Bound a success rate when some eligible outcomes are unknown

Freeze who and what the cohort includes

Define the unit, inclusion period and binary result before classifying outcomes. A person who never qualified for the task is different from an eligible person whose outcome was not recorded. Keep unknown eligibility separate; these simple bounds require a fixed known eligible denominator.

Record why outcomes are missing where legitimately known, without collecting unnecessary personal detail. An interrupted task, a lost record and a withdrawn permission can have different treatment under the actual protocol. Do not change that protocol after seeing which denominator produces a preferred percentage.

Keep the known-outcome rate and full cohort together

For s successes, f failures and m unknown eligible outcomes, the known-outcome rate is s/(s+f), when s+f is positive. It describes classified outcomes only. It need not estimate the whole eligible cohort well when missingness is related to difficulty, access or dissatisfaction.

The fixed cohort has N=s+f+m units. Its all-missing-fail lower bound is s/N; its all-missing-succeed upper bound is (s+m)/N. These are logically possible extremes under the declared binary classification, not predictions about the missing cases or probabilities that an extreme is true.

Use the local count review without filling the gap

The local tool asks for three explicit whole counts, including zero where actually measured. Confirm the distinct eligible binary unit before reviewing. It preserves empty versus zero, rejects invalid or oversized values and returns no rate for an empty cohort. Unknown eligibility cannot be solved by typing a convenient missing count.

The optional Wilson output uses known results under a separately declared binomial model. It does not incorporate missing-outcome uncertainty and must not replace these bounds. The text download labels both separately; inputs stay local and a changed input invalidates the previous review.

Resolve three constructed missing-outcome claims

U1 — Constructed fixed cohort:8 successes,2 failures,4 unknown. Known-outcome rate 80%; whole-cohort bounds 8/14≈57.14% to 12/14≈85.71%. Decision: report all three counts and both summaries. Outcome: a clear uncertainty gap, not permission to choose 80% as the verified result for all 14.

U2 — Constructed unknown eligibility: four additional records have neither eligibility nor outcome established. Decision: hold the full-cohort denominator and investigate the permitted eligibility record. Outcome: do not silently call them four eligible failures, successes or missing units to force a bound.

U3 — Constructed fully missing cohort:10 eligible units, no known successes or failures. Decision: report the absence of classified outcomes, no known-outcome rate and bounds 0–100%. Outcome: genuinely uninformative evidence, not measured 0% completion or a numerical success claim.

Use the width to choose the next evidence question

A wide range shows that the decision may depend strongly on unknown results. Identify the legitimate source or protocol step that could resolve them. Do not pressure nonrespondents, bypass withdrawal or use unrelated activity to infer a missing preference or task success.

Narrowing the range requires actual permitted information or additional explicit assumptions. If you adopt an imputation model, label and justify it separately, retain sensitivity to that model and keep the original unknowns. Replacing them with the average hides the very question these bounds expose.

Save a complete classification statement

The worksheet keeps eligibility rule, outcome definition, classified/unknown counts, missingness reasons if known, both denominators, bounds and permitted next action. AAPOR’s reporting guidance distinguishes partial/no responses from explicit none, don’t know or refusal categories. Preserve the actual meanings rather than recoding them all as failure.

Primary guidance was read on 1 November 2026. All cases are constructed; no real cohort was classified, contacted or changed. The exercise does not certify a real preference, task result or business outcome.