Condition Tested For One Outcome

What goes wrong

A true/false condition — a filter, a guard, an 'only active' rule — is tested only on inputs where it passes (or only where it rejects): the test data holds rows it keeps and none it must drop, or asserts refusal and never success. Making the condition always-true, deleting it, or widening it to refuse too much keeps every test green.

Example

A test of the 'active users only' filter feeds three active users and checks all three come back; delete the filter and the test still passes.

Not to be confused with

The condition runs but only one of its two outcomes is ever observed; not a numeric limit probed only from inside (Unchallenged Boundary Assertion), not test data that never reaches the condition (Unreached Branch Test), and not a negative case blocked by an earlier refusal (Overdetermined Negative Control).

Discovered by

@sks24git GitHub

Field notes · 25

@sks24git2026-10-29beat 3 review rounds

A new completion flag was tested only with true, never with the false, string or number a stored row can hold. Loosen the check to any truthy value and every test stays green while junk marks a run complete.

Test that only the exact proof value counts. A truthy check lets stored junk through.
Follow a new flag through to the final grade. A unit test of the rule misses the row that drops it.

@sks24git2026-10-25beat 5 review rounds

A watchdog reads the newest of several question markers on a thread, but every test fixture ever placed exactly one. Swap 'newest' for 'oldest' and the suite stays green while a stuck second question is never flagged.

Fixture the plural. If code picks 'the last one', one test must have at least two.
Test both sides of every threshold. A limit checked only above it can be deleted unnoticed.

@sks24git2026-10-23

A guard accepts a skip note only when it names the same period, but the one test feeds it a matching period. Drop the period check and a skip from an earlier quarter lets a later one ship without its required file.

Test each guard in both directions. A check proven only to pass hides the case it must reject.
Match skips by key, not by kind. "Some item was skipped" is not "this item was skipped".

@sks24git2026-10-23beat 2 review rounds

A rule takes a pluggable reducer and treats any an object it returns as a stop. The real reducer answers success as a an object too, so wiring it in turns every good result into a stop with no reason.

Pin a callback's return shape to the contract it wraps. A loose check matches success and failure alike.

@sks24git2026-10-21

A balance-chain rule said 'compare with the neighbouring statement' but never said the neighbour must be adjacent. One missing month and a correct file is rejected as a balance error instead of a coverage gap.

Define 'neighbour' by contiguity, not order. Gaps belong to the coverage check.
Copy every gate clause into the binding spec. Prose in a design doc is not a test target.

@sks24git2026-10-21

Tests drove a retry worker by hand and only checked its captured flag in one state. Drop the worker's start hook or hardcode the flag, and every test stays green while records stop retrying.

Start the real background loop in one test. Driving a worker by hand never proves it runs.
Test every value of a flag you capture. One state lets a hardcoded default pass.

@sks24git2026-10-21

Every link used one curve helper that goes flat when both ends share a row, and the layout packs unrelated items into one row. A link then runs straight through other items and draws a chain that does not exist.

Route same-row links around the row. A straight line through a packed row reads as a link.
Keep the guard when you swap renderers. The old one bent such links; the new one forgot.

@sks24git2026-10-21

A modal panel moved to a shared primitive, and its old backdrop was a tracked button. The new backdrop closes the panel but carries no tracking attribute, so every close by backdrop click drops out of the metrics.

Carry tracking hooks across a primitive swap. Every old close path had an event; list them before you delete the markup.

@sks24git2026-10-15

A query read both open and closed records, but every test fixture was open. Narrowing the query to open records kept all tests green while resolved records stopped counting.

Give each filter value its own fixture. A state no test holds is a state a refactor can drop.
Make test comments match fixtures. A comment that says closed on an open record hides the gap.

@sks24git2026-10-13

A catch turned one error into success and rethrew the rest, but tests only drove the caught case. Swallow every error there and the suite stays green.

Test both sides of a narrow catch. One test for what it catches, one for what it must rethrow.

@sks24git2026-10-13

An alert could be weakened until it never fired, and every test still passed. No test fed it the one state it exists to catch.

Give every alert a test case that must fire. An alert no test can trip is decoration.

@sks24git2026-09-21beat 3 review rounds

One admin response is built from two queries that must bind the same lock flag, and every test passed false to both. Flip the literal on one query and the two halves disagree while the suite stays green.

Test the arm that can fire. A flag only ever tested as false is a flag with no test.
Two queries, one truth, one test. Assert the halves agree under the same input.

@sks24git2026-09-19beat 7 review rounds

A new response field carried the one signal that a feature exists, and every test of that handler stubbed its source to null. Replace the expression with null and all tests stay green while the feature ships dark.

Drive every new field through its non-null branch once. A stub that is always null tests absence, not the field.
Mutate to a constant before trusting green. Nothing red means nothing covers.

@sks24git2026-09-14beat 1 review round

Every fixture in the suite starts with all collections non-empty, and the overrides only swap one non-empty set for another. The branches that positively state a section is empty never execute.

A branch that asserts absence needs a fixture that has nothing in it. Overriding one full set with another full set exercises the same arm twice.

@sks24git2026-09-11beat 1 review round

A «not enforced» suffix was tested present where it belonged, never absent where it did not, while its twin cell had both checks. Make it unconditional and every blocked account reads as merely measured, tests green.

Pin both polarities of conditional text. «Present when true» alone lets «always present» ship.
A sibling cell’s test is not this cell’s test. Each rendering of a rule needs its own pin.

@sks24git2026-09-10beat 4 review rounds

One change recorded in its header and runbook that a risky window had been refused and split in two, while the rationale comment beside the code still said the window was accepted. The operator reads the comment.

When a decision flips, a search tool its own name and fix every sentence. The rejected option survives nearest the code.
Say which text the operator reads first.

@sks24git2026-09-10beat 1 review round

Every test of a refusal planted rows matching all of its conditions. Delete one condition and the suite stays green while the check becomes an unconditional block on any instance holding pending work.

Plant a row on each side of every condition a guard names. One-sided fixtures make a widening mutation invisible.
Test that a guard passes, not only that it fires.

@sks24git2026-09-10beat 1 review round

A cleanup routine dropped a filter on the rows it deletes, and no test seeds the row shape that filter exists for. Putting the filter back changes nothing anyone can observe, so its removal was invisible.

Delete a guard only after a test goes red. If nothing fails, the guard was never covered, not never needed.

@sks24git2026-09-10beat 4 review rounds

A money figure is added back only when a marker row exists, and the suite's one fixture of that kind carries the marker. Drop the marker join and every test stays green while an opening balance loses the bonus.

A gate that decides money needs the fixture that can fire it. A fixture on the passing side proves the gate exists, not that it holds.
Check the twin branch — its coverage may be accidental.

@sks24git2026-09-10beat 2 review rounds

Every case in the suite seeds an uncovered balance on exactly one account. Drop the join key inside the correlated subquery and each row reports the instance-wide total, with the suite still green.

A correlation key needs a second tenant in the fixture. One account makes scoped and global sums identical.
Pin every WHERE term with a row it excludes.