Stored marker read as current condition

What goes wrong

A decision reads a stored marker that once summarised a condition instead of evaluating the condition itself. The marker is written at one moment and the underlying state keeps moving, so every later read answers from the stand-in. The reader wrongly relies on the marker still agreeing with the live state it was derived from.

Example

An account page decides whether someone belongs to a group by reading the first invitation code they ever redeemed, so a member who joined through a later invitation is asked to join again.

Not to be confused with

Not a stale cache of an expensive computation, and not the nearest catalog species, a frozen measurement served as current, because here the marker was never wrong when written and only diverges once the real state changes beneath it.

Discovered by

@sks24git GitHub

Field notes · 8

@sks24git2026-11-05beat 1 review round

A follow-up charge waived what an earlier one had billed, but only once that earlier charge was booked. When the follow-up came first, both charged the same defect.

Treat a charge that will fire on its own as already made. Waivers keyed on booked charges miss it.
Freeze a promised deadline when you show it. Recomputing it later can move it earlier.

@sks24git2026-11-05beat 3 review rounds

A flag that says a call was paid is set before the call runs, and the call can skip itself when its own budget is spent. Rows then burn their retries as if paid and end closed as errors, never classified.

Set a fact flag from the result, not the intent. A flag set before the call lies whenever the call skips.
Give one owner to a row's lifecycle. Two schedulers on one row fight each other.

@sks24git2026-11-04

A page cuts any reference to 12 characters and labels it as a commit hash. A branch name or tag passes through the same cut and shows up as a fake hash.

Check a value's shape before labelling it. A truncated name looks like a hash.

@sks24git2026-11-04beat 3 review rounds

A timeline marks its first unfinished step as running, and one mode hides a setup step. While that hidden step runs, the review step reads as running under a headline that says setup is still going.

Pick the in-progress step from the current stage only. "First unfinished" jumps stages when a step is hidden.
Test every mode's step list. A step hidden in one mode moves the marker there.

@sks24git2026-11-03beat 3 review rounds

A freshness check aged a last-known-good marker by its own commit date. After a quiet night, the first bad commit made a current marker look stale, and the guard switched itself off exactly when it was needed.

Age a marker from when it fell behind. Its own timestamp only says when it was made.
Test with more than one item in the list. A one-element fixture hides first-versus-last bugs.

@sks24git2026-11-03beat 2 review rounds

A list began to prefer a stored final result over the record's own state, and a result can be voided later. A hunt that used to read failed now reads done and waiting for a verdict nobody can give.

When a derived value overrides a stored one, carry its void rule too. Without it the override resurrects dead state.
Test the case your comment calls rare: rare paths regress silently.

@sks24git2026-11-02beat 1 review round

A page asked "already used?" by reading the first code an account redeemed, not whether it was in the group. A member who joined by a later invite was asked to join again, and old links bounced them off home.

Ask the fact you mean, not a field that once implied it. A first-use column is not membership.
Run the "already done" check before you redirect, not after.

@sks24git2026-10-28beat 1 review round

A weekly series was built against a hand-written sample, then wired to the real producer, which skips empty weeks and flags a sleeping source. The screen compares by position and hides the series with a false reason.

Compare time series by date, not by index. A missing row moves the baseline and flips the sign.
Re-read the contract when the mock goes live. The real source has states the sample never had.