Repository navigation
DESIGN: the discovered roster does not run unshrunk - 11.7% is declined by home policy - #8819
Merged
Merged
Conversation
…ed by home policy The CI paragraph claimed the 2026-08-13 operator directive means "the whole discovered roster runs unshrunk". That is true of the ROUTED roster and false of the DISCOVERED one, and the difference is load-bearing: a reader who takes it literally concludes that authoring a witness is sufficient for it to execute. It is not — a witness under a long or live home is discovered, counted, and never run. Measured on main run 32515441229 (e336acc, 2026-08-21T18:51Z): offered=11615 routed=10253 declined_long=551 declined_live=811 1362 discovered sites — 11.7% — are declined by HOME POLICY before the fold sees them, so `planned` is the routed roster and never the discovered one. The directive is about SELECTION, and the decline mechanism is not a selection flag, so the directive is not violated by it. Conflating the two questions is what made one true sentence and one false sentence read as the same claim. The floor's own line is already honest ("every discovered site is exactly one of these"); only this document overclaimed. Also refreshes the cited counts, which were run 32203661157 of 2026-08-19 and had gone stale by roughly 1200 witnesses. Replaced rather than annotated — two count sets in one clause is two accounts of one fact. Edited in the carrier (dag/gunbc/design_document.dag); DESIGN.md is the generated projection and is regenerated to match, verified by running generated_artifact_gate main_wet over the edited carrier. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FwPMTY6Myy3scaMNn33cg5
gunbai-bot
Bot
changed the title
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What was wrong
The Building-&-checks CI paragraph claimed the 2026-08-13 operator directive means "the whole discovered roster runs unshrunk".
That is true of the routed roster and false of the discovered one.
Measured on main run
32515441229(e336acc4f0, 2026-08-21T18:51Z):1362 discovered sites — 11.7% — are declined by home policy before the fold ever sees them.
plannedis the routed roster and never the discovered one.Why it is load-bearing rather than a wording nit
A reader who takes the sentence literally concludes that authoring a witness is sufficient for it to execute. It is not: a witness under a long or live home is discovered, counted, and never run.
That inference has a live cost. #8801 merged 7 new witnesses into
v2.test.long.fold_lowering; they are the evidence its fold-lowering split is correct, and they do not run. Thedeclined_longcounter moved 551 → 558 — exactly +7 — which proves discovery reaches them and proves they are declined, in one number.What is NOT claimed
The directive is about selection, and the decline mechanism is not a selection flag, so the directive is not violated by it — the fold still has no batch id, worker role, or selection flag to configure. Conflating the two questions is what made one true sentence and one false sentence read as the same claim.
The floor's own output is already honest about this (
every discovered site is exactly one of these). Only this document overclaimed.Also in this diff
The cited counts were run
32203661157of 2026-08-19 and had gone stale by roughly 1200 witnesses (planned=9008 … passed=8702 known_red_held=306→planned=10253 … passed=9946 known_red_held=206). Replaced rather than annotated: two count sets in one clause is two accounts of one fact.Provenance
Edited in the carrier,
dag/gunbc/design_document.dag.DESIGN.mdis the generated projection and is regenerated to match — verified by runninggenerated_artifact_gate main_wetover the edited carrier, which returned exit 0 with the corrected text still present inDESIGN.mdafterward.All counters in this diff were read first-hand from the run log, not carried over from notes.
known_red_held=206in particular is not what arithmetic over the other counters would have predicted.🤖 Generated with Claude Code
https://claude.ai/code/session_01FwPMTY6Myy3scaMNn33cg5