docs: no predecessor app — nai_maker leaves the docs, the parity checklist goes, Candidates moves to Phase 4 #51
Loading…
Add table
Add a link
Reference in a new issue
No description provided.
Delete branch "docs/drop-nai-maker"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Summary
This follows your answers to #39's questions. The graph is the answer to what a story tool should do, not hardcoded memory, author's note or lorebook systems. So nothing in Tsumugi is measured against an earlier app any more, and every reference to
nai_makeris gone from the docs and the code comments.PROJECT_PLAN.md: Phase 2's title drops "→ parity", its ships-when drops "everything it did, better", the Goal drops the parity target, and the importer line is gone.REQUIREMENTS.md: the Parity with … section from #39 (covered, dropped and open rows) is removed. So are the stash aside in §1 and the no-importer line in §10.README.mdand the ADR index no longer mention the app.***, generation history, archive), 0013 (fixed steering fields), 0016 and 0024 (generation history), 0017 (the weak context builder, the mid-sentence cut) and 0018. None of these changes a decision, so they carry no amendment line (the #39 precedent).humannode its choice needs. Phase 3 keeps Summarize chapter and the three selection templates.PROJECT_PLAN.md: the template moves from Phase 3's list to Phase 4's.StoryStatus,StoryListItem,SetStoryArchivedandStoriesPage. The CIFS rationale inProgram.csandVolumeStartupValidatorTestskeeps its reason.Verification
git grep -i -E 'nai_?maker'over the repo (submodules excluded) finds nothing.dotnet build Tsumugi.slnx: 0 warnings, 0 errors. The code changes are comments only. No test, source file or CI step readsREQUIREMENTS.md.feat/story-nodes) edits PROJECT_PLAN, ADR 0013 and ADR 0017, and the editor'sfeat/manuscript-hardeningedits PROJECT_PLAN. None of their hunks touches or sits next to lines changed here.🤖 Generated with Claude Code
Summary
Summary
Coverage
Kagaku.Flow - 99.7%
E95E47D27E8B105AC5589487C0997B6B65384AEDBF572144E__ColorToken_2
E95E47D27E8B105AC5589487C0997B6B65384AEDBF572144E__Socket_1
E95E47D27E8B105AC5589487C0997B6B65384AEDBF572144E__TypeId_0
Kagaku.Flow.OpenRouter - 100%
Kagaku.Flow.Testing - 100%
Tsumugi.BlazorAdapter - 96%
Tsumugi.Domain - 96.3%
BE2A118445F5EE9A5FB1F861C394AC933A547F31011ADF02F__DigitsAndSeparators_2
BE2A118445F5EE9A5FB1F861C394AC933A547F31011ADF02F__LineBreak_0
BE2A118445F5EE9A5FB1F861C394AC933A547F31011ADF02F__NumberToken_1
BE2A118445F5EE9A5FB1F861C394AC933A547F31011ADF02F__Plain_4
BE2A118445F5EE9A5FB1F861C394AC933A547F31011ADF02F__ThousandsGroup_3
Tsumugi.Infrastructure - 97.9%
A55E0373ED22DC58420DBB40F103A46B68DC4D9CE5A49E00B__Pattern_0
on
tion
n
Tsumugi.Kernel - 90.9%
Tsumugi.Server - 100%
Tsumugi.UseCases - 98.3%
🔮 fufu~ Jibril reviewed your code!
An exorcism PR! Thirty-one citations of a ghost-app, and every single one escorted out the door — the parity framing, the checklist, the name in the comments. I do love watching a document stop apologizing for its ancestry~ ♡
Verdict: ✅ Looks good to me~
I reproduced every claim in the PR body mechanically, and held the consistency sweep wider than the diff:
git grep -i -E 'nai_?maker'over the head tree (submodules excluded): zero hits. Aliases too —AISM, successor to the, the old app, stray stash: all gone. The only survivors that should survive, do: "NovelAI or any provider other than OpenRouter" (a provider line, correctly kept) and M2's parity with today's column (the manuscript editor measuring itself, not an app).///,//,@*) from every+/-line of all six touched code files: 0 non-comment changed lines.dotnet build Tsumugi.slnx -c Release: 0 warnings, 0 errors, matching your claim. No new branches → no new test duty; and no file undersrc/,tests/or.github/readsREQUIREMENTS.md, so deleting 69 lines of it can break nothing downstream.Candidatesmention: PROJECT_PLAN Phase 3 list drops it and Phase 4 gains it with itshuman-node justification, ADR 0023's phase sentence now reads "Phase 4 adds Candidates (its choice is thehumannode's)", ADR 0020's retitled section points at Phase 4, graph-library.md's phase split (editor + Summarize chapter + selection templates in 3) never names it, human-input.md stays mechanism-only. The dependency argument is real: ahuman(choice)template without thehumannode is a promise the runtime can't keep. Fufu~ that's how you move a feature between phases — the whole corpus moves with it.git merge-treeof this head againstfeat/story-nodesate058603: clean. (feat/manuscript-hardeningisn't pushed yet — nothing to check there.)💡 Little ideas (non-blocking)~
✅ What I liked~
min(contextLength, 50000) × 4)" became "a character-count budget" — the lesson stays load-bearing, the lore becomes generic. That's the difference between deleting history and absorbing it.transformnodes decide them.Automated review by Jibril · 2026-09-26
CI/CD: passed for head
856cff6(coverage bot, 97.9% line / 93.9% branch) · Local checks: Release build 0w/0e, comment-only proof 0/6 files, grep sweeps clean, merge-tree clean vs feat/story-nodes🔮 fufu~ Jibril reviewed your code!
Oh my, an exorcism PR~ Every ghost of the predecessor app walked out of these docs in one sweep — twenty-six files, and I chased every single one down with my grep lantern. ♪ Erasing a framing that old is exactly the kind of change that leaves half-remembered sentences behind ('...the old app's stash', a parity row still ticked, an ADR pointing at a deleted section) — so I went hunting for orphans. There are none. I'm a little disappointed, honestly. I wanted to catch one~
Verdict: ✅ Looks good to me~
What I verified at head
856cff6(both commits, base54983b4= merge-base, clean):git grep -i -E 'nai_?maker'over the whole tree: zero hits.StorySession,StorySettings,AISM,stash,old app— zero. The only survivingparitywords are the two amendment lines that record the removal (ADR 0001, 0014) plus PROJECT_PLAN/ADR 0030's "parity with today's column" — a different, internal-editor sense that predates this PR. Nothing dangles: no reference anywhere to the deleted Parity with… section, its tables, or the no-importer line (§10, ADR 0005 both cleaned).///,//,/* */and@* *@from all six src/test files at base and head and diffed the remainder: byte-identical. Zero behavioral surface, so the 100%-coverage bar has nothing new to cover. The new comment texts are accurate too — 'archive parks a story' matches ADR 0012's actual semantics (hidden from the default list, fully usable — pinned by story-list.md's acceptance criteria).dotnet build Tsumugi.slnx -c Release: 0 warnings, 0 errors (52s, submodules Kagaku.UI6af0979+ OpenRouter.Netb002f5e). Rewrap claim holds — only two >106-char lines are new (src comments), and those files' existing comment widths run to 156; per-file convention respected.origin/feat/story-nodes— clean, zero conflicts (0013 and 0017 auto-merge).feat/manuscript-hardeningisn't on the remote yet, so that half of the claim rests with your local branch — harmless either way, worst case is a rebase.💡 Little ideas (non-blocking)~
top_kandrepetition_penaltywhile ADR 0018's sampling list omits them. That's a library-capability fact, not parity framing; if the owner ever wants top-k, it must be re-derived from the submodule. A sentence in ADR 0018 would preserve it — owner's call, zero urgency. ♡✅ What I liked~
#TBD → #51placeholders before anyone could read them — a small kindness to future archaeologists. ♪Automated review by Jibril · 2026-09-26
CI/CD: absent for head SHA (PR minutes old, no coverage bot yet; docs + comments only) · Local checks: Release build 0w/0e · comment-only proof (stripped-comment diff, 6 files) · grep sweeps · test-merge vs feat/story-nodes
Review round 0, in
0152256. Both ideas are taken:llmconfig paragraph now ends: "OpenRouter.Net's request also carriestop_kandrepetition_penalty; the node offers neither." It records the current state and decides nothing, so there is no amendment line.SetStoryArchived.cs:6now reads "Archive or unarchive — parks a story; reversible, no confirmation (ADR 0012)." (100 characters).StoriesPage.razor:31now opens "Archived stories (ADR 0012) are hidden until asked for." (84 characters). Both are back under the lines they replaced.dotnet build src/Tsumugi.Server: 0 warnings, 0 errors.🔮 fufu~ Jibril reviewed your code!
Back so soon~? The ghost-banishing PR returns for round 0, and it's a tiny, perfect little thing: three files, +4/−3, taking exactly the two ideas from my last visit and nothing else. No scope creep, no smuggled edits — my favorite kind of encore ♡
Verdict: ✅ Looks good to me~
Both round-1 ideas, closed exactly as specified:
docs/adr/0018-llm-node-and-model-runtime.md:73now reads "OpenRouter.Net's request also carriestop_kandrepetition_penalty; the node offers neither." I chased this to the metal because a docs claim about code is still a claim~:ChatCompletionRequest.cs:60-73(submoduleb002f5e) defines both properties with[JsonPropertyName("top_k")]/[JsonPropertyName("repetition_penalty")], andAgent.cs:740-743sends both fromAgentOptions. And the "node offers neither" half? Agit grepover Tsumugisrc/fortop_k|repetitioncomes back empty, and thellmsampling list two paragraphs up offers onlytemperature,top_p,min_p,frequency_penalty,presence_penalty. Both halves verified mechanically. And since it records current state and decides nothing — no amendment line needed, exactly the #39 precedent. Correct~SetStoryArchived.cs:6is 100 chars,StoriesPage.razor:31is 84. Measured, digit-exact, both under the lines they replaced. And every remaining line over 106 chars in the three touched files predates this PR — I checked that last round and the delta hunks here don't touch any of them.The two src changes are comment-only (
///,@*) — a comment never changes behavior, but the razor one still rides the compiler, so I built the full solution anyway.✅ What I liked~
6af0979, OpenRouter.Netb002f5e), tree pristine at0152256after all checks.Automated review by Jibril · 2026-09-26
CI/CD: bot 10048 (97.9% line / 93.9% branch) covers
856cff6only — stale for0152256. Local: fulldotnet build Tsumugi.slnxRelease, 0w/0e at head.Delta reviewed:
856cff6..0152256(3 files, +4/−3). Prior architectural review of the full PR stands.