Table of contents
Published October 25, 2026.
This protocol examines whether the source is reconstructable and consistent with the recorded final state across records with a source artifact and a later outcome in Nonprofit Operations operations. It evaluates the reliability of the administrative record, not the quality of a professional judgment or the truth of an underlying claim.
Research question and decision use
The primary question is whether the source is reconstructable and consistent with the recorded final state. The unit of observation is one eligible record followed from its triggering evidence to a verified disposition or the end of the observation window. Leadership may use the result to improve forms, ownership, system configuration, reviewer coverage, or escalation rules. It should not use the result as an automatic ranking of people, clients, patients, or teams.
Population and eligibility
Define records with a source artifact and a later outcome from a fixed export. Record the query time, systems, date boundaries, statuses, exclusions, deduplication rule, and resulting count. Freeze a reproducible query or a copy before sampling. A record belongs because it meets the written rule, not because it appears easy or unusual. Preserve excluded counts by reason so the denominator can be challenged.
Sampling protocol
Draw a stratified sample across age, state, owner group, channel, exception type, and final disposition. Include completed and unresolved work. Record the random seed and every replacement. Oversample rare exceptions for learning only when weighted and unweighted results remain separate. Set the sample size before classification and disclose any subgroup too small for a stable, privacy-respecting estimate.
Evidence dictionary
Define each field used for source presence, match, missing state, correction, reviewer agreement, and time, including allowed values, source system, timestamp rule, and treatment of missing, unknown, and not applicable. Those states must remain distinct. Retain a restricted crosswalk to source identifiers. When sources conflict, keep both values and create a conflict flag rather than silently selecting the cleaner result.
Classification rubric
Score intake completeness, ownership clarity, exception visibility, decision routing, communication evidence, and completion evidence separately as present, partial, absent, or not applicable. Require a source reference and a concise rationale for every partial or absent classification. A completed status without matching evidence is an exception, not proof of completion.
Reviewer calibration
Train two reviewers on a small calibration set. Compare classifications field by field, discuss ambiguous definitions, revise the codebook, and repeat before the main sample. Double-review a prespecified subset and report agreement counts. Preserve initial ratings and adjudications so readers can distinguish genuine agreement from post-review harmonization. Escalate professional questions rather than answering them to complete a classification.
Analysis plan
Report counts, rates, medians, aging bands, denominators, and missingness for source presence, match, missing state, correction, reviewer agreement, and time. Use uncertainty intervals when the design supports them and show absolute counts beside rates. Segment only by prespecified operational variables with enough observations. Inspect case narratives behind important differences before assigning a cause. A higher exception rate may reflect stronger detection, a harder case mix, or a new procedure rather than weaker performance.
Sensitivity and exception analysis
Repeat the main calculations under documented alternative clock, missing-data, duplicate, and denominator rules. State whether the practical conclusion changes. Separate reopened from first-pass-complete records while retaining the reason for reopening. New information, changed direction, reviewer correction, and system failure have different implications. List material exceptions rather than forcing them into an average.
Privacy and role boundaries
Use the minimum information needed for records with a source artifact and a later outcome, limit access, log exports, and apply the retention schedule. Obtain approval before reusing the dataset. Patient or client communication should use neutral language, and population patterns must not become assumptions about an individual. The nonprofit operations support role may coordinate approved administrative steps; authorized professionals retain clinical, legal, financial, employment, and contractual decisions.
Limitations
The study cannot prove that the source is truthful or that the final decision was correct. The method observes documentation during a defined period and cannot prove that the workflow caused a later outcome. System migrations, policy changes, staffing shifts, missing off-system contacts, and reviewer error may influence findings. Public sources provide context and method guidance but do not validate local records or substitute for qualified professional review.
Replication package
Retain the population query, exclusion log, sample seed, codebook, reviewer assignments, source references, classifications, adjudications, calculations, and publication table. A later cycle should use the same definitions or disclose changes before comparison. Replication succeeds when a qualified reviewer can rebuild the sample and results without undocumented oral context.
Conclusion
A defensible result is a transparent measurement system with stable definitions, traceable evidence, visible uncertainty, protected information, and an authorized owner for every action that follows.