Project Finish Criteria: Define Done Before Starting

Treat execution as a field procedure rather than a collection of tips. The procedure should still work when the day is busy, the easy option is unavailable, or another person takes over. A useful productivity system converts an important result into a small observable behavior that survives a normal week.[1] [2] [3]

This guide is for a reader who has a real project finish criteria: define done before starting decision in front of them. It focuses on the sequence, evidence, and recovery path—not on claiming that one answer fits every material, person, location, organization, or appetite.[1] [2] [3]

By Mitch Russo · Updated 2026-09-10[1] [2] [3]

1. Define three observable outcomes.[1] [2] [3]

2. Use recent evidence to estimate effort.[1] [2] [3]

3. Protect focused work on the calendar.[1] [2] [3]

4. Assign communication and interruption rules.[1] [2] [3]

5. Capture completion evidence.[1] [2] [3]

6. Review the week and change one design variable.[1] [2] [3]

Read the list once before acting. Circle the checkpoint with the weakest evidence. That is where the plan needs attention; polishing a later step cannot compensate for an unresolved early constraint.[1] [2] [3]

Describe the result in observable terms. Include the person, object, or business process affected; the real environment; the acceptable range; and the point at which the work must stop. For execution, an unacceptable outcome includes a system that adds tracking but not progress, depends on ideal motivation, or hides overloaded commitments.[1] [2] [3]

Separate hard constraints from preferences. A hard constraint can disqualify the method even when it is faster or cheaper. Write assumptions as assumptions, attach an owner, and give high-consequence unknowns a deadline for resolution.[1] [2] [3]

Before scheduling the work, assemble calendar history, completed-work evidence, workload constraints, baseline behavior, and a clearly defined result. Ask which single missing fact could reverse the decision. Resolve that item first; lower-impact questions can remain in the test log.[1] [2] [3]

Set up the workspace and communication path before the demanding step. Define three observable outcomes; then confirm that use recent evidence to estimate effort. Make the stop authority explicit. The person who notices a problem should not need to negotiate permission while the exposure or failure is growing.[1] [2] [3]

1. Define three observable outcomes.[1] [2] [3]

Document the starting condition before you define three observable outcomes. Without a baseline, the team may notice change but cannot judge whether it is acceptable. The baseline for execution should be brief, observable, and saved with cue, action, duration, completion evidence, interruption, result, and one design adjustment.[1] [2] [3]

Checkpoint: before moving to “use recent evidence to estimate effort,” write one sentence describing what passed, what did not, and who owns the unresolved item.[1] [2] [3]

2. Use recent evidence to estimate effort.[1] [2] [3]

Treat this as the handoff checkpoint: use recent evidence to estimate effort. The person receiving the work should be able to state the result, the remaining risk, and the next review date. If the handoff requires hidden context, the execution instruction is not finished.[1] [2] [3]

Checkpoint: before moving to “protect focused work on the calendar,” write one sentence describing what passed, what did not, and who owns the unresolved item.[1] [2] [3]

3. Protect focused work on the calendar.[1] [2] [3]

Close the loop after you protect focused work on the calendar. Record the actual outcome, including friction and near misses, rather than only marking the task complete. Use that result to revise the next execution attempt while the details are still fresh.[1] [2] [3]

Checkpoint: before moving to “assign communication and interruption rules,” write one sentence describing what passed, what did not, and who owns the unresolved item.[1] [2] [3]

4. Assign communication and interruption rules.[1] [2] [3]

Use a two-person check when the consequence is meaningful. One person should assign communication and interruption rules; the other should compare the action with calendar history, completed-work evidence, workload constraints, baseline behavior, and a clearly defined result. The second check is not bureaucracy—it catches a mismatch while the work is still reversible.[1] [2] [3]

Checkpoint: before moving to “capture completion evidence,” write one sentence describing what passed, what did not, and who owns the unresolved item.[1] [2] [3]

5. Capture completion evidence.[1] [2] [3]

Document the starting condition before you capture completion evidence. Without a baseline, the team may notice change but cannot judge whether it is acceptable. The baseline for execution should be brief, observable, and saved with cue, action, duration, completion evidence, interruption, result, and one design adjustment.[1] [2] [3]

Checkpoint: before moving to “review the week and change one design variable,” write one sentence describing what passed, what did not, and who owns the unresolved item.[1] [2] [3]

6. Review the week and change one design variable.[1] [2] [3]

Ask what would make this action wrong in the present setting, then review the week and change one design variable. Compare the answer with calendar history, completed-work evidence, workload constraints, baseline behavior, and a clearly defined result. This counter-check is especially valuable when a familiar method is being reused with a different person, product, location, or workload.[1] [2] [3]

Checkpoint: before moving to “schedule the next inspection or review,” write one sentence describing what passed, what did not, and who owns the unresolved item.[1] [2] [3]

Do not launch the whole plan as the experiment. Trial one week or one deliberately bounded work cycle, observe without coaching the result toward success, and stop when a system that adds tracking but not progress, depends on ideal motivation, or hides overloaded commitments is present. A bounded failure is useful evidence.[1] [2] [3]

Keep the test honest. Do not add help, favorable conditions, or expert intervention that will be absent during normal use. If the difficult case cannot be tested responsibly, escalate it to the qualified person or authority who can evaluate it.[1] [2] [3]

• Choosing a tool, product, setting, contract form, or template before the execution requirement is defined.[1] [2] [3]

• Testing only the easiest condition and assuming the result represents normal execution use.[1] [2] [3]

• Changing several variables together, which hides the cause of success or failure.[1] [2] [3]

• Continuing after a system that adds tracking but not progress, depends on ideal motivation, or hides overloaded commitments because time or money has already been invested.[1] [2] [3]

• Finishing the visible task without recording cue, action, duration, completion evidence, interruption, result, and one design adjustment.[1] [2] [3]

When a mistake appears, stabilize first. Protect the person, material, rights, equipment, food, environment, or client experience involved. Return to the first checkpoint contradicted by the evidence, revise one variable, and create a new stop rule before trying again.[1] [2] [3]

This is an implementation framework, not medical or mental-health treatment. Reduce the workload when the system creates sustained distress, and seek qualified support when health or disability affects execution.[1] [2] [3]

Authoritative starting points:[1] [2] [3]

• CDC shared-goals worksheet[1] [2] [3]

• NIH behavior-change overview[1] [2] [3]

Confirm that a source applies to the exact model, jurisdiction, land manager, product category, transaction, clinical situation, or activity. Save the access date and pair general guidance with current manufacturer instructions or individualized professional advice when appropriate.[1] [2] [3]

Close the work with a short audit owned by the person who can change the commitment or environment. Preserve cue, action, duration, completion evidence, interruption, result, and one design adjustment. That record is the starting point for maintenance, training, renewal, or the next controlled test.[1] [2] [3]

Turn the final note into a short checklist for the next person. Include the six checkpoints, the approved range, a photograph or example where useful, the stop rule, and the escalation contact. A workflow is not delegated until another person can recognize both a good result and a reason to stop.[1] [2] [3]

Write the desired result and the unacceptable outcome. Complete checkpoint one using a current source or direct observation. Then prepare one week or one deliberately bounded work cycle. If the critical evidence is missing, use the time to send one precise question instead of improvising.[1] [2] [3]

The goal of this short session is not to finish execution. It is to reach the first defensible action with the stop rule already in place.[1] [2] [3]

• 14-day productivity experiment[1] [2] [3]

• weekly review template[1] [2] [3]

• results resource library[1] [2] [3]

What should be verified first?.[1] [2] [3]

Verify the fact that could disqualify the entire approach. In this workflow that usually means define three observable outcomes, followed by a check that you can use recent evidence to estimate effort under real conditions.[1] [2] [3]

How detailed should the written plan be?.[1] [2] [3]

Detailed enough that another capable person can perform the next checkpoint and recognize a system that adds tracking but not progress, depends on ideal motivation, or hides overloaded commitments. For most situations, one page plus the controlling sources and cue, action, duration, completion evidence, interruption, result, and one design adjustment is more useful than a long narrative.[1] [2] [3]

When is a small test not appropriate?.[1] [2] [3]

Skip informal testing when a recall, emergency, legal restriction, clinical concern, structural question, food-safety uncertainty, unknown hazardous material, or manufacturer prohibition requires an authoritative response first.[1] [2] [3]

What evidence should be saved afterward?.[1] [2] [3]

Save cue, action, duration, completion evidence, interruption, result, and one design adjustment. Add the source date, the person who approved the result, and the date or trigger for the next review.[1] [2] [3]

What if the first attempt fails?.[1] [2] [3]

Stop, protect the affected people and property, and preserve the evidence. Identify the earliest failed checkpoint, change one variable, and decide whether a second bounded test or qualified professional review is the responsible next step.[1] [2] [3]