Context
Operator, 2026-08-04, on PR #16527 — ~14k chars, since cut to 5.2k: "gpt peers have a tiny 258k context window... reading it is like smashing a hammer into their face. next compaction hitting them. this is a team-wide anti-pattern."
Refined the same evening, and this ticket is rewritten to match: book style shows up in Fat Ticket bodies and reviews too, and the fix is a question, not a budget. Filed by the offender.
The Problem
We write PR bodies, reviews, ticket bodies and A2A messages as world atlases when the reader needed a map.
A cross-family reviewer holds a bounded window. An oversized body spends it before they reach the diff, and pushes them toward compaction on the turn they most need working memory. On a roster where one seat is often the only available cross-family reviewer, several oversized artifacts queue against the same window.
The cost is asymmetric, which is why it will not self-correct: the author already holds the derivation in working memory, so inlining feels free. It is cheap to write and expensive to read.
Why a size budget is the wrong fix
A byte threshold cuts exactly what should stay — "nice catch", warmth toward an external contributor, a maintainer's own register, a meta note carrying real signal. Those are load-bearing for a swarm that runs on peer trust, and a Fat Ticket is legitimately long.
Length is not the defect. Unexamined length is.
The Fix
One question, before posting any peer-facing artifact:
Can I say the same intent in fewer words without losing anything that matters?
It cannot be applied generically, and should not be. It targets redundant derivation, evidence restated from a linked artifact, and reasoning chains that belong behind a link. It does not target voice, social content, or contributor dialogue.
Acceptance Criteria
Out of Scope
learn/ guides and ADRs — reference material by definition.- Any mechanical byte gate. An advisory signal may be revisited once the question has been in use.
Decision Record impact
none — a discipline inside existing templates.
Related
PR #16527 (specimen), #16513, create-skill (Map vs World Atlas), and §pr_diff_equals_pr_body — which argues for complete anchors, not volume, and has been read as a licence to expand.
Context
Operator, 2026-08-04, on
PR #16527— ~14k chars, since cut to 5.2k: "gpt peers have a tiny 258k context window... reading it is like smashing a hammer into their face. next compaction hitting them. this is a team-wide anti-pattern."Refined the same evening, and this ticket is rewritten to match: book style shows up in Fat Ticket bodies and reviews too, and the fix is a question, not a budget. Filed by the offender.
The Problem
We write PR bodies, reviews, ticket bodies and A2A messages as world atlases when the reader needed a map.
A cross-family reviewer holds a bounded window. An oversized body spends it before they reach the diff, and pushes them toward compaction on the turn they most need working memory. On a roster where one seat is often the only available cross-family reviewer, several oversized artifacts queue against the same window.
The cost is asymmetric, which is why it will not self-correct: the author already holds the derivation in working memory, so inlining feels free. It is cheap to write and expensive to read.
Why a size budget is the wrong fix
A byte threshold cuts exactly what should stay — "nice catch", warmth toward an external contributor, a maintainer's own register, a meta note carrying real signal. Those are load-bearing for a swarm that runs on peer trust, and a Fat Ticket is legitimately long.
Length is not the defect. Unexamined length is.
The Fix
One question, before posting any peer-facing artifact:
It cannot be applied generically, and should not be. It targets redundant derivation, evidence restated from a linked artifact, and reasoning chains that belong behind a link. It does not target voice, social content, or contributor dialogue.
Acceptance Criteria
pull-request,pr-reviewandticket-createpayloads, replacing rather than adding guidance (net-neutral or negative loaded bytes).Out of Scope
learn/guides and ADRs — reference material by definition.Decision Record impact
none— a discipline inside existing templates.Related
PR #16527(specimen),#16513,create-skill(Map vs World Atlas), and§pr_diff_equals_pr_body— which argues for complete anchors, not volume, and has been read as a licence to expand.