docs: the nai_maker parity checklist ADR 0001 promises, and the ticks the code has earned #39
Loading…
Add table
Add a link
Reference in a new issue
No description provided.
Delete branch "docs/requirements-parity"
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
ADR 0001 says the
nai_makerparity checklist is "recorded inREQUIREMENTS.md". It never was: the file namednai_makeronly for the stash and the missing importer. PROJECT_PLAN titles Phase 2 "→nai_makerparity", yet there was no list to check that against. The tracker also lagged the code in four places. Docs only; no behaviour changes.nai_maker, a new section at the end ofREQUIREMENTS.md. It covers every story setting and feature ofnai_maker: itsStorySessionandStorySettings, the Flutter client's story screens, and the three provider paths. It is read from nai_maker's code, not its settings entity alone.nearEnd(distance), counted in segments and honoured on every provider (nai_maker's NovelAI path alone used it; ADR 0017);humanchoice card (ADR 0020, 0023), with the phases from PROJECT_PLAN;llmnode (ADR 0018);max_tokens;non the request; a second generate while one runs (one live run per story, ADR 0016); user accounts; an importer.***; an entry's own trigger words; top-k; repetition penalty; min length; bracket ban; the cut to the last full sentence; insertions; regex output replacements; the per-story stash. nai_maker never built the last three: their sections were placeholders, the client always sent none, and the server read none.11b466e:UpdateStory,SetStoryArchived,TrashStory,RestoreStory,PurgeStory, the stories page and the trash page).CreateChapter,RenameChapter,ReorderChapters,MergeChapters, wired through the New chapter button, the chapter header and the chapter menu). The line is split from the run's target-chapter half, which stays open.95bef53).Questions for you, raised by writing the table. None of them is decided here:
nai_makerparity", and ADR 0014 says the Continue graph reaches parity. But N candidates with Select arrive only in Phase 3 (the Candidates template) and Phase 4 (thehumanchoice card).mapof three writers → ahumanchoice →commit) in Phase 3. PROJECT_PLAN puts thehumannode in Phase 4, and its Phase 3 ships-when commits the best "from the inspector". The table follows PROJECT_PLAN.Docs:
docs/REQUIREMENTS.md,docs/adr/0001-docs-first-mirror-orihon-reuse-two-libraries.md. PROJECT_PLAN, AGENTS.md and the other ADRs are untouched.Verification
REQUIREMENTS.mdor links to its anchors (grep overtests/,src/anddocs/).nai_makerclaim was checked in its code at~/AISM/nai_maker, read-only:<think>.LlmSamplingomit top-k and repetition penalty, while OpenRouter.Net'sChatCompletionRequestcarriestop_kandrepetition_penalty.DeleteSession. It now reads "never wired".***: the lines land after the break's first line, style before tags. The row now says so.LinesandCharactersas kinds of text.🤖 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 - 96.4%
Tsumugi.Domain - 96%
BE2A118445F5EE9A5FB1F861C394AC933A547F31011ADF02F__DigitsAndSeparators_2
BE2A118445F5EE9A5FB1F861C394AC933A547F31011ADF02F__LineBreak_0
BE2A118445F5EE9A5FB1F861C394AC933A547F31011ADF02F__NumberToken_1
BE2A118445F5EE9A5FB1F861C394AC933A547F31011ADF02F__Plain_4
BE2A118445F5EE9A5FB1F861C394AC933A547F31011ADF02F__ThousandsGroup_3
Tsumugi.Infrastructure - 97.7%
A55E0373ED22DC58420DBB40F103A46B68DC4D9CE5A49E00B__Pattern_0
on
tion
n
Tsumugi.Kernel - 90.9%
Tsumugi.Server - 100%
Tsumugi.UseCases - 98.3%
🔮 fufu~ Jibril reviewed your code!
Oh? Oh! A checklist PR~ ADR 0001 has been promising this parity table since before the first line of C# existed, and now the debt comes due — with ticks, no less! I do love a table that names its ADRs. I read every row, chased every tick into the wiring, and parsed both tables with my own little parser. ♪
Verdict: ✅ Looks good to me~
Every checkable claim verified at head
32cb70b:UpdateStory/SetStoryArchived/TrashStory/RestoreStory/PurgeStoryall exist as use cases and are wired — StoriesPage's archive/unarchive menu, the trash page's Restore, and Purge behind aConfirmDialogwithButtonVariant.Danger(ADR 0022's confirmation, honored). §2: create/rename/reorder/merge all dispatch through the New chapter button, the chapter header's rename, and the ChapterMenu's move/merge (Move →MoveChapterRequested→ReorderChapters.ExecuteAsync); the Chapters page lists them. §3:FieldKind.Image+ ADR 0029, and "seen by no model by default" matches ADR 0017's PR #30 amendment word for word. §9: the Appearance tab with its ThemeSwitcher is real; ADR 0009 is the right ref, and settings.md's story already claimed Theme done — the file catches up.ChatCompletionRequestat the pinned submodule (15a1f28) carriestop_kandrepetition_penalty— the Open rows state a real gap. ADR 0017 really does sayheader | nearEnd(distance)and really doesn't specify how a keyword names the document it wakes. The Candidates Phase 3/4 split follows PROJECT_PLAN ("commit the best from the inspector" ships Phase 3; thehumannode is Phase 4), and the conflict with ADR 0023 is raised in the body, not buried. "One live run per story" matches §5's own gloss (the story is the engine's owner — ADR 0016).REQUIREMENTS.mdanchors — the stories reference §numbers of unchanged sections; no breakage.95bef53precedent — I pulled that commit and checked: it body-edited 0001 without an amendment line too.One honest caveat, fufu~:
~/AISM/nai_makerdoesn't exist in my review sandbox, so I could not re-derive the nai_maker-side claims (line-distance-on-NovelAI-only, the sentence cut, the never-wired insertions/filters/stash…) from its code. The Tsumugi side is fully verified; the nai_maker side is internally consistent with ADR 0001's own description ("lorebook triggers honoured on one provider only, the story as a flat string") and with PROJECT_PLAN. I'm taking your read-only pass on faith here — and the seven precision fixes you listed read like someone who actually opened the files. ♡💡 Little ideas (non-blocking)~
95bef53, this one). Defensible as "pointer only, no decision change," but if the owner wants uniform amendment hygiene across all 31 ADRs, a one-liner would settle it. Owner's call~✅ What I liked~
Automated review by Jibril · 2026-09-26
CI/CD: absent for head SHA (PR just opened, no coverage bot yet; docs-only) · Local checks: full diff + head-tree read, table parse, wiring grep sweeps in review clone