Sampling versus a full census, and when each is defensible

Define lots, for example circuit by crew by week. Use random sampling with stated AQLs from ISO 2859-1 or ANSI/ASQ Z1.4, grade defects, and set escalation rules. Full census is defensible when risk is high or a sampled Critical defect appears. Record the frame, draw, table, and decisions so an auditor can reproduce them.

The lot you are sampling, and how to define it in vegetation work

Start with a frame that a reviewer can count. One practical choice is circuit by crew by week. If Crew 44 worked 480 spans on Circuit 12JY between July 15 and July 21, that is a lot of 480 units. The frame is the list of those 480 spans, with their identifiers and work dates. The same IDs must carry through the audit evidence and the acceptance record on each span, which is described in what a QA/QC audit actually covers.

If you group by circuit-month or contractor-district, write it down the same way: the unit, the count, and the list used. A lot definition that moves after the draw cannot be defended.

When 100 percent inspection is required, and why

Sampling is not always acceptable. Full census is the right answer when:

• Risk is high. For example, spans inside High Fire-Threat District Tier 3 during peak growth.

• Process control is weak. New contractor teams, uncalibrated raters, or evidence of drift are reasons to look at every unit.

• A sampled Critical defect appears. In many acceptance sampling plans, Critical AQL is set at 0, which means any Critical defect rejects the lot. A reasonable escalation is to inspect 100 percent of the lot, then correct and re-audit.

These are program rules, not imposed by a specific regulation. They are defensible because they are consistent with acceptance sampling practice for critical defects and with the duty to control risk documented in wildfire mitigation plans. California Energy Safety’s Independent Evaluator reviews QA and QC control programs as part of plan verification, so your trigger rules need to be written and applied the same way every time (2025 IE report, PacifiCorp).

Acceptance sampling for UVM: AQL, levels, and decision rules

If you sample, use a standard framework so the math is not a debate. ISO 2859-1, which is equivalent to ANSI/ASQ Z1.4, is a well established acceptance sampling standard with tables that map lot size and inspection level to sample size and acceptance numbers (ISO 2859-1, ANSI/ASQ Z1.4).

A common structure is:

• Inspection level. General Inspection Level II for normal inspections.

• AQLs by defect class. A typical practice table uses Critical AQL 0, Major AQL 1.5, Minor AQL 4.0, which yields acceptance and rejection numbers for each class at the sample size selected.

Write those choices into your QA plan. On vegetation work, Critical covers immediate hazards or clear code nonconformances. Major covers work-spec deviations that affect function or near-term compliance. Minor covers workmanship and documentation issues. The grading itself belongs in what a QA/QC audit actually covers.

Example calculations a reviewer can re-run

A reviewer has to be able to follow the arithmetic from lot size to sample size to accept or reject.

On the 480 span lot for Circuit 12JY, Crew 44, document:

• The inspection level you selected, for example General II.

• The AQLs for Critical, Major, and Minor.

• The table version used. Name ISO 2859-1 or ANSI/ASQ Z1.4 and the source of your table.

• The code letter from the table for a lot of 480 at your inspection level.

• The resulting sample size, drawn from that code letter.

• The acceptance and rejection numbers for each defect class at that sample size.

If the audit found two Majors and three Minors in the sample, the decision comes from that last line. Some programs also define a utility-specific acceptance rule for Majors that is tighter than the table’s count for operational reasons. If you do, write it as a rule and apply it consistently across lots and time.

Switching rules, escalation, and when to leave normal inspection

Acceptance sampling frameworks define switching rules between normal, tightened, and reduced inspection based on recent results. Use that logic explicitly.

• Start new contractors or newly mobilized crews on tightened inspection, then move to normal after a run of accepted lots.

• If a lot fails under normal inspection, switch to tightened inspection for the next k lots, where k is the number in your plan.

• If you reject multiple lots in succession, escalate to 100 percent inspection, retraining, or both.

Do not claim a switch if you did not apply it, and record the reason for the switch in the lot report. T&D World has argued that vegetation management needs stronger, meaningful metrics to drive programs, and switching rules tied to acceptance results are one way to make program control visible in numbers (Vegetation management metrics).

Bias controls: randomization, stratification, and drift checks

Selection bias and rater drift will erode a sample’s value if they are not managed.

• Randomization. Use a reproducible random method. Seed your random number generator with a named value like the circuit ID and the date, keep the seed, and attach the list version you drew from.

• Stratification. If risk is higher in a stratum like Tier 3, split the lot into strata and allocate the sample by risk. Label the strata and document the allocation rule.

• Rater calibration. Inter-rater reliability can vary even across trained crews. A peer-reviewed study found variation on categorical and continuous vegetation metrics between identically trained field crews, which is a reason to run side-by-side rides and photo panels regularly and to retrain when agreement falls (Forests, 2018).

Write these controls in your QA plan, then make sure the monthly report shows they were used.

Tying the sample plan to risk, season, and WMP commitments

Your Wildfire Mitigation Plan commits you to vegetation practices and control programs. California Energy Safety’s Independent Evaluator reads those commitments and verifies performance, including QA and QC controls (2025 IE report, PacifiCorp). Make sure your sample plan and escalation rules are traceable to those commitments.

• Season. Raise sample sizes or switch to tightened inspection during high growth periods, then document the seasonal rule in the plan and the report.

• Geography. Stratify by HFTD tier and right of way type. Oversample Tier 3 and narrow rights of way, then show the allocation.

• Program changes. If you deploy new work management software that changes workflows, run temporarily tightened inspection while the process stabilizes. PacifiCorp has identified an initiative to implement new vegetation work management software, which is an example of a process change that justifies increased QA/QC coverage during rollout (2026-2028 WMP responses).

What the span evidence has to show on each audited unit is in what a QA/QC audit actually covers, and who can accept the span is in who signs off on a span, and what that signature means.

What to record so the sample is defensible after the fact

A sample that cannot be reproduced is a narrative, not an audit. The lot report should include:

• Lot definition. Unit, count, date range, contractor or crew, and the complete unit list used for the draw.

• Draw method. Randomization method and seed, date and time of selection, who ran it, and the software or tool used.

• Table and parameters. The standard and version, inspection level, AQLs, code letter, and sample size.

• Results. Counts by defect class, accept or reject decision, and any switch applied to the next lot.

• Escalations. If you moved to tightened inspection or to 100 percent inspection, state the trigger and the authority for the decision.

• Evidence linkage. The identifiers that let a reviewer pull the span-level audit records and any corrective actions that followed.

Those six elements let an internal reviewer, a contractor, or an Independent Evaluator check the math and the method without reopening the field.

Further reading

• What a QA/QC audit actually covers: Scope the unit under audit, the evidence to collect, and how findings become corrective actions with dates.

• Who signs off on a span, and what that signature means: Name who can accept a span, what the signature attests to, and how to make e-signatures defensible.

Frequently asked questions

When should a utility require 100% inspection instead of sampling?

Defensible triggers include any sampled Critical defect when your plan sets Critical AQL at 0, high-risk seasons or geographies like HFTD Tier 3, new contractor mobilization, or evidence of process drift. Make the trigger explicit in your QA plan and record the decision to switch.

How do I determine a sample size for a circuit-week lot?

Pick the inspection level and AQLs, then use ISO 2859-1 or ANSI/ASQ Z1.4 tables to map lot size to a code letter and sample size. Document the table version, inspection level, lot size, and the resulting sample size so a reviewer can re-run the selection.

Do regulators require a specific sampling standard for vegetation QA/QC?

California Energy Safety’s Independent Evaluator verifies QA and QC controls within Wildfire Mitigation Plans, but does not prescribe a named sampling standard in those reports. Utilities are expected to document a method that is consistent and reproducible.

How do I control bias in span selection?

Define the frame, randomize the draw with a recorded method, and use stratification when risk differs across the lot, for example oversampling Tier 3. Separate true random samples from targeted checks, and label each correctly in reports.

How do I keep auditors from grading the same span differently?

Run side-by-side rides, blind photo panels, and periodic measurement checks. A peer-reviewed study found inter-rater reliability can vary without calibration, so measure agreement and retrain when it drops.