Question and scope

This protocol measures the share of accepted cross-team requests that carry a complete definition of done at the moment work begins. A definition of done is the set of conditions the requester and the receiving team agree will make the deliverable acceptable. It is not a due date, a priority, or a description of activity. The result is a percentage, a breakdown by request type, and a list of the specific elements that are most often missing.

The question matters because most intake reviews check whether a request has an owner and a date, then stop. A request can pass those checks and still leave the receiving team guessing about format, audience, approval, and evidence. The guess usually becomes rework, and the rework appears later as a missed commitment rather than as an intake defect. Measuring the definition of done moves the diagnosis back to the point where the ambiguity entered.

Scope is one coordination team and the requests it accepted during a fixed window. Include requests that arrived through any channel, because informal requests are often the least complete. Exclude ideas that were discussed but never accepted, and exclude requests that were rejected before an owner was assigned. Record the window, the source systems, and the reviewer names so the count can be rebuilt.

Why the definition is hard to measure

Teams rarely store a definition of done in a dedicated field. It appears across a ticket description, a comment thread, an email attachment, and a verbal agreement in a meeting. The reviewer has to reconstruct the expected outcome from those artifacts. That reconstruction is a judgment, which is why the rubric must be fixed before any request is read. That is why the reviewer must document the source of each phrase rather than rely on a general impression.

A second difficulty is that completeness depends on the type of work. A report needs a named audience and a review step, while a data migration needs a rollback condition and a reconciliation check. One universal checklist will either be too vague to score or too long to use. Build a small core rubric plus a short type-specific addition, and score only the elements that the type actually requires.

Defining the coding unit

The unit is one accepted request with one owner and one deliverable. Split a request when each part could be accepted, scheduled, and reviewed independently. Keep it as one unit when the parts share a single approval or must ship together. Write the split rule into the codebook before sampling, because bundled requests are the most common source of disagreement between reviewers.

Use the version of the request that existed when the owner accepted it. Later edits can improve the definition, but counting those improvements would measure how much the team learned during delivery rather than how well it scoped the work. If the acceptance moment is unclear, use the first status change that names an owner. Mark the source of that timestamp in the record.

Give every unit a stable identifier and note its channel, requester, receiving team, request type, and acceptance date. These fields support the breakdowns later and make the inventory auditable. Do not include the eventual outcome in the coding file, because knowing whether the request succeeded can bias the reviewer toward calling the definition adequate or inadequate. A record that cannot be traced back to its source weakens every later comparison.

Building the rubric

Score five core elements for every request: the deliverable, the audience, the acceptance condition, the evidence, and the owner-reviewer pair. The deliverable must be a named artifact or completed action rather than a topic. The audience names who will use the output and for what decision, while the acceptance condition states what a reviewer will check. The evidence is the file, link, or record that must exist at handover. The final element names who produces the work and who approves it.

Add one type-specific element where the work requires it, such as a source-of-truth definition for a recurring report. A process change may need a rollback path, and a customer-facing deliverable may need a tone or brand reference. Keep the addition to one or two items so the rubric remains usable. Publish the full rubric with worked examples before coding begins, and keep that version visible during the review.

Score each element as present and specific, present but vague, or absent. Present and specific means a new team member could act on it without asking a clarifying question. Present but vague means the element is mentioned but does not constrain the work, such as asking for a report "soon" or for "high quality." Absent means the artifact does not address the element at all. The middle category carries most of the operational risk.

Sampling requests

Use a census when the team accepted fewer than roughly one hundred requests in the window. A census removes sampling error and is usually practical for a single coordinator. Above that threshold, draw a random sample and publish the method. Stratify by receiving team and request type so a large volume of simple requests does not hide a small set of complex ones.

Freeze the sample before reading the request bodies. Reading first and selecting later invites the reviewer to choose the most interesting cases. Store the random seed or the selection procedure with the sample file. If a selected request turns out to be out of scope, replace it using the next item on the randomized list and record the substitution. That keeps the sample random rather than convenient.

Include requests accepted near the end of the window even if work has not finished. Their definitions are still measurable, and dropping them would bias the result toward requests that were easy to close. If the definition was revised during delivery, score the acceptance version and note the revision separately. That separation keeps the headline about intake quality rather than about project learning.

Reading the artifacts

Read the request in a fixed order: the original ask, the acceptance comment, the linked specification, and any attachment that defines the output. Stop when the five elements have enough evidence to score. Continue to the later channels only when an element is still unresolved, and record which channel supplied the answer. That record shows where definitions usually live and where they tend to hide.

Quote the smallest phrase that supports each score. A quote keeps the judgment anchored and makes a second review possible. When the evidence is a link, record the link and the section heading rather than a summary. Do not paraphrase a vague requirement into something more specific, because that would credit the request with clarity it did not have. A direct quotation also makes disagreement easier to resolve.

Treat a template as evidence only when it is filled in for this request. A blank template attached to a ticket does not define the deliverable. Similarly, a link to a general style guide does not by itself define the audience for this particular output. The reviewer should ask whether a new team member could act using this artifact alone, and score accordingly.

Handling partial and disputed cases

Some requests define the artifact clearly but leave the reviewer unnamed. Score the reviewer element as absent while keeping the other elements as they stand. Do not collapse the whole request into a single pass or fail. The element-level scores are what tell the coordinator which part of the intake form needs a new required field. A missing reviewer is a common source of late rejection.

When two reviewers disagree, return to the quoted evidence rather than debating the request's intent. If the quote supports both readings, the rubric is ambiguous and should be tightened. Recode the disputed case and every case like it after the revision. Record the change in a short codebook log so later periods can be compared fairly. The log should name the rule that changed, not just the case.

Exclude a request only when the artifact is genuinely unavailable, such as a deleted ticket with no export. Mark it unverifiable and report the count separately. Do not guess a score from memory or from the eventual deliverable. A visible gap in the records is itself a finding about intake hygiene, and it should not be hidden inside a completeness percentage.

Calculating the share

Report two numbers rather than one. The first is the share of requests with all required elements present and specific. The second is the share with no element absent, which allows vague mentions to count. The gap between the two numbers shows how many requests look complete at a glance but still leave room for interpretation. Both numbers should include raw counts beside the percentages.

Break the result down by request type and receiving team. A team that receives many small, well-understood requests may score well without having a strong intake process. A team that receives complex work may score poorly even when its requests are better documented than average. Suppress groups with fewer than five requests so one unusual case does not look like a pattern.

Report the most common missing element and the most common vague element separately. Missing audience and vague acceptance conditions usually call for different fixes. The first may need a required field, while the second may need examples of what a testable acceptance condition looks like. Pair each finding with the specific intake change it suggests, and avoid a general instruction to "be more specific."

Analysis and reporting

Present the result as a baseline before presenting any target. A first measurement tells the team where it starts, not how far it has to go. Show the distribution across elements so readers can see whether the problem is concentrated or spread out. A single completeness percentage hides the difference between a team missing one element everywhere and a team missing everything occasionally.

Keep the reporting period consistent. A window that includes a holiday, a system migration, or a leadership change may not be comparable to a normal month. Note those events next to the result. The coordinator should be able to explain a movement in the number without reaching for a story that the evidence does not support. Write the note in the same row as the result.

Limits and sources of error

The sources cited here describe internal control, project scheduling, and agile practice in general terms; they do not provide a benchmark for definition-of-done completeness. The protocol therefore produces a local baseline, not an industry comparison. A team should not present its percentage as evidence that it is better or worse than another organization. Readers should treat the percentage as a starting point for a local conversation.

Reviewer judgment remains the largest source of error. Two careful reviewers can read the same thread and reach different conclusions about whether an acceptance condition is specific enough. The quote requirement and the second-review sample reduce that risk but do not eliminate it. Report inter-rater agreement so readers can judge how much confidence the number deserves. A single reviewer should not be the only check on a published number.

Review cadence

Review the measure after two full cycles before setting a target. One cycle is a baseline, and a target set from a single period is likely to reflect the period rather than the process. At the end of the review, choose one intake change, one owner, and one check date. A practical first change is to make the audience and acceptance condition required fields that cannot be saved empty.

Retire the measure when most requests carry all required elements and the rework trend has stopped moving. If the team keeps it, report completeness and vagueness together, because a rising complete share with a rising vague share means the form is being filled in without being understood. Link the result to the request acceptance checklist so the coding rubric and the operating checklist stay aligned. The purpose is to catch ambiguity while the request is still cheap to rewrite.

Sources