and their sequence, rather than replacing them with later passing results. If initial testing cannot be completed, identify the gap and reason instead of reconstructing measurements.
4. Choose and justify the next action. Leave the instrument unchanged, adjust it, repair it, replace it or continue investigation. Record why that branch was selected and what was done. After an intervention, establish and record the final measurement condition under the applicable procedure. Repeated adjustment until a passing reading appears is not a substitute for retaining the complete event history.

No adjustment can be a legitimate outcome. Where the approved procedure permits, one unchanged set of observations may support both as-found and as-left states. Explicitly identify that treatment and the observations used; do not imply that a separate test occurred or invent a second dataset.

Workflow: Prepare → Capture initial readings → Leave unchanged / Adjust / Repair / Replace / Investigate → Establish final condition → Authorised review and disposition.

At final review, reconcile the selected branch with the retained observations and unresolved issues. Exception-based review can focus attention on gaps, but authorised people decide the instrument’s disposition and return to service.

An unacceptable as-found result creates uncertainty about measurements made before the calibration event. The practical question is which earlier measurements could have been wrong enough to change a manufacturing, testing or product decision. Open a risk-based assessment through the applicable deviation process; do not treat the calibration failure itself as proof that every associated batch failed.

Set a provisional review boundary

Start with the last suitable satisfactory calibration as the initial backward-review boundary. Check that it covered the relevant range, configuration and measurement function. Then examine whether evidence supports narrowing the candidate period or extending it further back. An intervening check may support a narrower window only for the behaviour it actually tested. If the earlier calibration did not address the affected function, it may not provide a defensible boundary. This is assessment practice, not a fixed look-back period imposed by the cited provisions.

Follow use, not just location

Build the chain from instrument identity and location history to measurement tasks, associated equipment, production or laboratory records, batches and consequential decisions. Use batch execution history to connect readings to actual operations rather than merely listing batches produced during the period. Account for shared instruments, instrument swaps, portable equipment and loans outside the instrument’s home department. Record gaps in use history explicitly; a current asset location cannot establish where earlier measurements occurred.

Evaluate error direction, magnitude, relevant range, measurement uncertainty and credible failure mechanism. Compare these with intermediate checks, maintenance, environmental events and independent measurements. Ask whether the error could have changed each consequential decision, considering the applicable process or product criterion. A single end-point failure neither establishes when drift began nor proves linear drift. Do not interpolate historical corrections without evidence supporting that model.

Keep three decisions separate

Separate immediate equipment control, assessment of earlier product and final product disposition. Authorised people make and document each decision under the applicable procedures. An acceptable final instrument result addresses its final condition; it does not resolve earlier-product questions. Link product conclusions to the relevant batch record, including the reasoning for any unaffected-product conclusion.

Conclude with an assessment record containing the candidate use window, affected records, assumptions, supporting and conflicting evidence, unresolved uncertainty, decisions, owners and follow-up actions. Make the rationale for the final review boundary explicit so another reviewer can follow why particular uses were included or excluded.

All numerical readings below are hypothetical teaching assumptions, not product specifications, tolerance recommendations or reported manufacturing outcomes. These examples illustrate industry practice, not additional regulatory requirements. Throughout, error = instrument indication − reference value; a positive error means the instrument reads high.

Case 1: Balance adjustment reduces observed error

Assume a reference value of 100.000 g. Before adjustment, the balance indicates 100.030 g. After adjustment, a fresh observation at the same reference value gives 100.005 g.

• As-found error: 100.030 g − 100.000 g = +0.030 g = +30 mg.
• As-left error: 100.005 g − 100.000 g = +0.005 g = +5 mg.
• Reduction in observed error magnitude: 30 mg − 5 mg = 25 mg.

The adjustment reduced the observed error at this test point. It does not establish that either result is acceptable, or that performance improved across the full weighing range. Acceptance needs the approved criteria and uncertainty treatment. Retain both observations rather than replacing the initial reading with the improved result.

Case 2: Temperature bias changes the historical interpretation

Assume a reference temperature of 80.0 °C and an instrument indication of 82.0 °C. The observed error is 82.0 °C − 80.0 °C = +2.0 °C. Under an explicitly assumed constant +2.0 °C bias, an earlier indication of 82.0 °C corresponds to an estimated temperature of 82.0 °C − 2.0 °C = 80.0 °C.

For a heating operation, this direction of bias could mean the process was cooler than the operator believed, potentially reducing the intended heat exposure. The estimate remains conditional: today's observation alone does not prove that the same bias existed at the earlier time or throughout the operating range.

Apply the backward review to actual use: identify the sensor, operations and timestamps in the batch execution history, then examine intervening checks, maintenance and independent temperature records to support the review boundary. For each relevant use, compare the conditional estimate with the process limits already approved for that operation. Consider duration and sensor location where they affect interpretation of temperature monitoring. Do not substitute the calibration acceptance limit for a process limit or make an automatic batch decision; authorised reviewers assess the evidence and disposition.

Case 3: One shared instrument, three products, one out-of-tolerance point

Assume a shared temperature probe is used for three products whose approved operating ranges differ, and the as-found check at one test point falls outside the procedure's tolerance while the other points are within it. These are assumptions for the example only; no acceptance limits are implied. The reviewer first establishes which test point failed and whether that point lies inside, near or outside each product's operating range, using the equipment-use log to list the batches measured since the last acceptable event.

Impact can then differ by product: a deviation confined to a point no product uses may support a documented low-impact conclusion, while a product operating at that point needs a batch-level assessment under the out-of-tolerance or deviation procedure. Adjustment and as-left results restore the probe for future use but do not answer the question about earlier measurements. The authorised reviewer records each product conclusion separately rather than one blanket statement for the instrument.

Map each instrument to its product and activity scope before choosing the applicable calibration procedure. As-found/as-left is industry-practice vocabulary, not a prescribed record template in these provisions. The table separates regulatory duties and official guidance from practical implementation recommendations. For broader context, see US drug-product cGMP and dietary-supplement cGMP.

This mapping covers US drug products, US dietary supplements and the cited EU medicinal-product GMP guidance. For other foods, devices, chemicals or other jurisdictions, check the applicable current product-specific rules rather than extending these provisions by analogy. Keep the recommended workflows distinct from the duties they help organise.

Recommended architecture connects the equipment register or maintenance system, calibration records, laboratory or production records, historian data and quality investigations through stable instrument and calibration-event identifiers. Plan these associations across maintenance interfaces and historian interfaces. This is a recommended integration model, not a statement of any product’s capabilities.

At each interface, check serial-number mismatches, unit conversions, timestamp and time-zone differences, duplicate imports, missing attachments and conflicting record versions. Route exceptions to reconciliation queues, preserve source files and keep unresolved status visible rather than silently accepting incomplete events. Retain original values alongside converted values so reviewers can trace transformations.

For computer or related systems within US drug-product operations, 21 CFR 211.68(b) requires accuracy checks on input and output of formulas, records or data. Interface reconciliation supports that duty; the architecture above remains a practical recommendation.

For an interrupted event or incomplete service certificate, preserve what exists, identify the missing state or record, seek contemporaneous supporting evidence, document the limitation and route it for authorised assessment. Contemporaneous recording helps distinguish observations made during the event from later explanations. A new measurement describes the current condition; it does not recreate an unrecorded earlier condition.

Use these recommended practice checks alongside the earlier sequencing, acceptance-limit and impact-assessment explanations. They target record and workflow weaknesses; they are not a separate regulatory checklist.

Use these suggested programme metrics for management review, not as proof of compliance. State reporting-period dates and show numerator and denominator counts alongside percentages.

• First-pass as-found failure rate: events with unacceptable initial results ÷ events with evaluable initial results × 100, within the reporting period.
• Missing-baseline rate: events missing required as-found evidence ÷ events expected by the applicable procedure to contain as-found evidence × 100, for the same period. Report separately; missing baselines must not disappear behind the failure rate.
• Open impact assessments: at period end, report count, age since opening and responsible owner. Use locally justified escalation criteria, not universal closure targets.

Use this practical checklist before closing the event:

• Intended use and test range are identified.
• Method and acceptance criteria versions are recorded.
• As-found and as-left states link to original observations.
• Units and calculations have been checked.
• Interventions, or the no-adjustment decision, are recorded.
• Reference-standard evidence is associated with the event.
• CMMS and other system links are reconciled.
• Evidence gaps are resolved or explicitly escalated through the local deviation process.
• Equipment status and applicable impact-assessment and disposition records are documented.

Name the responsible technical reviewer and authorised decision-makers in the local procedure. Use software to organise evidence and route work; keep approval, equipment return-to-use and product disposition decisions with authorised people.