One of the first questions we get from diagnostics teams moving off the bench and into real production volumes is deceptively simple: what has to happen before a lot can ship? At R&D scale, a scientist eyeballs the data and signs off. At 100,000 units a month, that judgment has to become a documented, repeatable release process that an FDA investigator can follow line by line. This piece is about building that process for IVD consumables, and about the decisions that determine whether it scales cleanly or becomes a bottleneck.
Lot release is the gate. Before finished product moves toward distribution, someone has to verify the lot meets its requirements and record that they authorized the release. That is not an LSO preference. It is the core of ISO 13485:2016 Clause 8.2.6, which requires organizations to monitor and measure product characteristics to confirm requirements are met, holds release until the planned arrangements are satisfactorily completed, and requires records that identify the person authorizing release (ISO 13485:2016 Clause 8.2.6). As of February 2, 2026, this is also U.S. law: the QMSR incorporates ISO 13485:2016 by reference into 21 CFR Part 820, so the ISO finished-device acceptance clauses are the operative device-QMS requirements for IVDs sold in the States (FDA Quality Management System Regulation (QMSR), 21 CFR Part 820 (2026)). And in case there's any doubt, FDA regulates in vitro diagnostics as medical devices, with IVD-specific overlays like labeling under 21 CFR Part 809 (FDA, Overview of IVD Regulation). Same QMS backbone, extra IVD requirements on top.
What a release panel actually covers
There is no universal test list, because the panel follows the product's requirements. For a lateral flow strip, a reagent blister, a molecular cartridge, or a microwell plate, the release panel generally spans three families of attributes.
The first is physical and dimensional: fill volume, component presence and placement, seal integrity on the sterile or moisture barrier, and closure. The second is functional or analytical: does the consumable produce the expected signal against known controls? This is where diagnostics diverges hard from a passive device. A blister can be dimensionally perfect and still fail if the reagent inside has drifted. The third is documentation and labeling: correct lot coding, expiry, and the QC instructions the end lab depends on. That last point matters more than it looks. ISO 15198 frames quality control as a shared responsibility between the IVD manufacturer and the laboratory user, including manufacturer-provided QC instructions and clear lot conventions (ISO 15198). Your release record and the QC guidance you ship are two ends of the same quality chain.
The honest translation problem for diagnostics teams is this: most of the readily available published detail on release testing lives in drug CGMP and sterile-packaging practice, not in IVD-specific guidance. You have to map that logic onto reagent fill, seal integrity, and assay signal yourself. That mapping is the work.
Where release criteria come from
Here is the part teams scaling up most often get backwards. Release criteria are not invented at release. They are set earlier, during process validation, and release testing simply confirms the process is still performing inside limits you already justified.
ISO 13485:2016 Clause 7.5.6 requires you to validate any production process whose output you can't fully verify by later inspection, and to establish criteria for reviewing and approving that process before use (ISO 13485:2016 Clause 7.5.6). For a heat-seal step or a reagent-dispense step, you cannot inspect quality into every finished unit after the fact, so you validate the process and then monitor it. FDA's process validation guidance describes the same pattern: you define lot acceptance criteria inside a process performance qualification protocol before you run the validation testing, then use statistical methods during continued process verification to catch intra- and inter-batch drift (FDA Guidance for Industry, Process Validation: General Principles and Practices). That guidance is drug-oriented, so treat it as a framework applied by analogy, not device gospel. The drug CGMP release rule makes the discipline explicit for anyone who wants the underlying logic: acceptance criteria and statistical accept/reject levels have to be adequate to ensure each batch meets specification before release, and the test methods themselves have to be validated (21 CFR 211.165). Again, that's the pharma citation. The device path routes the same rigor through Clause 8.2.6 and Clause 7.5.6.
Before your team starts blazing a new trail on acceptance limits, tie every criterion back to a validated process capability and a documented risk basis. A number with no justification behind it is the first thing an auditor pulls on.
100% inspection or statistical sampling
This is the decision that most affects your cost per unit and your throughput at scale, so it deserves real thought rather than a default.
One hundred percent inspection means every unit is checked against the attribute. It fits automated in-line checks that are cheap per unit, like vision inspection of fill or presence, and it fits high-risk or destructive-adjacent attributes where you cannot accept the statistical risk of passing a bad unit. The catch is obvious: for any test that consumes or destroys the consumable, or that runs the assay itself, you physically cannot test everything and still have product to ship.
That is where statistical sampling comes in. Instead of testing all of it, you test a defined sample and infer the lot's quality with stated confidence. The math is standard. In attribute (pass/fail) sampling, a plan of zero failures in 60 samples demonstrates 95% reliability at 95% confidence, a common way to pre-define pass/fail criteria before testing starts (MD+DI, Meeting FDA Process Validation Requirements). Note that worked example is a packaging and assembly case, so borrow the statistical relationship, not the exact target. The acceptance goal and sample size have to be justified per attribute against that attribute's risk.
Sampling conventions from sterile-barrier packaging illustrate how risk drives the number. When no mandated sampling plan exists, sample sizes are often chosen against AQL-style failure-rate objectives that differ by risk, on the order of 0.25% for invasive and 0.65% for non-invasive applications, and the same article is candid that statistics have real limits for detecting rare events (MD+DI, Sterile Packaging: Sample Sizes and Statistics). Those specific numbers are packaging-risk conventions, not reagent-performance criteria, but the principle carries: higher patient-impact attributes get tighter plans. A false result from a diagnostic consumable is a clinical event, so functional attributes usually warrant tighter sampling than cosmetic ones. Most real release panels end up mixed: 100% in-line checks on the attributes that are cheap and automatable, statistical sampling on the destructive and analytical ones.
Release testing as ongoing product monitoring, not a one-time gate
Release is not a stamp at the end of the line. It is one instance of the product monitoring that ISO 13485:2016 Clause 8.2.6 expects you to run continuously, with the record identifying who authorized release for each lot (ISO 13485:2016 Clause 8.2.6). The value shows up over time. Trend your release data across lots and you can see a fill step or a seal parameter drifting toward its limit before it produces a failing lot. That is the difference between catching a problem in-house and catching it in the field.
There is published precedent for how much a disciplined lot-testing program catches at diagnostic scale. The WHO/FIND centralized lot-testing programme for malaria rapid diagnostic tests evaluated 6,056 lots, roughly 1.6 billion RDTs, between 2007 and 2017 (PMC7359453). That program is post-market and independent, run for one RDT class in low-regulation settings, so don't read it as a template for your internal release design. Read it as evidence that lot-to-lot verification at volume is a mature, defensible practice, and that as manufacturer QMS maturity improved the field started debating how much centralized testing was still needed. The stronger your internal release process, the less anyone downstream has to re-test what you already proved.
That last point is where release testing and your lab accreditation intersect. Lab accreditation exists partly for this reason. ISO/IEC 17025 accreditation is intended to build confidence in test and calibration results, support wider acceptance of results across laboratories and regulatory bodies, and reduce the need for retesting, so data generated under an accredited scope carries more evidentiary weight when a release decision gets scrutinized. LSO's packaging test lab is accredited to ISO/IEC 17025:2017 (PJLA, certificate L26-623) for the seal-strength and integrity methods that sit inside many IVD consumable release panels, which means the seal and barrier data supporting a lot release comes from an accredited source rather than an unassessed bench.
Lot release for IVD consumables is not one test. It is a panel derived from product requirements, criteria justified during validation, an inspection-versus-sampling decision made per attribute against risk, and a record that says who released the lot and why it was safe to. Get that architecture right early and it scales from your first commercial lots to millions of units without a redesign.