New Member Missing From Parallel Registry

What goes wrong

A value — a status, a field, an error code, an enum case — must be listed in several hand-maintained parallel structures (an enum, a colour map, a translation table, a message switch, an audit list); a change adds it to some of them, so in the others it falls into the default branch or the 'other' bucket and silently lacks the behaviour or protection that list gives.

Example

A new order status 'on hold' is added to the enum but not to the colour map or the translations, so it renders grey with the raw key as its label.

Not to be confused with

The lists are code that runs, unlike a stale comment listing readers and writers (Obsolete Reader Writer Census); the newcomer is the gap, unlike a fix that misses an existing site (Sibling Site Missed By Fix); the test-side twin is Test Iterates A Hand-Copied List.

Discovered by

@sks24git GitHub

Field notes · 18

@sks24git2026-11-04beat 6 review rounds

A new kind of charge writes its link note in a format the balance audit does not know. Once enabled, every correct charge shows up in the audit as an orphan defect.

Teach every reader that classifies records about a new shape. Audits then flag correct data as broken.

@sks24git2026-11-03beat 6 review rounds

An erase promised to strip every row that names a person, but a table added later held free text the person typed. The erase never heard of it, so their own words outlive the account.

Erase by rule, not by list. A list of tables goes stale the day someone adds one.
Test every path that keeps a retry key. The one you skip is the one that pays twice.

@sks24git2026-10-26

A new result code went into the API model and two code sets, but not into the frozen contract schema those sets mirror. The schema then rejects every real result that carries the new code, and the parity tests go red.

List every registry keyed by a code before adding one. Hand-kept mirrors drift one at a time.
Test parity between mirrors, not membership. Only an equality check catches the missed copy.

@sks24git2026-10-25

A plan item cited a source for a "fingerprint before write" safeguard, but the task written from it hashed files in the agent's own sandbox, which nobody else edits. The ticket filed from that task would guard nothing.

Re-read the source when you name a mechanism. The label survives the copy; the mechanism does not.
Ask who else writes there. A fingerprint only guards a place with two writers.

@sks24git2026-10-24beat 2 review rounds

A spec made CI fail whenever a feature flag was off in production, while the flag checker's own rule says switching a flag off is always green. Enforce it and the kill switch for every flag breaks to police one.

A kill switch must always pass. Turning a feature off is the one move nobody may block.
Read the checker before you spec its rule. A spec that contradicts the tool ships a red build.

@sks24git2026-10-23

A new route with a fixed path segment sat where an older route takes a path parameter. The older route registers first, so every call to the new path lands in the wrong handler and is refused.

Check new paths against parameter routes of the same shape. The router takes the first match.
Spec the rounding rule. Rounded lines rarely sum to the rounded total.

@sks24git2026-10-20

A parser glued a whole numbered section onto the one before it, and the tests compared counts against totals made by that same parse. The data said complete, every test stayed green, and one section was missing.

Never check a parse against numbers from the same parse. A shared mistake agrees with itself.
Assert numbering has no gaps. A missing item shows up as a hole before it shows up as a bug.

@sks24git2026-10-16

A new state joined a shared enum, but one screen tracks busy slots in a plain string set the compiler never checks. Once the backend sends that state, those items vanish from the screen's counts.

Key status lists by the type, not by strings. A typed map breaks the build when a new state lands; a string set stays silent.

@sks24git2026-10-08

A response DTO copies fields from an internal record through an explicit allow-list, and a new field was added only to the internal record. The admin view showed it, while customers who were charged by it never saw it.

Grep every allow-list mapper when you add a field. A projection that copies by name drops new fields silently.
Test the customer path, not only the admin one.

@sks24git2026-10-06

A new price field reached the screen and the CSV, but each spreadsheet export kept its own hard-coded column list. The download dropped a billed rate, and the obvious fix would discount a price that never is.

Grep a sibling field across the repo. Each surface that prints it needs the new one too.
Flag the rate that breaks the shared rule. A shared formula applies itself to every column.

@sks24git2026-09-29

A new column is read by name, but the query selects an explicit column list that never got it. Every catalog read then throws at runtime, while tests that mock the repository stay green.

Add a new column to every explicit select list that reads it. A read by name fails only at runtime.
Test a new column against a real database. Mocks can't see a missing projection.

@sks24git2026-09-20beat 1 review round

A new wrapper script invoked the build tool, and the CI guard that enforces daemon-free runs only knew the old wrapper names. The script passed every check while its call sat outside the guard's view.

Register a new wrapper with every guard that lists wrappers. A guard keyed on names is blind to a name it never saw.
Pin the flag in the wrapper's own self-test. The guard trusts the list.

@sks24git2026-09-10beat 2 review rounds

A migration widened a column's allowed values and updated the constraint comment, leaving the column comment saying the default means 'everything else'. A reader trusting it counts a forfeiture as customer spend.

Widening an enum invalidates every 'everything else' sentence about it. Update the column comment in the same migration.
Docs aimed at money readers are part of the money path.

@sks24git2026-09-10

A new ledger operation settled its group with a total of zero while its own legs carried the whole moved amount. The journal is append-only, so every run leaves a permanent counterexample to a standing invariant.

Copy the sibling operation's exemption, not just its shape. An invariant that carved out one operation will select the next one you add.
Append-only means a wrong row is forever.

@sks24git2026-09-10beat 2 review rounds

A tick was wrapped in a timeout, but the work inside made blocking database calls on an IO dispatcher. The timeout cannot interrupt a socket read or a row-lock wait, so the promised bound on the tick is fiction.

A coroutine timeout only fires at a suspension point. Around blocking I/O it bounds nothing; set the deadline on the driver or the statement instead.

@sks24git2026-09-10beat 3 review rounds

A document bounded a fallback at one operation before the first successful read, but the read runs before every operation and a failing read keeps falling back. Every money split then ran the old arithmetic.

Bound a fallback by the failure that triggers it. "Once" holds only if the next read succeeds.
Count every fallback you log. A WARN per operation is invisible at fleet scale.

@sks24git2026-09-06beat 14 review rounds

A persist effect swallowed a storage write failure and never retried; on reopen an empty readable slot was read as 'resolved'. A live retry key was silently forgotten.

An empty slot is resolved only for the key you wrote. Distinguish 'never persisted' from 'persisted and cleared' or you drop live work.

@sks24git2026-09-06beat 1 review round

A response carried two verdicts built by two tables keyed on one state, and a new branch let a case reach one table its author never planned for. The body then refused the merge in one field and cleared it in the next.

One state, one verdict. Two lookups on one reading disagree the day a new branch reaches only one.
Pin the whole body, not the half you changed. The test read the refusal, never the clearance.