Season · AI, Trust & Human Oversight · 3 of 16
A clean medication list can hide the source gaps that matter most.
Concept of the Week · Completeness needs a boundary
An output can be internally tidy and still rest on an incomplete view of the work. Omission bias may make leaving a missing source unresolved feel less consequential than changing a medication, while confirmation bias may encourage a reviewer to read familiar entries as proof that the list is complete. An Input Coverage Review makes source availability, freshness, expected-but-missing evidence, and excluded time windows visible before the output shapes a decision.
Trust rule: Treat an output as complete only when the system also shows what it could not see.
At hospital admission, a clinical pharmacist reviews a medication-reconciliation list assembled by an artificial intelligence (AI) assistant before a prescribing decision. The list looks complete, but the assistant's connection covers only the hospital record through the previous evening; a recent community-pharmacy prescription and one delayed renal-function result sit outside its data window. Nothing in the list marks those gaps.
The calmer view places a coverage panel beside the medication list. It shows which sources were available, when each source last refreshed, which expected source did not respond, and which laboratory result remains pending. The pharmacist can separate verified entries from those needing a patient check, a pharmacy call, or accountable clinical review before the medication decision proceeds.
Showing source coverage changes what the pharmacist can recognize as missing, which items require a manual lookup, and whether a medication decision should proceed, pause, or be escalated.
Ship this by recording source availability, last refresh, covered time window, and one owner for every missing input in the next ten representative medication-reconciliation cases.
The two panels contrast a clean AI-generated medication list that leaves source gaps unmarked with a coverage view that shows the clinical pharmacist what to verify before the medication decision proceeds.
Coverage before synthesis
AI can reduce retrieval, comparison, and drafting work by assembling medication evidence across available sources. It cannot turn unavailable data into a negative finding or replace patient input, clinical judgment, representative testing, or risk review. Treat a clean list as a hypothesis: preserve unavailable sources and pending inputs, then check consequential gaps with the patient, the responsible clinician, and the source systems.
Asset & Resource
Design, human factors, and risk teams can use the Input Coverage Review Card when reviewing a workflow in which an AI-generated list or summary may influence a medication decision. Revisit it when the source set, connection, time window, model, workflow, or clinical population changes.
Decision and consequence: Name the decision the output informs, who could be affected, and what happens if a relevant input is absent.
Available evidence: List the expected sources, the sources actually reached, their last refresh, and the time window each one covers.
Missing and stale inputs: Mark unavailable, delayed, filtered, or outdated evidence without treating silence as a negative finding.
Authority and recovery: State who may proceed, pause, challenge, or escalate, and how the missing evidence will be recovered or documented.
Revalidation trigger: Review the card after a source, permission, interface, model, workflow, or clinical population changes.
The National Institute of Standards and Technology (NIST) Generative Artificial Intelligence Profile asks teams to document assumptions and limitations, upstream-data dependencies, provenance, data quality, and the completeness and coverage of training and evaluation sources. Applied to runtime inputs, those records can make “the system did not see it” reviewable rather than hidden.
Light Wisdom & Reflection
“Absence of evidence is not evidence of absence.”
Where does a quiet gap in your product look like proof that nothing is missing?
Mindful Practice
Notice one list you rely on at home—a shopping list, calendar, or set of messages.
What can you see?
What might be absent because it was never added or has not yet arrived?
Next time, notice when an empty space begins to feel like confirmation.
Best,
Andreas Walden
Share this with someone designing for visible limits.
Previous in the season · visible uncertainty: CALM: Let Uncertainty Stay Visible