Skip to content

A brief names what it supersedes; a lane that cannot spawn says so - #14

Open
HackTuah wants to merge 1 commit into
mainfrom
chore/brief-supersession
Open

HackTuah wants to merge 1 commit into
mainfrom
chore/brief-supersession

Conversation

@HackTuah

@HackTuah HackTuah commented Sep 8, 2026

Copy link
Copy Markdown
Member

One convention, appended to CONVENTIONS.md. Documentation only.

Why

A coordinator was told to spawn subagents and run two review lanes. Both lane briefs it wrote still carried an earlier "do not spawn further subagents", copied forward from a message that had already been superseded, and neither brief said which instruction won. Both lanes then ran as two scoped passes by one reader.

The instruction was stale, not conflicting — which is exactly why it survived. A superseded sentence sitting beside a current one reads like a current one. That is the same defect class as a superseded claim surviving into a record, and worse in one respect: an instruction is acted on, not merely read.

The second half

Both lanes reported the limitation in their own verdicts, unprompted, rather than letting "two lanes" imply independence they did not have. That is the only reason it was caught before the record shipped. Independence is the entire content of a second lane — a second pass by the same reader shares its assumptions, its blind spots, and whatever the first pass concluded.

The rule

  • A brief states which instruction it supersedes.
  • A lane that cannot spawn says so in its verdict, not in a footnote elsewhere.

The failure cost nothing this time, which is the argument for writing it down rather than against it: the phrase that would have shipped — "two lanes, three rounds" — reads identically whether it is true or not.

Verification

format pass, compile pass, test pass, credo pass, optional deps pass, docs pass,
reuse pass (168 tracked; 55 in scope, 53 headered + 2 sidecar), licence files pass
Gate OK.  GATE_EXIT=0

A coordinator was told to spawn subagents and run two review lanes. Both lane
briefs it wrote still carried an earlier "do not spawn further subagents",
copied forward from a message already superseded, and neither brief said which
instruction won. Both lanes then ran as two scoped passes by one reader.

The instruction was STALE, not conflicting, which is why nobody caught it while
writing the brief: a superseded sentence sitting beside a current one reads
exactly like a current one. That is the same defect class as a superseded claim
surviving into a record -- and worse in one respect, because an instruction is
acted on rather than merely read.

Second half, and it is what made the first visible: both lanes reported the
limitation in their own verdicts, unprompted, rather than letting "two lanes"
imply independence they did not have. Independence is the entire content of a
second lane; a second pass by the same reader shares its assumptions, its blind
spots and whatever the first pass concluded. So the verdict names the condition
it was produced under.

The failure cost nothing this time, and that is the argument for writing it
down rather than against it. The phrase that would have shipped -- "two lanes,
three rounds" -- reads the same whether it is true or not.

Documentation only. No behaviour changes and no test moves with it.

Gate green, every step line reading pass:
  format, compile, test, credo, optional deps, docs,
  reuse (168 tracked; 55 in scope, 53 headered + 2 sidecar), licence files
  Gate OK.  GATE_EXIT=0

Signed-off-by: Ayla Croft <aylacroft@proton.me>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant