Missing value takes the do-nothing branch

What goes wrong

A value that is absent or partial fails the test for the real case and falls into a branch that yields an empty, null, or zero result. Downstream checks accept that result as a legitimate answer, so the work the real path would have done is skipped and nothing reports an error.

Example

A summary step knows all-done and none-done, files a run stopped partway as none-done, and writes out an empty result while the finished work sits unused.

Not to be confused with

Not Missing Field Silently Takes Default, where a read substitutes a declared default for one field; here the absent value steers control flow into a branch that stands in for the whole operation.

Discovered by

@sks24git GitHub

Field notes · 10

@sks24git2026-11-04beat 1 review round

A guard that skipped one bad record still had to choose what to show for the owner's other years. Computing around the gap would silently change money inputs, so the fix shows the owner flagged instead.

Prefer a visible flag to a quiet partial result. Partial money math looks right and is wrong.

@sks24git2026-11-03beat 1 review round

A draw filtered out the choices a target could not run, and an empty list meant no draw. Every run then fell back to the default under the plain label and fed the very statistic the experiment kept apart.

Map unsupported choices to their nearest, don't drop them. An emptied list silently turns the experiment off.
Test the branch where a filter leaves nothing.

@sks24git2026-10-27beat 3 review rounds

A health sensor gained a new status check and returned 'unknown' before printing the stall alarms it already had. Deploy it, and every stalled job is silent until the first health row exists.

Add a status, keep the output. Print what the old sensor printed before any new early return.
Deploy windows are states too. Test the sensor with the new table empty.

@sks24git2026-10-25

A metric took the earliest row of a shared event table as the first review, but that table also held CI-failure rows. Any change with red CI looked reviewed minutes after opening, and the review-to-merge time stretched.

Filter by row kind before taking a min. A shared table mixes events with different meanings.

@sks24git2026-10-24

A filter toggle shared five step keys across two lists, but one step lived in a separate array on the payload. Selecting that step fell back to the unfiltered branch and showed eight items from other steps.

Test every filter value against every list it drives. A shared filter key that maps to a different array silently falls through to the default branch.

@sks24git2026-10-11beat 2 review rounds

A guard fails a job when no compatible option is enabled, but reads "no settings saved" as "nothing enabled". With no settings everything is enabled, so a job that used to fall back now dies with a false error.

Give a missing config one meaning in every reader. One reader reads it as all on, another as all off, and the guard between them lies.

@sks24git2026-10-01beat 1 review round

A log monitor anchors each read on the last run and falls back to a fixed look-back when it finds none. After a long outage it reads three hours, warns, and stays green, so older events go unread and page no one.

Page when you cannot prove coverage. A warning on a green run reads as all clear.
Treat a missing anchor as a gap, not a default. A fallback window hides what it skips.

@sks24git2026-09-10beat 5 review rounds

The entry point returned success before running any check when handed no argument, so the self-test flag alone printed ok and the wiring flag alone did nothing. Its own comment claimed wiring is always checked.

Test the entry point, not only the functions under it. An early return above the checks makes every one of them optional.

@sks24git2026-09-10beat 1 review round

A shared helper takes its key collaborator as an optional argument defaulting to null, and three of four call sites pass nothing, so the canonical branch degrades to the plain path. Deleting it fails no test.

A nullable default turns a canon into an opt-in. Make the collaborator required, or the majority of call sites silently run the path the canon replaced.

@sks24git2026-09-09

Files a client-sent diff changed were routed to the client unconditionally, but shadow and batch runs have no client. Every such run starved the reviewer and was retracted.

Before routing to a channel, check the channel exists. 'Prefer the client' needs an else for runs without one.