Truncation Bias
Reading a truncated result as if it were the whole result, so every counterexample is invisible.
Truncation bias is the error of drawing a conclusion about a whole set from a deliberately shortened view of it. It is endemic in command-line work, where piping a search through a "first twenty lines" filter is the reflex that makes long output readable.
The mechanism is brutal in its simplicity. If the first twenty matches all come from one directory, the shortened output looks like evidence that only that directory matches. The matches that would refute the conclusion are exactly the ones the truncation removed. The search returns a real result about a fake set.
The fix is mechanical: pipe to a count instead of a head, or scope the search to the exact region the conclusion is about, and do it as a separate deliberate step. Counting per directory is a two-second command and turns a guess into a fact.
The same shape appears wherever a view is partial by default: a paginated list, a log tail, a dashboard filtered to the last hour, a status page that shows only the checks registered so far. Any of them will support a confident, wrong summary. See Vacuous Truth for the closely related case where the set is empty rather than merely clipped.
See also5
Hand-picked in the note itself — the neighbours worth reading next.
Exhaustive Claim
A statement of the form "the last one" or "nothing else does this" — load-bearing, and only as good as the search behind it.
Method13 connections
Glob
Wildcard pattern matching over filenames, similar enough to regular expressions to mislead.
Systems & Tooling10 connections
Ground Truth
The measurement that arbitrates, as opposed to the artifact that merely reports.
Method30 connections
Observability
How much of a system's internal state can be inferred from what it emits.
Method15 connections
Silent Failure
A failure that produces no signal, so the absence of an error reads as success.
Method44 connections
Related3
Nearby in the graph rather than deliberately chosen. Looser, sometimes surprising.
Linked from12
Notes elsewhere in the wiki that reach for this one.
- Character EncodingSystems & Tooling
The mapping between characters and bytes, and the source of a specific family of invisible bugs.
- Coverage GateTesting & Verification
A build check that fails when coverage falls below a threshold, and what that pressure produces.
- CronSystems & Tooling
Time-based scheduling of recurring jobs, and the reasons scheduled jobs quietly stop working.
- DaemonSystems & Tooling
A long-running background process managed by the operating system's service supervisor.
- Exhaustive ClaimMethod
A statement of the form "the last one" or "nothing else does this" — load-bearing, and only as good as the search behind it.
- FalsifiabilityMethod
A claim is only worth something if you know what observation would refute it.
- GlobSystems & Tooling
Wildcard pattern matching over filenames, similar enough to regular expressions to mislead.
- Ground TruthMethod
The measurement that arbitrates, as opposed to the artifact that merely reports.
- ProcessSystems & Tooling
A running program with its own memory, environment, and identity.
- Reproducible CaseMethod
The smallest set of steps that reliably produces a fault, and the unit of real debugging.
- Silent FailureMethod
A failure that produces no signal, so the absence of an error reads as success.
- Vacuous TruthMethod
A statement that holds only because its subject set is empty — a green result that measured nothing.