Research question and scope
This protocol asks how much reviewed evidence a delegated work lane needs before it expands to new work types or higher volume. The unit is one reviewed sample cycle for a defined lane. The review measures the stability of the error pattern, not the absence of every error. A lane can have a small, known error rate and still be ready to grow.
Set the scope to one client and one lane with a written scoring guide. Record the work types, the sample size, and the reviewer for each cycle. If the scoring guide changes, the comparison starts again. A new guide can make the lane look better or worse without any change in the work.
Record the reason the lane wants to expand, such as new volume, a new work type, or a client request. The reason shapes the evidence that matters. A volume expansion needs a stable error rate under load, while a new work type needs a separate risk review. Write the reason before the sample so the decision stays focused.
Write the expansion proposal in one page with the current scope, the proposed scope, and the risk that changes. A one-page proposal is easier to review than a long report. It also forces the owner to name the specific change being approved. Record the approval date with the proposal.
Population and cycle unit
Define the population as the records available in the lane during each cycle, and the unit as the cycle. A cycle is a fixed period, such as two weeks, with a fixed sample size and a fixed reviewer. Include cycles with no errors, because stability includes the absence of new error types. Exclude cycles that were interrupted by a system outage, and note why.
Each cycle should report the number of records reviewed, the number of blocking errors, the number of minor issues, and the most common cause. Keep the same fields in every cycle so the comparison is possible. A cycle that reports a different set of measures cannot be compared with the one before it. Store the cycle report with the scoring guide version it used.
Keep the cycle length long enough to include a normal mix of work. A one-day cycle may miss the rare case that the sample is meant to find. Two weeks is a common starting point for a steady queue. If the work is seasonal, compare cycles from the same season rather than adjacent weeks.
Keep the same work mix in each cycle if possible. A cycle that happens to include more difficult cases will look worse even when the team has not changed. Record the mix and note any shift. If the mix changes materially, treat the cycle as a new baseline rather than a comparison.
Definitions of stability and readiness
Define stability as two consecutive cycles with no new blocking error type and a similar overall pattern. Define readiness as stability plus a written regression trigger and a named owner for the expanded scope. Stability alone is not readiness if nobody owns the next review. Write both definitions before looking at the data so the decision is not shaped by a good week.
Separate a blocking error from a minor issue using the scoring guide. A blocking error changes a customer outcome, a payment, or a private record. A minor issue is a formatting or tone problem that does not change the outcome. Only blocking errors should stop an expansion. Counting every minor issue equally will keep a lane from ever growing.
Write the regression trigger before expansion so it cannot be negotiated after a bad result. The trigger should name the measure, the threshold, and the person who will act. A trigger that depends on a vague feeling will not be used. A written one makes the rollback decision faster and less personal.
Define what counts as a new error type and who decides. A new type should be something the guide did not anticipate, not a variation of a known issue. Record the decision and the example. This prevents every unusual case from being treated as a new risk that blocks expansion.
Sampling and comparison
Keep the sample size and selection rule fixed across cycles. If the volume changes, report the change but do not change the sample to match it. A sample that grows with volume can hide a rising error rate because the reviewer sees more routine work. Use the same random or interval rule in every cycle.
Compare only cycles with the same work types and the same scoring guide. If a new work type is added, treat it as a separate lane until it has its own two cycles. Do not average a high-risk work type with a low-risk one. Report the results by work type and then give a combined view with the weighting stated.
Record the reviewer's agreement rate on a small overlap sample in each cycle. A score is only comparable if the reviewer applies the guide the same way. If agreement falls, recalibrate before comparing the cycles. The agreement rate is a quality check on the review itself.
Record the time between the sample and the feedback to the team. A slow feedback loop makes the next cycle less useful because the team may repeat the same error. Report the median feedback time. If it is long, fix the review cadence before adding more work.
Measuring readiness
Report the blocking error rate per cycle, the number of new error types, and the share of errors that were caught before the customer saw them. A lane that catches errors internally is healthier than one that relies on the customer to report them. Also report the time from sample to feedback, because a slow feedback loop makes the next cycle less useful. Put these numbers in one table so the owner can compare like with like.
Weight the result by the consequence of an error, not only by the count. A single blocking error in a payment lane may matter more than ten formatting issues in a newsletter lane. State the weighting so the owner can see how the conclusion was reached. A simple table is better than a composite score that nobody can explain.
Report the number of records that were corrected before delivery as well as the number that reached the customer. A lane that catches errors internally may still be ready to expand if the catch rate is stable. A lane that relies on the customer to find errors is not ready. Show both numbers.
Show the blocking error rate with a confidence interval when the sample is large enough to support one. A single rate can look precise while the underlying sample is small. If the interval is wide, say so and recommend another cycle. A cautious report is more useful than a confident one built on thin data.
Limitations and expansion decision
The sample cannot prove that an unobserved error type does not exist. It also cannot show whether the team will maintain the same quality when volume doubles. Treat readiness as a decision with a review date, not as a permanent certificate. Build a regression trigger that returns the lane to the earlier scope if the rate rises.
Give the owner the two-cycle evidence, the regression trigger, and the name of the reviewer for the expanded lane. Expand one work type or one volume step at a time. Record the date and the first cycle result after expansion. If the lane regresses, fix the cause before adding more work.
Set the first review after expansion close enough to catch a regression but far enough to include real volume. Two weeks is often enough for a steady lane, while a seasonal lane may need a full cycle. Name the reviewer and the date in the expansion note. An expansion without a review date is just a hope.
Name the first work type or volume step for expansion and leave the rest for later. Expanding everything at once makes it impossible to tell what caused a change. One step at a time keeps the evidence clean. Record the next review date and the person who will run it.
