Repository navigation
spec(FIX-1833): record best fit's question as "Who should take the post?" - #3006
Conversation
…ke the post?" The implementation (#2997) changed the evaluator's question from #2955's "which delegate" to a wording that names nobody, at review's request. The goal held at 6/6 on legs a, c, g and h. This records the change in the retained decisions, as the PR said it would. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01XMg2dz8sMEdyzCQTRsaMTV
|
The latest updates on your projects. Learn more about Vercel for GitHub. 3 Skipped Deployments
|
There was a problem hiding this comment.
Problem this PR solves
Recording in retained spec what implementation already shipped: once the coordinator is a best-fit choice, #2955’s delegate-only evaluation question is the wrong shape; neutral “Who should take the post?” matches coordinator-flow.ts and DOCS.md.
/simplify verdict
The approach did not go too far — a post-implementation DECISIONS.md correction is the right size of change (vs reopening the whole FIX-1833 spec or a full FIX-1828 EVOLUTION pass). What adds complexity is stopping at one file while other retained surfaces still say the opposite.
| Approach | Trade-off |
|---|---|
| This PR (DECISIONS bullet only) | Smallest diff; fastest to merge. Risk: readers hit two canonical truths (PLAN S3 vs new bullet). |
| Same PR, thin sync | +PLAN.md S3, optional one-line ## How it got here (FIX-1481 pattern). Still no EVOLUTION table unless you want #2955 lineage. |
| Full lineage amend | EVOLUTION.md row + Settled refresh + POC README caveat (“#2955’s question” = POC set label only). Most files, clearest history. |
Recommendation: keep this PR’s scope (behavior already proved in #2997 / goals), but fold in PLAN.md S3 before merge. Shorten the new bullet and move goal leg scores to Settled or How it got here so “Locks in” stays rules-only like its neighbors.
What looks good
- Inline
*(Amended … in implementation)*matches other FIX specs and honestly separates approved wording from shipped code. - DOCS.md and code already agree; you are aligning DECISIONS with them, not inventing a new design.
- Separating “#2955 question wrong shape” from “S2: naming the coordinator in the prompt hurts confidence” is the right distinction.
Before merge (coherence)
PLAN.mdS3 still says “#2955's question stays as it is” — direct conflict with this amend. Point S3 at the DECISIONS bullet (and keep “recent lines” as the #2955-retained piece).DECISIONS.md· Settled bullet “holds under #2955's question” is fine as POC set S1 evidence if labeled that way; as written it reads like today’s production question. One clarifying phrase avoids misreads.- POC README conclusion “#2955's question needs no change” is now historical; a footnote or strikethrough-style note there (optional follow-up) reduces future confusion.
Nit (optional)
Shipped ROUTE_QUESTION includes the recent-lines suffix (coordinator-flow.ts); the quoted stem in DECISIONS is enough for intent — no need to duplicate the full string if PLAN points at DOCS.
Performance / reuse: N/A and no runtime impact.
Happy to approve after PLAN S3 is aligned (and ideally a shorter lock-in bullet). Direction is correct; incomplete cross-surface sync is the only meaningful concern.
Sent by Cursor Automation: Review For Best Approach
| with no delegate to choose: then there is no choice to make, and today's misses apply. | ||
| - **Best fit's question stays as #2955 words it.** A question that names the coordinator lowered | ||
| Jev's confidence on both kinds of ask (POC, set S2 against S1). | ||
| - **Best fit's question asks "Who should take the post?"** *(Amended 2026-10-10, in |
There was a problem hiding this comment.
Simplify: This bullet is doing normative lock-in + amendment provenance + POC narrative + goal evidence (#2997). Neighboring “Locks in” bullets are mostly one rule each; consider:
- Lock-in: question stem + one sentence why (fix(workforce): a best-fit follow-up keeps its conversation's context (FIX-1828) #2955 = delegate-only; coordinator is now a choice).
- History: one line under
## How it got here(FIX-1481 does this after merge). - Evidence: move “legs a,c,g,h 6/6” to Settled next to the other Jev/goal citations.
Same facts, less cognitive load when someone scans D1.
| - **Best fit's question asks "Who should take the post?"** *(Amended 2026-10-10, in | ||
| implementation; this spec first kept #2955's wording.)* #2955's question asks which delegate | ||
| should answer, which no longer fits once the coordinator is one of the choices. The POC's | ||
| objection was to a question that *names* the coordinator: that lowered Jev's confidence on both | ||
| kinds of ask (set S2 against S1). The new wording names nobody. On the goal it held: the chief | ||
| of staff picked itself for its own asks at 0.8 to 1, plain asks still went to the EM, and legs | ||
| a, c, g and h each passed 6 of 6 (#2997, `goal.md`). |
There was a problem hiding this comment.
Worth cross-linking implementation once in the amend PR body (#2997) so this does not read like a silent doc tweak — behavior-bearing, even though code already shipped.


The problem
FIX-1833's retained decisions say best fit's question "stays as #2955 words it". The implementation, #2997, changed the question to "Who should take the post?" at review's request. The spec now disagrees with what shipped.
What this does
It amends that one bullet in
specs/issues/FIX-1833/DECISIONS.mdto say what shipped and why.goal.md).No code changes.
What's asked of you
Approve the amendment. It records an engineering call already made and shipped. It doesn't change D1's direction: the coordinator is still a choice on a person's post, with the same floor.
Spec
specs/issues/FIX-1833/· Linear FIX-1833 · original spec PR #2977 · implementation #2997🤖 Generated with Claude Code
https://claude.ai/code/session_01XMg2dz8sMEdyzCQTRsaMTV
Generated by Claude Code