Guard banding means accepting a measurement only if it falls inside an acceptance limit set a little inside the tolerance limit. The gap between the two is the guard band w, so the acceptance limit is A = L − w for a tolerance of ±L. The guard band is usually sized from the expanded measurement uncertainty U, for example w = U. It lowers the chance of passing an instrument or part that is actually out of tolerance (a false accept), and raises the chance of failing one that is actually fine (a false reject) (Global ACI-TECH-1-002, formerly ILAC G8, §5.1 and §6.4).
ISO/IEC 17025 doesn’t require a guard band. It requires a documented decision rule whenever a lab states pass or fail: a rule that says how the measurement uncertainty is taken into account (clauses 3.7 and 7.8.6.1). Guard banding is the most common way to write one.
What guard banding is, and why “in tolerance” isn’t enough
Every result has an uncertainty, so the true value could be anywhere in a band around the reading. Far from the limit, that doesn’t matter. Near the limit, the band crosses it, and the pass/fail call becomes a bet.
Take a digital thermometer with a tolerance of ±0.50 °C. At one test point its error is +0.40 °C: in tolerance. With a calibration uncertainty U = 0.10 °C (k = 2), the band runs from 0.30 to 0.50 °C: it reaches the limit but does not cross it. With U = 0.25 °C, the same reading has about a 21 % chance of being out of tolerance (normal distribution, our calculation).
The vocabulary comes from Global ACI-TECH-1-002, the decision-rule guidance that accreditation bodies follow (§2.1–2.9, §5.1):
- Tolerance limit (TL): the specification limit, here ±0.50 °C.
- Acceptance limit (AL): the limit you actually compare the result with.
- Guard band (w): the distance between them, w = TL − AL.
- w > 0, guarded acceptance: the acceptance limit is inside the tolerance. Fewer false accepts, more false rejects. JCGM 106 also calls it stringent acceptance (§8.3.2).
- w = 0, simple acceptance: accept anything inside the tolerance. Also called shared risk, because a result on the limit is a 50/50 call for both sides (§5.1).
- w < 0, guarded rejection: the acceptance limit is outside the tolerance. You only fail items that are clearly out (JCGM 106, §8.3.3). Global ACI’s Table 1 lists w = −U as an “uncritical” rule (§6.2).
Guard band methods and formulas
There is no single guard band. These are the methods you will meet, all written for a symmetric tolerance ±L, an expanded uncertainty U = k·u at about 95 %, and TUR = L / U.
| Method | Acceptance limit A | What it controls | Source |
|---|---|---|---|
| Simple acceptance | A = L | Nothing by itself. Valid only with a limit on U (for example TUR ≥ 4:1) | Global ACI-TECH-1-002, §5.2.1; A2LA G136 |
| Multiple of U | A = L − r·U (r = 1 is the common choice) | Specific risk at the acceptance limit | Global ACI-TECH-1-002, §6.2 and Table 1 |
| Target specific risk | A = L − kw·u, with kw = Φ⁻¹(1 − PFAmax) | A chosen maximum specific false-accept probability | UKAS LAB 48, Appendix C |
| Root-sum-square (RSS) | A = √(L² − U²) | Global false-accept risk of about 2 % | Global ACI-TECH-1-002, Appendix B, Example 3 |
| Managed guard band | A = L − M·U, with M = 1.04 − e(0.38·ln TUR − 0.54) | Global false-accept risk of at most about 2 %, without knowing the EOPR | Dobbert 2008, eq. 4 |
| Fixed or indirect | Set by a regulator, a standard or the customer | Whatever the rule-maker decided | Global ACI-TECH-1-002, §6.1 |
Multiples of U: the Global ACI table
The most common rule is w = r·U. Global ACI’s Table 1 gives the maximum specific false-accept risk (PFA) for a result sitting exactly on the acceptance limit, for a single limit and normally distributed results (§6.2). The exact figures for k = 2 are ours, checked in our tests:
| Guard band w | Name in Table 1 | Maximum specific PFA at the acceptance limit |
|---|---|---|
| 3 U | 6 sigma | below 1 ppm |
| 1.5 U | 3 sigma | about 0.13 % |
| 1 U | ILAC G8:2009 rule | about 2.3 % |
| 0.83 U | ISO 14253-1:2017 | about 4.8 % |
| 0 | Simple acceptance | 50 % |
| −U | Uncritical | about 2.3 % false reject at the rejection limit |
Global ACI attributes 0.83 U to the 2017 edition of ISO 14253-1. JCGM 106 describes w = U as the default of the earlier 1998 edition (§8.3.2), so check which edition your contract cites.
A target false-accept probability
If you would rather choose the risk than the multiple, UKAS shows how: pick the largest specific PFA you accept, take the matching one-sided normal factor kw = Φ⁻¹(1 − PFA), and set the acceptance limit kw·u inside the tolerance (UKAS LAB 48, Appendix C). In UKAS’s Zener example, a maximum of 0.5 % gives kw = 2.58, and with u = 0.05 V an upper limit of −5.40 V becomes an acceptance limit of −5.53 V (Example 8). With a two-sided tolerance that is narrow compared with u, the other tail adds risk too, and the factor has to grow (Appendix C).
Global-risk methods: RSS and Dobbert’s managed guard band
The methods above control the risk for one result. The next two aim at the global risk: the share of all instruments calibrated this way that end up wrongly accepted, about 2 %, the limit made popular by the former ANSI/NCSL Z540.3.
- RSS: A = √(L² − U²). Global ACI gives it as an example of a binary rule for a global PFA of 2 % or less (Appendix B, Example 3).
- Managed guard band: Dobbert showed that, for a normal and centered population, the global PFA under simple acceptance peaks at an in-tolerance probability (EOPR) of about 60–65 % for each TUR in his Table 1 (1.5:1 to 4:1). Sizing the guard band for that worst case keeps the PFA at or below about 2 % for any EOPR, so you don’t need calibration history. His fitted multiplier M becomes negative above about 4.6:1 (our calculation from eq. 4; his Table 2 already gives a negative M at 5:1): the acceptance limit then sits slightly outside the tolerance.
Which method lowers which risk: a side-by-side comparison
A guard band trades one risk for the other: with a pass/fail rule, anything that cuts the consumer’s risk pushes the producer’s risk up (JCGM 106, §9.5.4; Global ACI-TECH-1-002, §6.4). Choose a TUR below to see how far each method pulls the acceptance limit in, and what that does to both risks.
Simple acceptance w = 0
±1.00 L
Managed guard band Dobbert 2008
±0.86 L
RSS A = √(L² − U²)
±0.87 L
w = 0.83 U ISO 14253-1:2017, per Global ACI
±0.58 L
w = U ILAC G8:2009 rule
±0.50 L
accept (±A) guard band (rejected although inside tolerance) tolerance limits ±L
PFA and PFR: global risk across a population that is 95 % in tolerance before calibration; k = 2, normal distributions, centered process.
Show all values as a table
| TUR | Method | A / L | PFA | PFR | Worst-case PFA (any EOPR) |
|---|---|---|---|---|---|
| 1.5:1 | Simple acceptance | 1.000 | 1.54 % | 6.62 % | 5.33 % |
| 1.5:1 | Managed guard band | 0.760 | 0.61 % | 16.9 % | 1.91 % |
| 1.5:1 | RSS | 0.745 | 0.57 % | 17.7 % | 1.78 % |
| 1.5:1 | w = 0.83 U | 0.447 | 0.10 % | 41.5 % | 0.29 % |
| 1.5:1 | w = U | 0.333 | 0.04 % | 53.5 % | 0.12 % |
| 2:1 | Simple acceptance | 1.000 | 1.34 % | 4.18 % | 4.18 % |
| 2:1 | Managed guard band | 0.859 | 0.65 % | 8.70 % | 1.92 % |
| 2:1 | RSS | 0.866 | 0.68 % | 8.43 % | 2.00 % |
| 2:1 | w = 0.83 U | 0.585 | 0.08 % | 25.4 % | 0.22 % |
| 2:1 | w = U | 0.500 | 0.04 % | 32.9 % | 0.09 % |
| 3:1 | Simple acceptance | 1.000 | 1.05 % | 2.29 % | 2.91 % |
| 3:1 | Managed guard band | 0.948 | 0.72 % | 3.44 % | 1.94 % |
| 3:1 | RSS | 0.943 | 0.69 % | 3.58 % | 1.85 % |
| 3:1 | w = 0.83 U | 0.723 | 0.06 % | 12.8 % | 0.15 % |
| 3:1 | w = U | 0.667 | 0.03 % | 16.4 % | 0.06 % |
| 4:1 | Simple acceptance | 1.000 | 0.86 % | 1.55 % | 2.24 % |
| 4:1 | Managed guard band | 0.987 | 0.76 % | 1.79 % | 1.96 % |
| 4:1 | RSS | 0.968 | 0.63 % | 2.16 % | 1.61 % |
| 4:1 | w = 0.83 U | 0.792 | 0.05 % | 8.19 % | 0.12 % |
| 4:1 | w = U | 0.750 | 0.02 % | 10.4 % | 0.05 % |
What the comparison shows (our calculation, same assumptions):
- w = U is the safest for the customer and the most expensive for the producer. At 2:1 it brings the PFA down to 0.04 %, but about 33 % of the population is falsely rejected, against 4 % with simple acceptance.
- RSS and the managed guard band are almost the same: at 2:1 they accept up to ±0.87 L (RSS) and ±0.86 L (managed), for a PFA near 0.7 % and a PFR near 8.5 %. A 2021 Sandia evaluation of guard banding methods reached the same conclusion (Delker 2021, abstract).
- “About 2 %” is not a guarantee. In our worst-case check, RSS reaches 2.07 % at TUR 2:1 when U = 1.96 u, slightly above 2 %. The managed guard band stays at 1.99 %. With k = 2, as in the chart, RSS peaks at 2.00 %.
- At 4:1 the differences shrink, and simple acceptance already peaks at about 2.2 % global PFA (k = 2; 2.28 % with Dobbert’s U = 1.96 u). That is why a TUR of 4:1 with simple acceptance remains a common rule.
Worked example: five acceptance limits for one thermometer
Tolerance ±0.50 °C, expanded uncertainty U = 0.25 °C (k = 2), so u = 0.125 °C and TUR = 0.50 / 0.25 = 2:1.
| Method | Calculation | Acceptance limit | Specific PFA for a result on the limit |
|---|---|---|---|
| Simple acceptance | A = 0.50 | ±0.50 °C | 50 % |
| Managed guard band | M = 1.04 − e(0.38·ln 2 − 0.54) = 0.282; A = 0.50 − 0.282 × 0.25 | ±0.43 °C | about 29 % |
| RSS | A = √(0.50² − 0.25²) = √0.1875 | ±0.43 °C | about 30 % |
| w = 0.83 U | A = 0.50 − 0.83 × 0.25 | ±0.29 °C | about 4.8 % |
| w = U | A = 0.50 − 0.25 | ±0.25 °C | about 2.3 % |
Now the reading at +0.40 °C. It passes under simple acceptance, RSS and the managed guard band, and fails under w = 0.83 U and w = U. Its own specific risk is about 21 %.
The two global-risk methods accept a reading with a 1-in-5 chance of being out of tolerance. That is not a bug: they are designed to keep the average false-accept rate low across many calibrations, not the risk of each result. Global ACI makes the same point: an instrument that passes a 2 % global criterion can have a specific risk close to 50 % (§6.3). Which kind of risk you want is part of the decision rule.
Decision rules: simple acceptance, guarded acceptance and rejection, binary and non-binary
A guard band is a number. A decision rule is the full statement: which acceptance limits, which outcomes are reported, which type of risk, and under which assumptions. Global ACI describes three formats (§5.2):
- Binary, simple acceptance: pass or fail, with AL = TL (§5.2.1).
- Binary with a guard band: pass or fail, with AL = TL − w (§5.2.2).
- Non-binary: four outcomes. Pass below TL − w, conditional pass between TL − w and TL, conditional fail between TL and TL + w, and fail above TL + w (§5.2.3).
Simple acceptance binary, w = 0
Guarded acceptance binary, w = U
Guarded rejection binary, w = −U
Non-binary four outcomes, zones of width U
Measured result increases to the right. Each zone is one expanded uncertainty U wide (k = 2). Risk figures are for a result sitting exactly on that boundary.
Here are four example results under simple acceptance and under a guard band equal to U:
measured result ± expanded uncertainty U guard band (AL to TL) beyond the tolerance limit
The same part, two rules
UKAS LAB 48 shows how much the rule matters with a surface-roughness specification of 1.5 to 1.9 and a standard uncertainty of 0.05 (Example 13). Rule 1 uses a guard band of 2u and accepts 1.6 to 1.8. Rule 2 uses simple acceptance with u limited to 0.05. For results at the upper end:
| Result | Rule 1 (guard band 2u) | Rule 2 (simple acceptance, u ≤ 0.05) | Chance the true value is above 1.9 |
|---|---|---|---|
| 1.70 | Pass | Pass | about 0.01 % |
| 1.75 | Pass | Pass | about 0.14 % |
| 1.80 | Pass | Pass | about 2.3 % |
| 1.85 | Fail | Pass | about 16 % |
| 1.90 | Fail | Pass | 50 % |
Both are valid rules. Rule 1 keeps the false-accept risk at or below about 2.3 %; rule 2 accepts up to 50 % at the limit and rejects less (UKAS LAB 48, Example 13).
“Simple acceptance, uncertainty not taken into account” is not a rule
By definition, a decision rule describes how uncertainty is taken into account (ISO/IEC 17025, clause 3.7). A2LA therefore treats “simple acceptance; measurement uncertainty not taken into account” as not meeting the requirement. Simple acceptance with a limit on the uncertainty is fine, for example “with a TUR of at least 4:1” or “with U no more than 10 % of the tolerance” (A2LA G136).
UKAS explains why with a tolerance of ±1 and a result of 0.5 that “passes”. With u = 0.1 the true value is almost certainly in tolerance. With u = 2 the chance falls to about 37 %, and with u = 10 to about 8 %. All three are a pass under unconstrained simple acceptance, which is why the risk is undefined without a limit on u (UKAS LAB 48, Appendix D). If a customer asks a lab to “ignore the uncertainty”, UKAS’s answer is for the lab to turn the customer’s expectation into a limit on U, or on the capability index, during contract review (Example 3 and Appendix D).
Specific risk or global risk
Specific risk is the chance that this one result is wrong. Global risk is the average over a population of instruments, and it needs to know how often they arrive in tolerance (the EOPR). Without calibration history, a lab can only give specific risk. A customer that manages its intervals to a target reliability can ask the lab for a global-risk rule (Global ACI-TECH-1-002, §6.3). See TUR isn’t risk for how the EOPR changes the numbers.
What ISO/IEC 17025 requires
ISO/IEC 17025:2017 is a paid standard. The clauses below are summarized from Global ACI-TECH-1-002 (§3), A2LA G136, PJLA PL-3 and NIST’s crosswalk, which quote or summarize them:
| Clause | What it means for guard banding |
|---|---|
| 3.7 | Defines the term: a decision rule explains how the uncertainty of the measurement is handled when a lab declares conformity with a requirement. |
| 7.1.3 | When the customer asks for a statement of conformity, the specification and the decision rule must be clearly defined. If the rule is not built into the specification or standard, the lab communicates it to the customer and agrees it. |
| 7.8.6.1 | The lab documents the decision rule it applies, taking into account the level of risk (false accept, false reject and statistical assumptions). A note says no further risk consideration is needed when the rule is prescribed by the customer, a regulation or a normative document. |
| 7.8.6.2 | The report states which results the statement applies to, which specification is met or not met, and which decision rule was used, unless the specification already builds it in. |
| 6.2.6 | Staff who issue statements of conformity must be authorized to do so. |
Two consequences people often miss:
- A statement of conformity is not compulsory. In calibration, the lab issues one when the customer asks for it. Without that request, it reports the result and its uncertainty (Global ACI-TECH-1-002, §7 step 1 and §8; UKAS LAB 48, “Is a conformity decision needed?”). That is why some accredited labs no longer print “pass”.
- The rule is agreed before measuring. The same result can pass with one guard band and fail with a larger one, so the rule must be in place before the work starts (Global ACI-TECH-1-002, §5.2.3 and §8). Labs may offer standard services with different guard bands, including zero, so customers can choose their level of risk (§8).
How to state the rule on a certificate and agree it with your customer
A good rule can be checked by someone who wasn’t there. Global ACI’s examples all name the same elements (Appendix B): the acceptance limit or guard band, the uncertainty basis, the outcomes reported, the probability distribution assumed, whether the risk is specific or global, and the resulting risk level.
Examples, in our own words:
- Simple acceptance with a capability limit: “Pass/fail by simple acceptance (acceptance limit = tolerance limit), applied only where the TUR is at least 4:1. Where the TUR is below 4:1, results are reported without a statement of conformity.” Global ACI asks you to say what happens when a measurement misses the ratio (§7, step 4 note).
- Guarded acceptance, specific risk: “Pass if the measured error is within the tolerance reduced by the expanded uncertainty U (k = 2); otherwise fail. Normal distribution, specific risk: a pass has at most about a 2.5 % chance of being out of tolerance.”
- Non-binary: “Pass, conditional pass, conditional fail or fail, with a guard band w = U on each side of the tolerance limit (Global ACI-TECH-1-002, §5.2.3).”
- Global risk: “Pass if the measured error is within the acceptance limit √(TL² − U²), for a global false-accept probability of about 2 % or less, assuming a normal, centered population.”
On the certificate, check three things against clause 7.8.6.2: the results the statement covers, the specification (your tolerance, or only the manufacturer’s?), and the rule. Our calibration certificate guide walks through the rest of the certificate.
Decision rule agreement checklist
The checklist follows the sample checklist in Global ACI-TECH-1-002, Appendix A, in our own words.
How to choose a decision rule
Global ACI’s selection chart goes through four questions in order (§7):
Is a statement of conformity needed?
If not, report the result and its measurement uncertainty. For calibration, that is the default unless the customer asks for pass/fail.
No: report result + U
Is there a legal or regulatory rule?
Use it as prescribed (for example, legal metrology rules that fix the uncertainty at a fraction of the maximum permissible error).
Does a published standard set the rule?
Standards such as ISO 14253, ISO 8655 or ISO 6508 often build the guard band into their limits, so a further guard band is usually not needed.
Otherwise, choose a standard rule or document your own
Simple acceptance with a TUR limit, a guard band such as w = U, or a global-risk rule. Agree it with the customer before measuring.
When you reach step 4, this is how the methods usually fit:
| Situation | Reasonable choice | Why |
|---|---|---|
| TUR of 4:1 or better, routine instruments | Simple acceptance with the TUR limit written into the rule | A2LA gives it as an acceptable rule; global PFA peaks at about 2.2 % at 4:1 |
| A wrong pass is costly (safety, flight hardware, legal) | w = U or larger, or a target specific PFA | Controls the risk of each result; accept the extra rejects |
| A wrong fail is costly (product recalls, investigations) | A global-risk rule (RSS or managed guard band) with your reliability data, or a non-binary rule | Fewer false rejects; Global ACI notes that each “fail” can trigger an impact investigation (§6.4) |
| Customer wants to see the borderline cases | Non-binary with w = U | Conditional results flag them without forcing a pass or a fail |
| TUR below 1:1 | Usually decline the work, or report without a statement | A2LA notes the risk of a wrong call can exceed 50 % |
A “fail” on the as-found data starts an impact assessment on everything the instrument measured. See out-of-tolerance calibration results.
Common mistakes and myths
| Myth | What the sources say |
|---|---|
| “ISO 17025 requires a guard band (or 4:1, or 2 %).” | It requires a documented decision rule that accounts for uncertainty and risk (clauses 3.7, 7.8.6.1). The ratio and the 2 % limit come from industry practice and the withdrawn Z540.3. |
| “Simple acceptance means ignoring uncertainty.” | Simple acceptance is only a valid rule with a limit on the uncertainty (A2LA G136; UKAS LAB 48, Appendix D). |
| “Shared risk is a reduced acceptance zone.” | Shared risk is simple acceptance, w = 0 (Global ACI-TECH-1-002, §5.1). |
| “ILAC G8 Type A and Type B rules.” | No such categories exist in ILAC G8 or Global ACI-TECH-1-002. |
| “A guard band of U always keeps the risk below 2.5 %.” | That is the specific risk for a result on the acceptance limit, for one limit, a normal distribution and k = 2 (§6.2, Table 1). With a narrow two-sided tolerance, the other tail adds risk (UKAS LAB 48, Appendix C). |
| “RSS guarantees 2 % global risk.” | It gives about 2 % or less; in our worst-case check it reaches 2.07 % at TUR 2:1 with U = 1.96 u. |
| “ILAC G8 has been withdrawn.” | It was reissued unchanged as Global ACI-TECH-1-002 on June 26, 2026. Only the section numbers moved. |
Back to the measurement uncertainty guide for where U comes from.
FAQ
What does guard banding mean?
Guard banding means accepting a measured result only if it sits inside an acceptance limit that is set some distance (the guard band) inside the tolerance limit. The distance is usually based on the measurement uncertainty. It lowers the chance of passing an item that is actually out of tolerance, at the cost of failing some items that are actually in tolerance.
Does ISO/IEC 17025 require guard banding?
No. When a lab issues a statement of conformity, ISO/IEC 17025 requires a documented decision rule that takes measurement uncertainty and the level of risk into account, agreed with the customer when it is not set by a specification or standard. A guard band is one common way to do that. Simple acceptance with a stated limit on the uncertainty, such as a TUR of at least 4:1, is another.
What is the most common guard band?
A guard band equal to the expanded uncertainty (w = U, with U at about 95 %). Global ACI's guidance lists it as the former ILAC G8:2009 rule. For a single limit, normally distributed results and k = 2, a result sitting on the acceptance limit then has about a 2.3 % chance of being out of tolerance. The cost is a high false-reject rate when the TUR is low.
My measurement is right on the tolerance limit. Pass or fail?
It depends on the decision rule you agreed before measuring. Under simple acceptance, a result exactly on the limit is in tolerance, but the chance that the true value is outside is about 50 %. Under a guarded acceptance rule it fails. Under a non-binary rule it is reported as a conditional pass. If you round results before comparing them with the limit, write the rounding into the rule and apply it every time.
Why doesn't my ISO 17025 calibration lab say pass or fail anymore?
ISO/IEC 17025 does not require a statement of conformity on a calibration certificate unless you ask for one. If you do, the lab has to agree a decision rule with you and report it. Many labs now report results and uncertainty only, unless the purchase order asks for a pass/fail statement and names the rule.
What does conditional pass mean on a calibration certificate?
It comes from a non-binary decision rule. The result is inside the tolerance, but closer to the limit than the guard band (often U), so part of its uncertainty interval reaches past the limit. The chance that the instrument is actually out of tolerance can be up to about 50 % for a result right at the limit. Treat it as a pass with a warning, and decide in advance what you will do with it.
What is the 4:1 rule, and is it a guard band?
The 4:1 rule asks for a test uncertainty ratio of at least 4, meaning the tolerance is at least four times the expanded uncertainty. It is not a guard band but a limit on the uncertainty. A2LA describes it as a generally accepted industry standard and gives simple acceptance with a TUR of 4:1 or better as an acceptable decision rule.
Is ILAC G8 still valid?
Its content is. Since January 2026, Global Accreditation Cooperation (Global ACI) has taken over ILAC's role, and on June 26, 2026 it reissued ILAC G8:09/2019 as Global ACI-TECH-1-002, with no changes to the content. Every section number went up by one (for example, the guard band table moved from section 5.2 to section 6.2).
Is there a guard band calculator?
Yes. Our free TUR calculator computes the TUR, the specific risk for a measured value, and the global false-accept and false-reject risk with simple acceptance, w = U, a custom guard band, or the smallest guard band that meets a target false-accept probability.
Sources
- Global Accreditation Cooperation (Global ACI). Global ACI-TECH-1-002 (G), Guidelines on Decision Rules and Statements of Conformity (formerly ILAC G8:09/2019). 2026 — §2 definitions, §3 ISO/IEC 17025 clauses, §5.1–5.2.3 guard bands and statements, §6.1–6.4 uncertainty and risk (Table 1), §7 selection, §8 documentation, Appendices A–B
- UKAS. LAB 48, Decision rules and statements of conformity (Edition 5). 2024 — Examples 3, 5, 8, 9, 13; Appendix C (guard band factor); Appendix D (simple acceptance)
- A2LA. G136, Guidance on Decision Rules in Calibration. 2022
- Keysight Technologies. A Guard-Band Strategy for Managing False-Accept Risk (M. Dobbert, NCSLI 2008; white paper 5991-1267EN). 2008 — Managed guard band, eq. 3–5, Tables 1–2
- JCGM / BIPM. JCGM 106:2012, Evaluation of measurement data – The role of measurement uncertainty in conformity assessment. 2012 — §8.3 guarded acceptance and rejection, §9.5.4 producer's and consumer's risk
- PJLA. PL-3, Policy on Measurement Uncertainty for Calibration and Testing Laboratories (Rev 1.13a) — Quotes ISO/IEC 17025 clause 6.2.6 b)
- NIST Office of Weights and Measures. ISO/IEC 17025 Crosswalk – Reporting the Results (clause 7.8). 2018
- Sandia National Laboratories. Evaluation of Guardbanding Methods for Calibration and Product Acceptance (C. J. Delker). 2021 — Abstract only
- NCSL International. ANSI/NCSL Z540.3 status page