docs(roadmap): line 2 is delivered, so it leaves the roadmap - #205
Merged
Conversation
"Say real things" shipped: the off-loop topic pool, weighted by the listener's spoken language and their timezone, landed via #201 and #203. The file's own rule is that a delivered line is deleted rather than archived, so the row and its section go, and spec 13 becomes the record — its two pointers into the deleted section are retargeted at itself. What survives the line is tracked elsewhere: #202 joins the by-ear list, and #44 closes on that issue's first box. Line 0's other half is resolved too, though not the way the file predicted. It called for the parallel clock and grounding work to be reconciled into one change; what actually happened is that #165 landed the grounding, #191 then landed only the weekday-and-date half of #162, and the play-progress half was deliberately dropped because the look-ahead makes a countdown false by the time it airs. Recorded as a decision, not a merge, so nobody re-proposes the progress fields. What is left of line 0 is the dropped first line, unchanged. Two cross-references pointed at the deleted line. #98's eval track is the sharper of them: line 2 shipped a prompt whose whole job is stochastic, and the only thing that caught its first draft getting that wrong was a person reading a dev log — so the eval is overdue, not upcoming. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018KVNGUoCkDfWFdLwkGWsAi
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Keeps
ROADMAP.mdtrue after spec 13 landed. Related:specs/spec13/13-real-world-topics.md,specs/STATUS.md.What changed
Line 2 "Say real things" is delivered, so it leaves the file. The off-loop topic pool — weighted by the listener's spoken language and their timezone, fetched off the live loop — shipped in #201 and #203. The roadmap's own rule is in its header: a line is deleted when it is done, not archived. So the table row and the whole
## 2section go. Nothing that survives the line is lost: #202 (the by-ear gate) joins the by-ear list under "Not on this roadmap", and #44 is recorded there as closing on #202's first box. Both are already indexed inspecs/STATUS.md.Line 0's first half is resolved — but not the way the file predicted. It called for the parallel clock and grounding work to be "reconciled into one change before either merges". What actually happened: #165 landed the grounding (the four-state music union and the anti-fabrication rules), #191 then landed only the weekday-and-date half of #162, and that PR's own message records why the other half was dropped — a beat is composed two deep ahead of air, so a played/remaining countdown stamped at compose time is already false when spoken, and it works against the grounding rules #165 had just added.
MusicStatecarries no progress fields today, confirming it. The section now records that as a decision rather than a merge, so nobody re-proposes the progress fields. The only PR still open is #168, a conflicting STATUS.md-only chore that blocks nothing. The dropped-first-line half of line 0 is untouched — still open, still P0, still carrying its 2026-09-04 re-check.Two cross-references pointed at the deleted line. Line 4's P column said its eval half is "P2 once lines 1/2 land"; and the #98 bullet said lines 1 and 2 "are entirely stochastic behaviour". Both now read against reality — and #98 gets sharper for it: line 2 shipped a prompt whose entire job is stochastic, its first draft got it wrong (the material scrubbed back to mood), and the only thing that caught that was a person reading
.dev/dev.log. That makes the eval overdue rather than upcoming.Spec 13 absorbs the deleted section's pointers. Its header called itself "the 'say real things' line of ROADMAP.md §2" and its non-goals cited "ROADMAP §2" for language-is-not-region. Both now point at the spec itself, which is the durable record once the roadmap line is gone.
One doc-debt note. #104 already tracks DESIGN.md's stale "three network calls"; spec 13's topic fetch makes it four in shipped code, so the bullet says so.
Counts: five lines, not six. No renumbering — the header says the numbers are names, so lines stay 0, 1, 3, 4, 5 and only the order moves.
Verification
Every claim was checked against the repo and GitHub rather than from memory:
gh pr list(one open PR, #168, CONFLICTING, STATUS.md only),gh issue viewon all seventeen issues the file names (all still open except #145, which the file already records as closed in its favour), andgit logfor #199/#200/#201/#203.pnpm test,pnpm run typecheckandoxlintall exit 0 — run bare, not piped.Peer review
Peer review (codex gpt-5.6-sol): 2 findings, 2 applied, 0 dismissed — the reconciliation account overstated what merged (the play-progress half was deliberately dropped in #191, not landed), and deleting §2 would have left two dangling pointers in spec 13. Both verified against
git logandsrc/contracts.tsbefore applying.AI coding brief
Original request: "再开一个 pr 更新下我们的 road map，应该有些是做完了的" — open a PR bringing the roadmap up to date, since some of it is now done. The why: spec 13 merged an hour earlier and the file still described it as upcoming work, in future tense, across a full section.
Manual interventions: none. One judgement call taken without asking: deleting line 2 outright rather than marking it done, because the file's header states that rule for itself.
Retro: checking the state of every issue and PR the file names, before editing, is what surfaced the two stale cross-references. But the peer review still caught a claim I had taken from the roadmap's own prose rather than from
git log— the reconciliation that never happened as written. A roadmap edit is a set of factual claims; next time read the history for each one first, not just the issue tracker.🤖 Generated with Claude Code
https://claude.ai/code/session_018KVNGUoCkDfWFdLwkGWsAi