From ca84b8df766f4f9783403fdf373603294dff0e9e Mon Sep 17 00:00:00 2001 From: Serina Mcfall Date: Mon, 24 Aug 2026 11:46:30 +1200 Subject: [PATCH 1/2] docs(launchpad): a caveat belongs where the claim is (#506) Adds rule 7 to AGENTS.md section 5. Rules 4 and 6 already cover not claiming a check you did not run and not fabricating; neither covers where a known limitation is written down. In the research notes reviewed on 2026-08-22 the limitation was almost always present and almost always in the wrong place -- stated in "Confidence and what is still unknown" while the conclusion it qualified sat in the Summary a hundred lines above. Six notes carried it, and each was filed as its own issue. This is the rule those six share. The rule gives a test rather than a preference: does the conclusion survive the caveat? "I did not measure logs" bounds scope and belongs at the bottom. "The artifact I measured cannot be identified" changes what the figures mean and belongs at the top. No check is added. Whether a caveat changes a conclusion is a judgement, which is what ADR-0019 says a deterministic script may not decide. Scope: one numbered rule in an existing list. The research-note layout, naming convention and index remain #132's, which is open and untouched here. Closes #506 Co-Authored-By: Claude Opus 5 Claude-Session: https://claude.ai/code/session_01LuDy9K8WDtowz5e2Uz2wjz Signed-off-by: Serina Mcfall --- launchpad/AGENTS.md | 8 ++++++++ 1 file changed, 8 insertions(+) diff --git a/launchpad/AGENTS.md b/launchpad/AGENTS.md index 179c4d5441d..e8dc40b7dd5 100644 --- a/launchpad/AGENTS.md +++ b/launchpad/AGENTS.md @@ -176,6 +176,14 @@ These are hard constraints, not style preferences. write `N/A - `. 6. **Do not fabricate.** No invented file paths, issue numbers, model names, or command output. If you do not know, write that you do not know. +7. **Put a caveat where the claim is, not where the discussion of it is.** If a + limitation changes how a Finding or Summary should be read, it goes *in* that Finding + or Summary. "Confidence and what is still unknown" is for what you did not check -- + it is not a place to qualify what you did claim. The test is whether the conclusion + survives the caveat: "I did not measure logs" bounds scope and belongs at the bottom; + "the artifact I measured cannot be identified" changes what the figures mean and + belongs at the top. A caveat present but sitting below the conclusion it qualifies + reads as agreement to everyone who stops at the headline. ### Filing an issue From 97a29f6b03ad92df9c8f022ab82feea423214956 Mon Sep 17 00:00:00 2001 From: Serina Mcfall Date: Mon, 24 Aug 2026 13:42:13 +1200 Subject: [PATCH 2/2] docs(launchpad): rule 7 names the confidence section by function, not by title Addresses the review panel's Medium 1 on #508, plus two adjacent defects in the same clause. The rule said "Confidence and what is still unknown" is for what you did not check, and is not a place to qualify what you did claim. Read literally, which is how section 5 asks to be read ("hard constraints, not style preferences"), that excludes the positive confidence assessments merged notes open that section with: 331-grafana-stack-cost.md:139 "High confidence in the numbers", and 327-grafana-stack-components.md:93 "High confidence on the component list". It also quoted one heading of four in circulation. 14 of 35 notes use "Confidence and what is still unknown", 12 use "Confidence and limits", 9 a "was not checked" variant, so an agent reading it literally on a note headed otherwise had to infer that the rule applied. Naming the section by function fixes both, and separates the two things it may legitimately do (assess trust in what was checked, record what was not) from the one it must not (qualify a claim made above it). The operative test, does the conclusion survive the caveat, and both worked examples are unchanged. The rewrap also drops the ASCII double hyphen for an em dash, matching the rest of section 5. Verified: python3 -m unittest discover -s launchpad/scripts -t launchpad/scripts -> Ran 215 tests, OK. Closes #561 Closes #562 Closes #564 Co-Authored-By: Claude Opus 5 Signed-off-by: Serina Mcfall --- launchpad/AGENTS.md | 13 +++++++------ 1 file changed, 7 insertions(+), 6 deletions(-) diff --git a/launchpad/AGENTS.md b/launchpad/AGENTS.md index e8dc40b7dd5..8c5821d95c9 100644 --- a/launchpad/AGENTS.md +++ b/launchpad/AGENTS.md @@ -178,12 +178,13 @@ These are hard constraints, not style preferences. output. If you do not know, write that you do not know. 7. **Put a caveat where the claim is, not where the discussion of it is.** If a limitation changes how a Finding or Summary should be read, it goes *in* that Finding - or Summary. "Confidence and what is still unknown" is for what you did not check -- - it is not a place to qualify what you did claim. The test is whether the conclusion - survives the caveat: "I did not measure logs" bounds scope and belongs at the bottom; - "the artifact I measured cannot be identified" changes what the figures mean and - belongs at the top. A caveat present but sitting below the conclusion it qualifies - reads as agreement to everyone who stops at the headline. + or Summary. A closing confidence section may assess how far to trust what you did + check and record what you did not — what it must not do is qualify a claim made above + it. The test is whether the conclusion survives the caveat: "I did not measure logs" + bounds scope and belongs at the bottom; "the artifact I measured cannot be identified" + changes what the figures mean and belongs at the top. A caveat present but sitting + below the conclusion it qualifies reads as agreement to everyone who stops at the + headline. ### Filing an issue