docs: Modellwahl nach Pruefbarkeit; stack-dev bricht an den Phasenwechseln fuer den Model-Switch (4.3.3)

Files changed:
- CHANGES.md
- VERSION
- instructions/claude-code-model-selection.md
- instructions/dev/stack-dev/SKILL.md

AGENTS.md

@@ -82,6 +82,7 @@ What a file is called says who it is for and how it is loaded. This is a rule, n
 | `instructions/<name>.md` | Agents | By link, or on explicit request |
 | `instructions/<name>/SKILL.md` | Agents | By the harness, once published |
 | `types/<name>.md` | Agents + validator | Via `tools/wikitool types describe`. Split by `root:`: a page type-spec (`root: kb`) belongs to the instance and ships as `.template`; one describing a stack artifact ships verbatim |
+| `docs/<name>.md` | Agents and humans | By link, or on explicit request - never automatically, and never as instruction |
 | `INDEX.md` | Both | Generated - never hand-edited |
 
 A stage may carry both a `README.md` and a `CONTRACT.md`: different readers, different
@@ -89,6 +90,16 @@ documents. What it may not carry is the same content twice - a README that resta
 contract is a second copy that drifts. `docs verify` enforces the specific case that already
 happened once: no README may hold a copy of the `wikitool` command table.
 
+**`docs/` carries no normative sentence.** It holds why the stack is built the way it is -
+background a session consults in passing, not a rule it must follow. Anything that would bind
+belongs in a `CONTRACT.md` instead, which is what keeps invariant 8 intact here: `docs/` is
+never a second place a rule could live, only prose about rules that live elsewhere. That is also
+why nothing verifies its content - there is no rule in it to check. It has no frontmatter, no type, no index, no lint, no decay, no provenance, and no
+`COLLECTION.md` - which [kb/CONTRACT.md § Collections](kb/CONTRACT.md#collections) forbids
+outside `kb/` anyway, but the point holds independently: `docs/` stays a plain directory of
+prose, invisible to everything `tools/wikitool` does except `dist export`, which copies it
+verbatim. A fresh instance needs the reasoning as much as this one does.
+
 ## Personalization
 
 `USER.md` and `SOUL.md` are read at session start, if the runtime has not already injected
@@ -145,7 +156,8 @@ input       schema + compiler     output   derived (gitignored)
                   work/   tracked scratch, deleted when the run closes
 ```
 
-Alongside it, not part of it: `instructions/` (what agents are told to do) and this file.
+Alongside it, not part of it: `instructions/` (what agents are told to do), `docs/` (why the
+stack is built the way it is - see [File naming](#file-naming)), and this file.
 
 **By stage** - read the contract for the stage you are writing in:
 
@@ -263,3 +275,10 @@ and `tools/README.md` are part of the change that introduced a stage, a command
 not follow-up work: nobody comes back for them, and a document that describes a repo which no
 longer exists is worse than none. The mechanical half - command tables, contracts, ignore
 canaries - is checked by `tools/wikitool docs verify`; the prose half is yours.
+
+`docs/` pages are held to a different clock than those three. A README goes stale on every new
+flag; a `docs/` page goes stale only when the reasoning it wrote down stops holding - a gate
+that stops living in code, an ownership line that moves, a boundary redrawn - which is rarer
+and not tied to any one commit. Nothing checks this by construction: a page there carries no
+normative sentence (see [File naming](#file-naming)), so there is no rule for `docs verify` to
+check, only a rationale for a session to notice has gone stale and to update or retire.

CHANGES.md

@@ -20,6 +20,220 @@ their date-only headings.
 
 ---
 
+## 4.3.3 - 2026-09-03 - Modellwahl nach Pruefbarkeit statt nach Aufgabenname; stack-dev bricht an den Phasenwechseln fuer den Model-Switch
+
+**Author:** Torben Nehmer
+
+`instructions/claude-code-model-selection.md` routete bisher nach Skill: eine Zeile "Stack
+development -> Opus/high" fuer alles, was `tools/`, `types/` oder `instructions/` anfasst. Das ist
+zu grob in beide Richtungen - es verteuert die lange, mechanische Mitte einer Stack-Sitzung, und es
+sagt nichts darueber, dass Anfang und Ende derselben Sitzung anders zu behandeln sind.
+
+**Die neue Achse ist "was faengt hier einen Fehler ab".** Wo ein Check in Code steht - `pytest`,
+`docs verify`, `instructions verify`, CI, die Gates - kostet der Fehler eines schwaecheren Modells
+eine Runde und faellt auf. Wo die einzige Durchsetzung eine Sitzung ist, die Prosa liest, faellt
+derselbe Fehler gar nicht auf: er wird ausgeliefert und bleibt stehen. Das ist dasselbe Argument,
+das `docs/why-gates-are-code.md` fuer Gates fuehrt, angewandt auf die Modellwahl.
+
+Stack-Entwicklung ist damit **nicht mehr eine Zeile, sondern drei**:
+
+| Phase | Was einen Fehler faengt | Modell |
+|---|---|---|
+| Design, Versionsstelle, Grenzuebertritts-Urteil | nichts | Opus/high |
+| Code, Tests, mechanische Doku-Synchronisation | pytest, CI, `docs verify` | Sonnet/high |
+| Issue-Abschluss, `docs/`-Veralterung, Changelog-Prosa | nichts, per Konstruktion | Opus/high |
+
+Die Mitte ist die lange Phase und die mit den Checks - dort liegt die Ersparnis. Die beiden
+Raender sind kurz (Minuten, nicht Stunden), haben aber keinen maschinellen Waechter: `wikitool`
+kennt den Issue-Tracker bewusst nicht, und eine `docs/`-Seite traegt keinen normativen Satz, also
+gibt es dort nichts zu verifizieren. Sie oben zu lassen ist billig und schuetzt genau die Arbeit,
+die still scheitert.
+
+Zwei Praezisierungen dazu: **Effort ist der billigere Hebel als das Modell** - `medium` steht fuer
+Stack-Arbeit bewusst in keiner Zeile, weil Mehrdatei-Konsistenz das ist, was ein reduzierter
+Effort zuerst aufgibt; `high` ist die Untergrenze, sobald mehr als eine Datei oder ein Contract
+betroffen ist. Und die Asymmetrie ist benannt: eine unnoetige Opus-Phase kostet einmal Geld, eine
+ungepruefte Sonnet-Phase kann etwas ausliefern, das nie wieder jemand ansieht.
+
+**Damit die Tabelle ueberhaupt wirksam wird, braucht sie Haltepunkte.** Eine Sitzung kann ihr
+eigenes Modell nicht wechseln - das ist `/model` und gehoert dem Nutzer. Eine Empfehlung, die
+niemand zum richtigen Zeitpunkt ausspricht, aendert nichts. `instructions/dev/stack-dev/SKILL.md`
+bekommt deshalb zwei ausdrueckliche Breaks:
+
+- **Neuer Schritt 3** - "Settle the design before building", mit dem Angebot zum Wechsel nach
+  unten, sobald der Plan steht und die Arbeit mechanisch wird. Einmal aussprechen, dann so oder
+  so weiterarbeiten.
+- **Schritt 6 (Abschluss) bricht in die Gegenrichtung** - ab dort greift wieder kein Check. Mit
+  der ausdruecklichen Auflage, die Arbeit **unabhaengig von der Antwort** zu tun: nach dem Publish
+  auf einen Modellwechsel zu blockieren wuerde genau den Zustand hinterlassen, den Schritt 6
+  verhindern soll. Lief die Phase auf dem billigeren Modell, gehoert das in die Uebergabe statt
+  ins Schweigen.
+
+Ein auftauchender Grenzuebertritt ist unter den Decision points ebenfalls als Anlass zum Wechsel
+nach oben benannt: `docs verify` prueft, dass ein Uebertritt sich dokumentiert, nie dass die
+Stelle richtig gewaehlt war.
+
+Die uebrigen Schritte sind unveraendert und nur umnummeriert (alt 3-5 -> neu 4-6).
+
+---
+
+## 4.3.2 - 2026-09-03 - stack-dev: Issue-Abschluss ist ein nummerierter Schritt, kein Zeiger in einer Routing-Liste
+
+**Author:** Torben Nehmer
+
+Nachfassen zu 4.1.2 (#44), das die Regel geschaerft, aber den Weg zu ihr nicht geaendert hat.
+`instructions/dev/issue-tracking.md` bekam damals Schritt 7 ("Closing is the last body update,
+not a comment"); `instructions/dev/stack-dev/SKILL.md` bekam nur eine umformulierte Zeile in
+seiner Routing-Liste. Eine Stunde spaeter schloss #45 auf exakt dieselbe Weise: gruendlicher
+Abschlusskommentar ueber einem Body mit unangehakten Kriterien.
+
+**Die Ursache lag nicht am Text der Regel, sondern an ihrer Erreichbarkeit.** Die nummerierten
+Schritte des Skills endeten bei "Verify before publishing". Ein Issue zu schliessen war ueberhaupt
+kein Schritt - es hing an einem Zeiger *innerhalb* von Schritt 2, und Schritt 2 ist eine
+Routing-Tabelle aus fuenf "read X before Y"-Eintraegen, keine Checkliste. Eine Sitzung folgt dem
+Spine, den sie im Kontext hat; was nur hinter einem Link steht, wird genau in dem Moment nicht
+aufgeschlagen, in dem es greift - am Ende einer langen Sitzung, wenn der Kontext am vollsten und
+die verbleibende Instruktionsflaeche am duennsten ist.
+
+Verschaerfend arbeitete der Blurb gegen seine eigene Regel: fett gesetzt war "keep it current as
+the state moves, **not at the end**". Wer den Body unterwegs ungefaehr gepflegt hatte, las daraus
+Konformitaet - der eigentliche Abschlusstest stand nur in der verlinkten Datei.
+
+Geaendert:
+
+- **Neuer Schritt 5 in `stack-dev/SKILL.md`** - "Close the issue with a body rewrite, not a
+  comment", mit dem Test inline (Kriterien abgehakt oder mit Begruendung gestrichen,
+  Entscheidungen als entschieden formuliert, kein Praesens ueber einen behobenen Defekt,
+  Verifikation benannt) und dem Verweis auf Schritt 7 fuer die volle Form. Damit steht der
+  Abschluss auf dem Spine.
+- **Schritt-2-Blurb rebalanciert** - beide Haelften binden jetzt sichtbar: fortlaufende Pflege
+  *und* der Rewrite vor dem Schliessen, mit Verweis auf Schritt 5.
+
+Nichts davon ist maschinell pruefbar, und das bleibt richtig so: `wikitool` kennt den Tracker
+nicht und darf ihn nicht lernen, weil es an Instanzen ausliefert, die kein Board haben
+(`issue-tracking.md` § "What no tool checks"). Der Skill-Spine ist die einzige Durchsetzung, die
+es geben kann - was der Grund ist, den Schritt zu nummerieren statt ihn zu verlinken.
+
+Verallgemeinerbar: eine Regel, die in eine verlinkte Instruction geschrieben wird, erreicht
+Sitzungen nur, wenn die nummerierten Schritte des zustaendigen Skills sie in dem Moment
+ansteuern, in dem sie greift.
+
+---
+
+## 4.3.1 - 2026-09-03 - docs/ befuellt - Stack-Hintergrund fuer vier Themen, Pflegeklausel in AGENTS.md ergaenzt
+
+**Author:** Torben Nehmer
+
+Gitea #45: die von #38 angelegte, bis dahin leere `docs/` bekommt ihre ersten vier Seiten - frisch
+geschrieben, nicht durch Umzug aus `kb/` befuellt, jede ohne normativen Satz und mit Verweis auf
+das bindende Dokument statt einer Wiederholung seiner Regeln:
+
+- `docs/pipeline-rationale.md` - warum `raw -> types/tools -> kb -> reports` vier getrennte Stufen
+  sind und was "never re-derive, always compile" praktisch bedeutet
+- `docs/why-gates-are-code.md` - warum Mass-Update-, Publish-Remote- und Iteration-Budget-Gate in
+  `tools/wikitool` statt in einer Instruktion stehen
+- `docs/ownership-and-templates.md` - der Unterschied zwischen stack-eigenen, verbatim
+  ausgelieferten Dateien und instanz-eigenen `.template`-Dateien
+- `docs/version-model.md` - warum Drop-in-Kompatibilitaet und Migrationsbedarf zwei unabhaengige
+  Fragen sind, illustriert an der 2.0.0-Fallstudie
+
+**AGENTS.md § Changelog:** neue Klausel zur Pflege von `docs/`, ergaenzt neben der bestehenden
+Regel zu `README.md`/`EVALS.md`/`tools/README.md`. Eine `docs/`-Seite veraltet nicht wie ein
+README bei jedem neuen Flag, sondern nur, wenn die aufgeschriebene Begruendung selbst nicht mehr
+traegt - per Konstruktion ungeprueft, da die Seite keinen normativen Satz enthaelt, den
+`docs verify` pruefen koennte.
+
+---
+
+## 4.3.0 - 2026-09-03 - docs/ als ausgelieferter Hintergrund-Ort; Decision-Seiten bleiben in kb/, Decay-Skip fuer concept_type: decision
+
+**Author:** Torben Nehmer
+
+Gitea #38: `dist export` lieferte bislang keine einzige `kb/`-Seite aus - eine frische Instanz
+bekam den Stack, aber keinen Grund für seine Form. Die dokumentierte `adr-NNN-`-Konvention in
+`kb/concepts/COLLECTION.md` existierte zudem nur auf Papier: keine der sieben
+`concept_type: decision`-Seiten folgte ihr, und `confidence_decay()` lief bedingungslos über sie
+- ein Kategorienfehler, weil Zeitablauf eine Entscheidung nicht falscher macht, nur Supersession
+tut das.
+
+**Neu:** `docs/` - ein inertes Verzeichnis für Stack-Hintergrund (warum der Stack so gebaut ist,
+nicht was diese Instanz entschieden hat). Keine Frontmatter, kein Typ, kein Index, kein Lint,
+keine Decay, keine Provenance, keine `COLLECTION.md`. `dist export` liefert es verbatim aus, wie
+`instructions/` und `types/`. Befüllung folgt in Gitea #45.
+
+**Verworfen, nach Prüfung:** ein Umzug der sieben Decision-Seiten nach `decisions/`. Der
+Subtyp-Floor aus #28 verlangt mindestens eine Seite je deklariertem `concept_type`, und ein
+Umzug hätte `decision` auf null gebracht; dazu zeigen 89 Wikilinks aus `kb/` sowie
+tool-eigene Frontmatter-Arrays auf die sieben, und `links.py`/`xref add` kennen kein Ziel
+außerhalb `kb/`. Die sieben bleiben in `kb/concepts/`, ebenso ein zweiter, separat erwogener
+Rename (`docs verify` → `parity verify`) - der wäre nur nötig gewesen, wenn ein Befehl auf das
+Verzeichnis `docs/` wirkt, und keiner tut das.
+
+**Geändert:**
+- `confidence_decay()` überspringt `concept_type: decision` strukturell (kategorische Ausnahme,
+  nicht als Brücke gebaut - Begründung im Docstring).
+- `kb/concepts/COLLECTION.md` § Decisions ersetzt die tote ADR-Vorlage durch die real gelebte
+  Form: eine Entscheidung ist eine gewöhnliche Concept-Seite, organische Prosa, kein
+  `adr-NNN-`-Präfix, `**Status:**` optional, Supersession per `supersedes`-Link.
+- `kb/CONVENTIONS.md` § Naming und `instructions/kb-profiles.md` (Profil `german`) korrigiert -
+  beide dokumentierten noch die verworfene `adr-NNN-`-Namensregel.
+- `AGENTS.md` § File naming und § Routing: `docs/`-Zeile, plus die Regel, dass `docs/` keinen
+  normativen Satz trägt (das hält Invariante 8 heil - was binden würde, gehört in einen
+  Contract).
+- `tools/CONTRACT.md`: Klarstellung, dass `docs verify` Dokumentations-Parität prüft, nicht das
+  `docs/`-Verzeichnis, sowie `docs/` in der `dist export`-Zeile ergänzt.
+
+Additiv und in beide Richtungen drop-in: eine bestehende Instanz ohne `docs/` exportiert
+weiterhin identisch (leerer `_copy_tree`-Treffer), eine Instanz mit `docs/` bekommt es ab jetzt
+mitgeliefert. Kein Feld, kein Kommando ändert sein Verhalten für bestehenden Inhalt.
+
+**Migration:** none required.
+
+Berührt: `tools/chemenu/commands/confidence_decay.py`, `tools/chemenu/commands/dist_cmd.py`,
+`tools/chemenu/tests/test_confidence_decay.py`, `tools/chemenu/tests/test_dist_cmd.py`,
+`kb/concepts/COLLECTION.md`, `kb/CONVENTIONS.md`, `instructions/kb-profiles.md`, `AGENTS.md`,
+`tools/CONTRACT.md`.
+
+---
+
+## 4.2.0 - 2026-09-03 - Korpus-Kuratierungsrichtlinie: Untergrenzen und Leitplanke für reaktive Fixes
+
+**Author:** Torben Nehmer
+
+Ein Demo-Korpus will klein und stabil sein, ein Testbett groß, unordentlich und in Bewegung -
+dieses Repo verlangt seit der Veröffentlichung beides vom selben `kb/` (Gitea #28). Die Sitzung
+vom 2026-09-02 hatte Fixture, `--with-demo` und ein zweites Repo bereits verworfen; offen blieb
+nur, wie kuratiert "kuratiert genug" heißt und welche Leitplanke reaktive Fixes bekommen.
+
+**Neu:** `instructions/dev/corpus-policy.md`. Fünf Untergrenzen, jede mit einer bestehenden
+`wikitool`-Prüfung messbar, keine davon durch neuen Tool-Code: jeder Seitentyp und jeder
+deklarierte Subtyp mit mindestens einer Seite, mindestens fünf Seiten mit mindestens drei
+Quellen, ein bis zehn Orphan-Seiten, im Schnitt mindestens vier ausgehende Wikilinks pro Seite.
+Gemessen am 2026-09-03: 181 Seiten, alle Typ-/Subtyp-Floors erfüllt, 12 Seiten mit ≥3 Quellen, 3
+Orphans, Ø 6,2 ausgehende Links - der Korpus war bereits groß genug, ohne dass eine einzige
+Seite eigens dafür angelegt werden musste. Eine Untergrenze wird nie durch eine erfundene Seite
+gefüllt, sondern durch eine echte Quelle beim nächsten passenden Ingest - Invariante 3 gilt
+unverändert.
+
+**Die Leitplanke für reaktive Fixes** unterscheidet drei Stufen: punktuelle Änderungen (immer
+erlaubt, gewöhnliche Arbeit), korpusweite Änderungen (nur geplant, mit eigenem Issue und
+`work/`-Run - trifft eine Session das Mass-Update-Gate während sie etwas anderes tat, holt sie
+sich nicht den `--confirm`-Token, sondern stoppt und legt ein Issue an) und reaktive Eingriffe
+in Korpusinhalt, um einen Test grün zu machen oder einen Tool-Bug zu umgehen (nie erlaubt,
+Invariante 7). Das Verhältnis zu `kb_dir`/`raw_dir` und `test_pipeline_l0.py` bleibt wie im
+ursprünglichen Befund: kleiner, isolierter Fall in der Fixture, großer, vernetzter Fall in
+`kb/` - keine Fixture-Extraktion aus dem Korpus.
+
+Dev-only und rein additiv - kein Feld, kein Kommando, keine Datei außerhalb von
+`instructions/dev/` ändert sich, daher `--minor` ohne `--breaking`.
+
+**Migration:** none required.
+
+Berührt: `instructions/dev/corpus-policy.md` (neu),
+`instructions/dev/stack-dev/SKILL.md` (Schritt 2, Routing-Zeile).
+
+---
+
 ## 4.1.2 - 2026-09-03 - Issue-Abschluss ist ein Body-Rewrite, nicht nur ein Kommentar
 
 **Author:** Torben Nehmer

VERSION

@@ -1 +1 @@
-4.1.2
+4.3.3

docs/ownership-and-templates.md

@@ -0,0 +1,71 @@
+# Ownership and Templates
+
+Chemenu ships two kinds of files side by side, and at a glance they look the same: both are
+plain markdown, both sit in the repo root or under `kb/`, both get read at session start. But a
+stack upgrade treats them completely differently. Some - [AGENTS.md](../AGENTS.md),
+[kb/CONTRACT.md](../kb/CONTRACT.md), the per-stage contracts - are identical in every instance
+that runs this stack and can simply be overwritten by the next release. Others - `USER.md`,
+`SOUL.md`, `kb/CONVENTIONS.md`, `ENVIRONMENT.md` - describe one particular instance, and
+overwriting them would silently erase a choice someone made on purpose.
+
+## Two different kinds of truth
+
+The stack-owned files describe how the tool works. `kb/CONTRACT.md` opens by saying it holds
+what `tools/wikitool` enforces or what follows mechanically from how it operates - see
+[kb/CONTRACT.md](../kb/CONTRACT.md), lines 10-13. That kind of statement doesn't vary by
+instance: the compiler behaves the same way regardless of who is running it, so the sentence
+describing that behavior can be copied byte-for-byte into every checkout without becoming
+wrong anywhere.
+
+The instance-owned files describe a choice: which language pages are written in, what tone the
+agent takes, who the operator is, which git remote is authoritative, which MCP servers are
+reachable. None of that follows from the tool's mechanics - two instances of the identical
+stack can answer all of these differently and both be correct. [AGENTS.md § Personalization](../AGENTS.md#personalization)
+frames the split the same way for `kb/CONTRACT.md` versus `kb/CONVENTIONS.md`: "the split is by
+who may change the sentence, not by what it is about." A rule about page structure could in
+principle have been written per-instance too, but then every instance answering "not German" to
+setup would be hand-editing a file the stack also ships, and the next `dist export` merge would
+hand the instance's own file back to it, discarding the customization.
+
+## Why silent overwrite is the failure being designed against
+
+A stack update is meant to be a routine, low-risk operation: pull the latest release, get
+whatever fixes and features shipped since the last one. That only stays low-risk if the update
+knows which files it's allowed to touch. If `USER.md` or `kb/CONVENTIONS.md` were treated the
+same as `AGENTS.md` - shipped and periodically re-copied - an upgrade would quietly replace a
+description of *this* operator, in *this* language, with whatever placeholder or default the
+stack maintainers wrote. The damage wouldn't be loud: nothing crashes, the files still parse,
+the agent just starts acting on the wrong premises until someone notices the voice or the
+language changed.
+
+Keeping the boundary at the file level, rather than trying to merge changes within a shared
+file, means an upgrade never has to guess which lines are "stack" and which are "instance" -
+the file itself already answers that.
+
+## Why a `.template`, not just an absent file
+
+The mechanism for instance-owned content is a `.template` file the distribution ships instead
+of the real one - `USER.md.template`, `SOUL.md.template`, `kb/CONVENTIONS.md.template`,
+`ENVIRONMENT.md.template`. An alternative would have been to ship nothing at all and let a
+brand-new instance start from a blank page. The template exists because a blank page doesn't
+tell [instructions/setup-instance.md](../instructions/setup-instance.md) what shape the answer
+should take, and it gives nothing for a validator to check afterward.
+
+A template carries a placeholder value - a sentinel - in the fields that need a real answer.
+Setup interviews the operator and replaces the sentinel with what they actually said. That
+gives `doctor` a mechanical way to tell "personalized" from "not yet": a file that still
+contains the sentinel hasn't been through setup, regardless of whether the file exists. That's
+also why `ENVIRONMENT.md` only warrants a WARN rather than a FAIL when absent - see
+[AGENTS.md § Environment](../AGENTS.md#environment) - while a missing or unfilled
+`USER.md`/`SOUL.md`/`kb/CONVENTIONS.md` is a harder failure: `ENVIRONMENT.md` describes one
+checkout among possibly several and is gitignored for that reason, so its absence is a normal
+state rather than a sign setup was skipped.
+
+## The consequence in practice
+
+Running a stack upgrade against an existing instance boils down to: overwrite the verbatim
+files, leave the `.template`-sourced files alone. The verbatim files are safe to replace
+wholesale because they were never instance-specific to begin with - identical content going
+back in changes nothing an instance actually decided. The template-sourced files were filled in
+once, by a person, for a reason, and nothing about a newer release of the stack's mechanics
+gives it standing to override that.

docs/pipeline-rationale.md

@@ -0,0 +1,64 @@
+# Why the pipeline has four stages
+
+Chemenu could, in principle, be one directory: drop a file in, ask a question, get an answer
+computed fresh each time. It isn't built that way. The pipeline in
+[AGENTS.md](../AGENTS.md#routing) - `raw/` -> `[types/ + tools/]` -> `kb/` -> `reports/`, with
+`work/` alongside rather than inside it - separates *material* from *meaning* from
+*byproduct*, and each seam exists because collapsing it costs something specific.
+
+## Why raw material stays untouched
+
+[raw/CONTRACT.md](../raw/CONTRACT.md) keeps a source exactly as it arrived. The reasoning is
+simple once stated: the moment someone "cleans up" or reformats a source on the way in, the
+thing later claims get checked against is no longer the thing that was actually said. An
+immutable `raw/` means a citation always resolves to the original, not to somebody's tidied
+memory of it. It also draws a trust boundary in one place instead of scattering it - everything
+past `raw/` can be treated as reviewed, because nothing upstream of it silently already was.
+
+## Why extraction happens once, through a schema
+
+[types/type-spec.md](../types/type-spec.md) is what stands between a raw file and a `kb/` page:
+a type-spec defines what a conforming instance of a page looks like, and the compiler
+(`tools/wikitool`) applies it. The alternative - every query re-reading and re-interpreting the
+source on demand - would mean paying the cost of understanding the material every single time,
+and getting a slightly different answer each time depending on how the question was phrased.
+Extracting once, against a fixed schema, turns "re-read and re-guess" into "look up what was
+already compiled." That is the "never re-derive, always compile" principle from
+[AGENTS.md](../AGENTS.md): understanding a source is expensive and worth doing exactly once,
+after which it becomes a cheap, stable lookup.
+
+## Why a `kb/` page has to stand on its own
+
+[kb/CONTRACT.md](../kb/CONTRACT.md) sets the bar for the compiled layer: a page should answer a
+future question without sending the reader back to the source it came from. That's the payoff
+of compiling in the first place - if every answer still bottomed out in "go re-read the raw
+file," the `kb/` layer would just be a pointer with extra steps, and the cost of extraction
+would have bought nothing. A page that stands alone is what makes the corpus fast and
+consistent to query: the work of understanding is already sitting there, done.
+
+## Why `reports/` doesn't need to be maintained
+
+[reports/CONTRACT.md](../reports/CONTRACT.md) treats most of what lands in `reports/` -
+lint output, telemetry traces - as disposable. The structural content of a lint report can be
+recomputed from the tree at any commit, so keeping an old copy around would just be a second
+version of something the tool can already answer on demand, and a second copy is exactly the
+kind of thing that quietly goes stale. Treating it as derived output rather than a fourth thing
+to maintain means there is nothing there to fall out of sync - regenerating it is cheaper than
+reconciling it. The one part that genuinely can't be recomputed - the judgment a pass produced -
+is carried out into `kb/` or `kb/log.md` before the report itself is discarded, which is the
+distinction between what's recomputable and what isn't.
+
+## Where `work/` fits
+
+[work/CONTRACT.md](../work/CONTRACT.md) describes a workshop, not a fifth pipeline stage: a
+place for the notes, extracts and open decisions of a task that spans more than one session, on
+its way toward becoming a `kb/` page. It sits beside the raw -> kb -> reports flow rather than
+inside it - closer in spirit to a desk than to a conveyor belt.
+
+## The shape this produces
+
+Four stages, each answering a different question: `raw/` - what was actually said; `types/` +
+`tools/` - how to turn that into structured understanding; `kb/` - what is now known;
+`reports/` - what a pass over the corpus noticed in passing. Keeping them separate is what lets
+each one be trusted for what it is, instead of every layer having to double as all four at
+once.

docs/version-model.md

@@ -0,0 +1,84 @@
+# Why the stack version splits compatibility from migration
+
+A stack version number looks like it answers one question. It actually answers two, and the two
+are independent of each other.
+
+## Two questions, not one
+
+The first question is whether the new version is a drop-in replacement for the old one - whether
+an existing instance can install it, and can also go back, without anyone doing hand-work. That
+is what a version number *is*: a promise. The second question is whether the existing corpus in
+`kb/` needs to change shape to keep working under the new version. These sound like the same
+question, because most of the time a change that breaks compatibility also happens to touch
+content, and most of the time a change that leaves content untouched also happens to be
+compatible. The correlation is real; it just is not a law. `instructions/dev/version-parts.md`
+carries the actual test for telling them apart and the steps that follow from it - this page is
+about why the split exists at all.
+
+## Why "kb/ untouched" is not proof of anything
+
+The tempting shortcut is: if no page in `kb/` had to change, the bump can't be that serious. This
+is exactly backwards for a class of changes that live entirely outside the corpus - a renamed
+release artefact, a Python import path, an environment variable, the URL an instance's own
+updater points at. None of those touch a single page. All of them can strand an existing
+instance just as thoroughly as a rewritten type-spec would. The corpus is the part of the stack
+that looks at itself; the compatibility question is about everything an instance depends on to
+keep functioning, most of which the corpus never sees.
+
+## Reading compatibility off the leftmost non-zero component
+
+Semantic versioning gives every component a job, but only one of them is where an existing
+instance's tooling actually looks to decide "is this safe." On a `2.x` stack that is MAJOR; on a
+still-pre-1.0 `0.x` stack, by the same convention, it's MINOR - the leftmost slot that isn't
+pinned to zero is the one an automated updater treats as the compatibility boundary. Bump
+anything to its left, or bump that slot itself, and the promise changes. Everything to the right
+of it can move as freely as the project likes without touching that promise. This is why the
+question "is it boundary-crossing" always resolves to one specific digit, not to a feeling about
+how big the change is.
+
+## Downgrade is half the promise
+
+It's natural to test compatibility by only asking "does the upgrade work." The other half -
+"can an instance that upgraded put the old version back and land where it started" - carries
+equal weight, and it's the half that's easy to forget because forward motion is what everyone is
+testing for anyway. A state file the old version can no longer parse, a generated index in a new
+shape, a stamp file that got renamed: none of these have to break the upgrade to break the
+downgrade. An instance that can go forward but not back has already lost the property a
+compatible version number is supposed to guarantee.
+
+## A promise made to a machine, not only to a person
+
+A human reading a changelog can absorb "this technically isn't compatible but it's fine, just
+update those two things by hand." An instance's own update mechanism cannot. It reads a version
+number, decides whether to pull the new release, and has no channel for nuance - which is exactly
+why the update path itself is one of the sharpest ways to cross the boundary invisibly: if the
+new version moves where updates come from, the very channel that would have told an instance to
+adjust is the channel that just broke. The version number isn't documentation aimed at a reader;
+it's an input consumed by code that has no other way to ask.
+
+## The 2.0.0 story
+
+This isn't hypothetical for this stack. The rebranding that produced Chemenu renamed the repo,
+the release artefact, and the Python package - and left every page in `kb/` untouched. The first
+instinct was a MINOR bump, on the reasoning that nothing in the corpus needed migrating. That
+reasoning was correct on its own terms and answered the wrong question. Three things broke
+underneath it: every existing instance's `update_url` pointed at a repo path that no longer
+existed and, because it's a machine-written file, couldn't be hand-repaired; the release artefact
+name changed, breaking every download script and pin against it; and the import name changed,
+breaking anything importing the package from outside the shipped tree. The corpus had nothing to
+say about any of this, because none of it lived in the corpus.
+
+What caught the mistake was a person looking at the diff and asking whether it really was a
+drop-in replacement, not a validator. No check in `docs verify` or anywhere else confirms that a
+version part was chosen correctly - it only confirms that a boundary-crossing bump documents
+what it breaks. The 2.0.0 entry in `CHANGES.md` carries the corrected reasoning in full, and the
+version bump that shipped it was `--major --no-migration`: boundary-crossing and untouched
+corpus, at the same time, which is precisely the combination the two-question split exists to
+make visible.
+
+## Where the procedure lives
+
+The drop-in test, the catalogue of changes that cross the boundary with no page touched, and the
+steps for a boundary-crossing bump - the `--breaking` line, the migration document or
+`--no-migration` reason, talking to the user before bumping - are one procedure, kept at one
+place: [instructions/dev/version-parts.md](../instructions/dev/version-parts.md).

docs/why-gates-are-code.md

@@ -0,0 +1,57 @@
+# Why gates are code
+
+Chemenu has three hard limits - the Mass-Update Gate, the Publish-Remote Gate, and the
+Iteration Budget Gate - and all three live inside `tools/wikitool`, not in a paragraph of
+instructions an agent reads and follows. The rules themselves, and what to do when one trips,
+are in [AGENTS.md § Gates](../AGENTS.md#gates) and [instructions/gates.md](../instructions/gates.md).
+This page is only about the design choice underneath them: why code, and why these three
+mechanisms in particular.
+
+## A suggestion an agent can talk itself past
+
+An instruction like "don't publish too much at once" or "don't loop forever" lives in the same
+place as every other piece of guidance a session is holding - alongside the task, the user's
+last message, and whatever context made the moment feel urgent. Under pressure, or with a
+plausible-sounding reason ("this batch is different, it's mechanical"), that guidance can be
+reasoned around without anyone deciding to break a rule. Nothing enforces it; it just competes
+for attention with everything else in the context window, and sometimes loses.
+
+A check compiled into the tool doesn't have that problem, because it isn't part of the
+conversation at all. It runs before the command dispatches, regardless of how convincing the
+case for skipping it seemed a moment earlier. The difference isn't that code is smarter than a
+well-written instruction - it's that code doesn't get talked into anything.
+
+## Why three different mechanisms, not one
+
+The three gates ask three different questions, and each one's shape follows from what kind of
+question it is.
+
+The Mass-Update Gate asks *is this change too large to publish unreviewed* - a judgment that
+varies changeset by changeset, so it clears with a `--confirm` token tied to the specific
+output the user just read. Approval is scoped to that one publish.
+
+The Publish-Remote Gate asks something underneath that: *is this even the right repository*.
+That's not a per-push judgment, it's a standing property of the checkout - true or false for
+every publish that checkout will ever attempt, not just this one. A confirm token would let an
+agent clear it once and then treat the answer as settled, which is exactly backwards for a
+question whose answer shouldn't move at all mid-session. The only way past it is the user
+editing `.wikitool-remotes.json` directly, outside the gate's own flow.
+
+The Iteration Budget Gate asks a third kind of question - not "is this instance correct" but
+"has this session stopped making progress." That's read from the shape of the call history
+itself (call count, repeated identical calls), not from anything about the content of any one
+call.
+
+## Numbers that come from measurement, not intuition
+
+The iteration ceiling didn't start where it sits now. It used to run 15-25, borrowed from a
+general rule of thumb, until four real ingest runs measured 24, 26, 29 and 30 calls apiece -
+every one of them an ordinary workflow doing nothing wrong, and every one of them at or past
+where the old ceiling would have refused it. A limit that the normal case keeps tripping stops
+functioning as a limit; it becomes background noise a session learns to route `--override-budget`
+around as a matter of course, and the whole point of a hard-coded check is that it isn't supposed
+to feel routine.
+
+That's the deeper reason these numbers live in a tool rather than in prose: prose is read once
+and remembered loosely, but a threshold enforced every call is tested by every call, and a
+threshold that fails its own test gets noticed and re-measured rather than quietly ignored.

instructions/claude-code-model-selection.md

@@ -6,10 +6,24 @@ description: Which Claude model and effort level to run a Claude Code session, a
 
 # Pick the Claude model and effort level for the task at hand
 
-Scale the model and effort to how much judgment the task actually needs. Running everything at
-the most capable model and highest effort is safe but wasteful: the gates in [gates.md](gates.md)
-are enforced in code, not by model judgment, so a weaker model cannot bypass them - it can only
-do a worse job of the calls the gates don't cover.
+Scale the model and effort to **what catches a mistake in this part of the work** - not to how
+important the task feels, and not to its name. Running everything at the most capable model and
+highest effort is safe but wasteful: the gates in [gates.md](gates.md) are enforced in code, not
+by model judgment, so a weaker model cannot bypass them - it can only do a worse job of the calls
+the gates don't cover.
+
+That last clause is the whole rule, turned into a test. Where a check lives in code - `pytest`,
+`docs verify`, `instructions verify`, CI, the gates - a weaker model's mistake surfaces and costs
+one more round. Where the only enforcement is a session reading prose, the same mistake does not
+surface at all: it ships, and it stays until someone happens to notice. The two are not the same
+risk, and they should not get the same model. This is the argument
+[docs/why-gates-are-code.md](../docs/why-gates-are-code.md) makes about gates, applied to who is
+holding the keyboard.
+
+Both directions cost something, which is why the axis matters rather than a blanket answer:
+over-provisioning is a standing cost paid every session, while under-provisioning in an unchecked
+phase is a silent error with a long tail. A corrective session, its bump, its CI runs and its
+release together cost more compute than the model difference they were saving.
 
 Claude-Code-only, and imported by CLAUDE.md rather than linked from AGENTS.md: the model names,
 the `/code-review` effort dial and the `Agent` tool's `model:` override have no equivalent in the
@@ -28,17 +42,41 @@ to *make*, not a setting to apply.
 
 ## Steps
 
-1. **Recommend the session's model and effort by the skill in use**, when asked or when the
-   mismatch is worth one sentence. Say it once and continue working either way - a session that
-   argues about its own model instead of doing the task has already cost more than the model
-   difference:
+1. **Recommend the session's model and effort by what catches a mistake in the phase it is in**,
+   when asked or when the mismatch is worth one sentence. Say it once and continue working either
+   way - a session that argues about its own model instead of doing the task has already cost
+   more than the model difference:
 
-   | Skill / task | Model | Effort |
-   |---|---|---|
-   | `wiki-status`, simple `wiki-query` lookups | Sonnet | default |
-   | `wiki-lint` | Sonnet | default |
-   | `wiki-ingest`, `wiki-manage`, judgment-heavy `wiki-query` | Sonnet | high |
-   | Stack development: `tools/`, `types/`, `instructions/` as code | Opus | high |
+   | Phase / task | What catches a mistake here | Model | Effort |
+   |---|---|---|---|
+   | `wiki-status`, simple `wiki-query` lookups | the answer is re-checkable against the corpus | Sonnet | default |
+   | `wiki-lint` | `lint` itself is the check | Sonnet | default |
+   | `wiki-ingest`, `wiki-manage`, judgment-heavy `wiki-query` | `lint` and `docs verify`, partly - the judgment about a claim is not covered | Sonnet | high |
+   | Stack dev: design, the version part, a boundary-crossing judgment | nothing - `docs verify` checks that a crossing documents itself, never that the part was right | Opus | high |
+   | Stack dev: code, tests, mechanical doc sync (command tables, contract rows) | `pytest`, `docs verify`, `instructions verify`, CI | Sonnet | high |
+   | Stack dev: closing an issue, `docs/` staleness, changelog prose | nothing, by construction - see below | Opus | high |
+
+   **Stack development is not one row**, which is the point of splitting it. The middle phase is
+   where the tokens are and where the checks are, so it is the phase worth running cheaper. The
+   two around it have no mechanical guard at all - a `docs/` page carries no normative sentence,
+   so there is nothing for `docs verify` to check ([AGENTS.md](../AGENTS.md) § File naming), and
+   the same holds for whatever tracker an instance keeps its open work in, which `wikitool`
+   deliberately knows nothing about. Those two phases are short - minutes, not hours - so keeping
+   them on the stronger model is cheap, and it protects the only work in the session that fails
+   silently.
+
+   **Effort is the cheaper lever than the model.** Reach for it first: `medium` deliberately does
+   not appear in this table for stack work, because multi-file consistency is exactly what a
+   reduced effort level gives up. Sonnet at `high` is the floor for anything touching more than
+   one file or a contract; `default` is for a single-file mechanical edit with a test behind it.
+
+   **A session cannot switch its own model**, so these rows only become real if someone offers the
+   switch at the moment the phase changes - once, without arguing about it, and never as a reason
+   to stop work that is already underway.
+<!-- dist:strip-start -->
+   In this repo those moments are named: the `stack-dev` skill breaks for them at its steps 3
+   (design settled, work turns mechanical) and 6 (publish done, the unchecked tail begins).
+<!-- dist:strip-end -->
 
 2. **Pick a spawned subagent's model by what it does**, via the `Agent` tool's `model:`
    parameter - the values are `haiku`, `sonnet`, `opus`, `fable`:
@@ -66,8 +104,14 @@ to *make*, not a setting to apply.
   mechanical one - `wikitool` carries the mechanical part regardless of which model is
   supervising it.
 - **Unsure which row applies?** Default to Sonnet at high effort, not the most capable model at
-  the highest effort. Under-provisioning costs one worse answer in one session; reflexively
-  over-provisioning is a standing cost paid every session.
+  the highest effort. Under-provisioning *where a check exists* costs one worse answer in one
+  session; reflexively over-provisioning is a standing cost paid every session.
+- **Unsure whether the phase is checked?** Treat it as unchecked. The asymmetry is not symmetric:
+  a needless Opus phase costs money once, an unchecked Sonnet phase can ship something nobody
+  looks at again.
+- **Mid-session and the phase changed, but nobody switched?** Do the work anyway - never block a
+  publish or an issue close on a model the session cannot change itself. Say which phase ran on
+  which model in the handover, so the gap is visible rather than silent.
 
 ## Scope
 

instructions/dev/corpus-policy.md

@@ -0,0 +1,94 @@
+---
+type: types/instruction.md
+name: corpus-policy
+description: What "curated enough" means for kb/ when it is demo and testbed at once, the measurable floors that define it, and what a reactive fix to the corpus may and may not do.
+---
+# Keep kb/ curated enough to develop against, without a second corpus
+
+This instance runs one `kb/` for two purposes at once: a public demo and the testbed this stack
+is developed against. There is deliberately no fixture corpus, no `--with-demo` export, and no
+second repository - see Gitea #28. The corpus's size and shape are set by what targeted
+development needs, not by a synthetic fixture size or a demo aesthetic.
+
+## When to run
+
+- Before judging whether the corpus can exercise a change under development - ranking, index
+  scaling, orphan detection, a new label, a new type-spec.
+- Before a reactive fix touches `kb/` content rather than the failing code - the floors below
+  are what decides whether the fix may proceed as-is.
+- Picking up Gitea #28 or #30, or any issue that references this file.
+
+## The floors
+
+Each is mechanically checkable with an existing `wikitool` command; none needs new tool code.
+A floor exists to keep some class of bug observable, not to describe an aesthetic target - so
+when a session is about to make one of these numbers *worse*, that is the signal to stop and
+think, not a number to defend for its own sake.
+
+| Floor | Check | Why this number |
+|---|---|---|
+| Every page type has ≥1 page | `wikitool search --field type=types/<t>.md` | A type with zero pages means its schema, its collection contract and its lint rules are unexercised |
+| Every declared subtype has ≥1 page | `wikitool search --field <x>_type=<v>` | Same reasoning, one level down - `entity_type`, `concept_type`, `source_type` |
+| ≥5 pages corpus-wide with ≥3 `sources:` entries | one-off script, see below | Provenance fan-in - multiple sources backing one claim - is a real case only a handful of pages exercise; fewer than 5 and a provenance-index bug can hide |
+| Orphan pages (no inbound link) between 1 and 10 | `wikitool lint` | Zero orphans makes orphan detection itself unobservable; more than 10 means the corpus stopped being curated |
+| Average outbound wikilinks per page ≥4 | one-off script, see below | Below this, ranking and graph-traversal work has too little structure to exercise |
+
+A floor is a lower bound only. There is no upper bound on page count or on any of these numbers
+except the orphan ceiling above - a corpus that outgrows these floors through real ingests is
+not a problem this file cares about.
+
+**Measured 2026-09-03** (see Gitea #28): 181 pages, 14/14 types and subtypes covered, 12 pages
+with ≥3 sources, 3 orphans, 6.2 average outbound links. All floors held without any manufactured
+content - the corpus was already big enough when the question was asked.
+
+A type or subtype sitting at exactly the floor - one page - shows no set-level bugs, only that
+the type is *reachable*. That is a soft target for the next `wiki-ingest` that happens to
+produce a matching page, never a reason to write one: filing an unsourced page to clear a floor
+is exactly what AGENTS.md invariant 3 forbids, floor or no floor. The same holds for an
+authorised link label with zero live uses (`wikitool xref` reports these) - fill it when a real
+edge calls for it, never manufacture one to exercise the label.
+
+To check the two floors without a dedicated command, walk `kb/**/*.md` (excluding
+`INDEX.md`/`COLLECTION.md`/`CONTRACT.md`/`CONVENTIONS.md`), parse frontmatter, and: count pages
+whose `related:` array (resolved against page titles) has ≥3 entries for outbound density; count
+`sources:` array length ≥3 for the provenance floor. `wikitool search` and `wikitool lint`
+cover everything else in the table.
+
+## What a reactive fix may do to kb/ content
+
+Three tiers, by how much of the corpus a change touches:
+
+1. **Pointwise - always allowed.** Creating, updating, renaming or deleting a single page
+   through the normal tools (`new`, `touch`, the page-lifecycle procedure), below the
+   Mass-Update Gate's threshold. This is ordinary work and needs no special permission.
+2. **Corpus-wide - planned only, never reactive.** A migration, a vocabulary sweep, a bulk
+   `touch` across many pages. This needs its own issue and, per `work/CONTRACT.md`, a `work/`
+   run - never a same-session reaction to whatever the session was originally doing. If a
+   session hits the Mass-Update Gate (exit 42, see `instructions/gates.md`) while working on
+   something else, it does not fetch the `--confirm` token to push through: it stops, opens an
+   issue for the corpus-wide change, and finishes the original task without it.
+3. **Reactive - never allowed.** Deleting or reshaping a page to make a failing test pass;
+   restructuring corpus content to route around a tool bug (AGENTS.md invariant 7); using
+   `kb/` as a scratch surface for a tool experiment. If a stack change under development needs a
+   corpus shape that does not exist, build it as a pytest fixture (see the next section) -
+   never manufacture it in `kb/`.
+
+## Relationship to the test fixtures
+
+`tools/chemenu/tests/conftest.py`'s `kb_dir`/`raw_dir` fixtures and `test_pipeline_l0.py` cover
+the **small, isolated** case: a handful of pages, built fresh per test, hermetic. `kb/` covers
+the **large, connected** case: 181+ pages, grown link density, real provenance history that no
+per-test fixture reconstructs economically. The cut: if a `tmp_path` tree can reproduce what the
+test needs, it belongs in a fixture; if the test needs density or scale that only a grown corpus
+has, it belongs against `kb/`. Neither absorbs the other's job - see
+[testing-conventions.md](testing-conventions.md).
+
+## Decision points
+
+- **A floor would be violated by an in-progress change - is that a blocker?** Only for the
+  orphan ceiling and the type/subtype floors, since those two can go to zero. The density and
+  provenance floors move gradually with ordinary ingests and are not gating on any single
+  session.
+- **Corpus is "too small" for a feature under development?** That is not this file's problem to
+  solve by adding pages - see tier 3 above. Either the feature waits for a real ingest to supply
+  the shape, or it gets a pytest fixture.

instructions/dev/stack-dev/SKILL.md

@@ -45,18 +45,44 @@ stack development happens in the origin repo instead (see AGENTS.md's routing li
    [issue-tracking.md](../issue-tracking.md) - open work lives in Gitea issues, one per work
    package, labelled `area/`, `kind/`, `prio/` and `size/`. There is no `TODO.md`. **The body
    of the issue you are working on is this session's plan file:** keep it current as the state
-   moves, not at the end, so an interrupted session leaves a body the next one can resume from.
-   Read it before filing something for later, before editing or closing an issue, or before
-   deciding what to pick up next.
+   moves, so an interrupted session leaves a body the next one can resume from, *and* rewrite it
+   to its final state before closing. Both halves bind; the second is step 6 below. Read it
+   before filing something for later, before editing or closing an issue, or before deciding
+   what to pick up next.
    [testing-conventions.md](../testing-conventions.md) - the suite runs against a deliberately
    empty machine; what the autouse fixture already neutralizes, and what a test still has to
    establish itself. Read it before adding or changing a test.
    [version-parts.md](../version-parts.md) - which part a change bumps: the drop-in test, the
    catalogue of breaks that cross the compatibility boundary with `kb/` untouched, and what to
-   put in front of the user before a breaking bump. Read it before step 3.
+   put in front of the user before a breaking bump. Read it before step 4.
+   [corpus-policy.md](../corpus-policy.md) - what "curated enough" means for the shared
+   demo/testbed `kb/`, the measurable floors that define it, and what a reactive fix may and may
+   not do to corpus content. Read it before judging whether the corpus can exercise a change, or
+   before any fix that would touch `kb/` content.
    More instructions are added here incrementally as stack-development needs come up - this
    list grows without needing this skill file to change shape.
-3. **Raise the version, if the change ships.** A change under `tools/`, `types/`,
+3. **Settle the design before building - and break there for the model switch.** These are two
+   different kinds of work, and the split is not stylistic: design, the version part and any
+   boundary judgment have **no** mechanical guard, while the code and tests that follow have
+   `pytest`, `docs verify`, `instructions verify` and CI behind them.
+
+   So when the design is settled - the issue body says what will be built, the open questions are
+   answered - stop and say so, in one sentence:
+
+   > Der Plan steht, ab hier ist die Arbeit mechanisch und durch Tests/CI abgedeckt. Wenn du auf
+   > Opus bist, ist jetzt der Moment für `/model sonnet` bei Effort `high`.
+
+   **You cannot make this switch yourself** - the session's model is the user's `/model`, not a
+   setting an agent applies. Offer it once and keep working either way; a session that argues
+   about its own model has already cost more than the difference. If the design turns out not to
+   be settled after all - a boundary crossing surfaces, an assumption breaks - that is a reason to
+   offer the switch back up, not to decide it alone.
+
+   Effort is the cheaper lever than the model, and `high` is the floor for anything touching more
+   than one file or a contract. Full table and reasoning:
+   [claude-code-model-selection.md](../../claude-code-model-selection.md).
+
+4. **Raise the version, if the change ships.** A change under `tools/`, `types/`,
    `instructions/`, `AGENTS.md` or a `CONTRACT.md` reaches every future instance, so it needs a
    version and a changelog entry:
 
@@ -93,13 +119,48 @@ stack development happens in the origin repo instead (see AGENTS.md's routing li
    Prose-only changes (`README.md`, `INSTALL.md`, `EVALS.md`) and the workflows under `.gitea/`
    do not need a bump - CI's version gate is scoped to what changes behaviour.
 
-4. **Verify before publishing.** `tools/wikitool docs verify`, `tools/wikitool instructions
+5. **Verify before publishing.** `tools/wikitool docs verify`, `tools/wikitool instructions
    verify`, and the relevant `pytest` run in `tools/` - the same checks any stack change must
    pass, run explicitly rather than assumed. CI (`.gitea/workflows/ci.yml`) runs these plus a
    full `setup-instance.md` replay against a fresh `dist export`; a push to `main` that moves
    `VERSION` additionally triggers a tagged release. **CI does the tagging** - a session never
    creates a tag, which is what keeps AGENTS.md invariant 5 intact.
 
+6. **Close the issue with a body rewrite, not a comment.** The last act of a session that
+   finished a work package, and the one most easily skipped: by here the change is published and
+   the issue feels done. It is not. The body is the version everyone reads afterwards and nobody
+   revisits, so it is the one place the debt comes due at the worst moment.
+
+   **Break here too, in the other direction.** Everything left in the session - this rewrite,
+   whether a `docs/` page's reasoning went stale, the changelog prose - is the unchecked kind of
+   work again, the mirror of step 3. If the session dropped to Sonnet there, say so now:
+
+   > Ab hier greift kein maschineller Check mehr - Issue-Body, `docs/`-Veralterung und
+   > Changelog-Prosa prüft nichts. Wenn du zurück auf Opus willst, ist jetzt der Moment.
+
+   Then **do the work regardless of the answer.** Never block a close on a model switch: the
+   change is already published, and a session that stops here leaves exactly the state this step
+   exists to prevent. If it ran on the cheaper model, name that in the handover rather than
+   leaving it silent.
+
+   Rewrite it to its final state *first*, then close. The test is what a reader who opens the
+   closed issue tomorrow would conclude:
+
+   - every acceptance criterion ticked, or struck with the reason it was dropped
+   - proposals that were decided read as decided; a "to decide" section has become the decision
+     and its reasoning
+   - nothing left in the present tense about a defect that no longer exists
+   - what was verified is named - which checks ran, which CI run - not a commit hash alone
+
+   Then one short comment naming what changed against the previous state, and nothing else.
+
+   **A closing report in a comment does not satisfy this**, however thorough: it reads as
+   complete to whoever writes it and leaves a body still phrased as open work. Nothing
+   mechanical catches it - `wikitool` does not know this tracker exists and must not learn it,
+   since it ships to instances that have no board - so this step is the only enforcement there
+   is. #44 and #45 both closed exactly this way, the second an hour after the rule was written.
+   [issue-tracking.md](../issue-tracking.md) step 7 has the full shape.
+
 ## Decision points
 
 - **Touches both stack code and wiki content in one session?** Apply this skill's rules to the
@@ -110,7 +171,10 @@ stack development happens in the origin repo instead (see AGENTS.md's routing li
   user decides whether it is worth that: show them what breaks, what an instance has to do about
   it, and the alternatives (avoid the break with a shim, defer and batch it with the next one,
   or split it behind a deprecation window), then recommend one and wait for a go-ahead.
-  [version-parts.md](../version-parts.md) step 4 has the full shape.
+  [version-parts.md](../version-parts.md) step 4 has the full shape. A surfacing boundary crossing
+  is also a reason to offer the model switch back up (step 3): the judgment it needs has no
+  mechanical guard, and `docs verify` only checks that a crossing documents itself, never that the
+  part was chosen correctly.
 
 ## Scope
 

instructions/kb-profiles.md

@@ -68,7 +68,7 @@ stack's hardcoded behaviour until the conventions file existed.
 |---|---|
 | `language:` | `de` |
 | `sections:` | `Beziehungen` / `Siehe auch` / `Fußnoten` |
-| Naming | Human-readable titles with spaces; singular for entities; `adr-NNN-` for decisions; `X vs Y` for comparisons |
+| Naming | Human-readable titles with spaces; singular for entities; a decision named like any other concept, no `adr-NNN-` prefix; `X vs Y` for comparisons |
 | Tone | Wikipedia register, with a German buzzword and filler list |
 | Relationship labels | `hängt ab von` · `verwendet` · `implementiert` · `erweitert` · `ersetzt` · `steht in Konflikt mit` · `benötigt` · `erzeugt` · `konsumiert` · `besitzt` · `pflegt` · `läuft auf` · `verwandt mit` |
 | Confidence rubric | 0.5 base, +0.2 per supporting source (max +0.6), recency and source-quality bonuses; hedge with "möglicherweise"/"kann" below 0.6, "unsicher"/"unbestätigt" below 0.4 |

kb/CONVENTIONS.md

@@ -62,7 +62,8 @@ There is no `## Siehe auch` region any more. It was the reciprocal half of a bid
 - Human-readable titles with spaces: `Hybrid Search.md`, `Gitea Actions.md` - not kebab-case.
 - Singular for entities: `ha-core.md`, not `ha-cores.md`.
 - Comparison pages read as a comparison: `Go vs Rust.md`.
-- ADRs are prefixed: `adr-001-use-go-modules.md`.
+- A decision (`concept_type: decision`) is named like any other concept - no `adr-NNN-` prefix.
+  See [kb/concepts/COLLECTION.md § Decisions](concepts/COLLECTION.md#decisions).
 - Prefer readability over convention when the two conflict.
 
 What to name a thing: projects use their repository or common name; systems a descriptive

kb/concepts/COLLECTION.md

@@ -27,19 +27,27 @@ tone, relationship labels, the confidence rubric. Neither is restated here.
 
 `concept` (`tools/wikitool types describe concept`).
 
-## Decisions and ADRs
+## Decisions
 
-An architectural decision is a concept page, prefixed as
-[kb/CONVENTIONS.md § Naming](../CONVENTIONS.md#naming) says. It records:
+An architectural decision is an ordinary concept page with `concept_type: decision`
+(`tools/wikitool types describe concept`) - not a separate format, and not a separate location.
+There is no `adr-NNN-`-prefixed filename and no dedicated directory: naming follows
+[kb/CONVENTIONS.md § Naming](../CONVENTIONS.md#naming) like every other concept, and the page
+lives in `kb/concepts/` like every other concept.
 
-- **Context** - what forced a decision.
-- **Decision** - what was chosen.
-- **Consequences** - what this costs, not only what it buys.
-- **Status** - proposed / accepted / deprecated / superseded.
-- Links to every entity the decision affects.
+The body is organic prose under this collection's usual sections, not a fixed template. What it
+still has to carry: what was decided, what forced the decision, what it costs (not only what it
+buys), and a link to every entity the decision affects. A `**Status:**` line is optional - most
+decision pages in this instance carry none, because the page's own prose already says whether the
+decision stands.
 
-A superseded ADR is never deleted or rewritten. The new one declares `supersedes` pointing at
-it; the old one needs no edge back, because its inbound view renders the replacement.
+A decision superseded by a later one is never deleted or rewritten. The new page declares
+`supersedes` pointing at it; the old one needs no edge back, because its inbound view renders the
+replacement.
+
+`concept_type: decision` is also the one subtype [kb/CONTRACT.md](../CONTRACT.md)'s confidence
+machinery treats differently: `confidence decay` skips it structurally, because elapsed time does
+not falsify a decision - only a later decision superseding it does.
 
 ## Authorised labels
 

tools/CONTRACT.md

@@ -68,10 +68,10 @@ tools/wikitool <command> --help
 | `instructions sync [--force]` | Publish every `instructions/<name>/SKILL.md` into `.agents/skills/` and `.claude/skills/` as **copies**, and delete published skills whose source is gone. Both targets are gitignored, so a fresh clone runs this once - see `instructions/bootstrap.md`. Re-running is also how a drifted copy is repaired: the source always wins. `--force` is required only to replace a target directory that is not a published skill at all (no `SKILL.md` in it) |
 | `instructions verify` | Check the instruction layer: flat instructions validate against `types/instruction.schema.yaml`, each `SKILL.md` carries the frontmatter its harness reads, every published copy is byte-identical to its source, no instruction is left that nothing references, and nothing under `instructions/dev/` is referenced from outside it (a `<!-- dist:strip-start/end -->` block is exempt - see [instructions/CONTRACT.md](../instructions/CONTRACT.md)). Missing *every* copy is reported as "run sync", not as drift - that is a clean checkout |
 | `instructions list [--json]` | List the flat instructions with their descriptions. This is how the layer is discovered; `search` deliberately covers `kb/` only |
-| `docs verify` | Check the docs that mirror the code: every CLI command documented here (and vice versa), every directory under `kb/` has a `COLLECTION.md` and no directory outside it does, every collection declaring `profile:` and a `required_by_stack:` that agrees with the stack's own list, `kb/CONVENTIONS.md` naming all three tool-owned section headings if it exists at all, every stage contract present, no pre-migration `type: entity` blocks left in the contracts, and the `.gitignore` canaries clear in both directions (nothing ignored under `raw/`/`kb/`, everything ignored under `reports/` and the published skill directories) |
+| `docs verify` | Check the docs that mirror the code: every CLI command documented here (and vice versa), every directory under `kb/` has a `COLLECTION.md` and no directory outside it does, every collection declaring `profile:` and a `required_by_stack:` that agrees with the stack's own list, `kb/CONVENTIONS.md` naming all three tool-owned section headings if it exists at all, every stage contract present, no pre-migration `type: entity` blocks left in the contracts, and the `.gitignore` canaries clear in both directions (nothing ignored under `raw/`/`kb/`, everything ignored under `reports/` and the published skill directories). The name is about documentation parity, not about the `docs/` directory - it neither reads nor requires one, the same way `kb/` predates the collection it now checks |
 | `eval sessions [--json]` | List the sessions that have a trace under `reports/telemetry/`, most recent first. Read-only and exempt from the Iteration Budget Gate |
 | `eval score [--session <id>] [--json] [--markdown out.md] [--save] [--fail-on-error]` | Score one traced session: structural state from `lint`'s own checks (L1) plus trajectory rules over the trace (L2) - was a refused call repeated unchanged, was a gate flag passed without that gate having refused anything, did a publish of `kb/` pages go unlogged. Defaults to the current session. `--save` writes `reports/evals/<date>/<session>.{json,md}`. Read-only over `kb/` and exempt from the budget; see [../EVALS.md](../EVALS.md) |
-| `dist export <target> [--dry-run] [--source-repo U] [--source-commit SHA] [--release-url U] [--update-url U]` | Write a contentless, distributable copy of this repo's machinery into an empty `<target>` directory: `AGENTS.md`/`README.md`/`EVALS.md` with any `<!-- dist:strip-start -->...<!-- dist:strip-end -->` region removed, `instructions/` (minus `instructions/dev/`), `types/` (the `root: kb` page type-specs and their schemas re-keyed as `.template`, the stack's own verbatim), `tools/` (no venv/caches), the `.github/hooks/`+`.vibe/` session-tracing config plus `.claude/settings.json`, `kb/CONTRACT.md` (no pages, no areas), empty `raw/{articles,documents,notes,assets}/`, `VERSION`, `USER.md.template`/`SOUL.md.template` plus `kb/CONVENTIONS.md.template` and each collection's contract re-keyed as `kb/<name>/COLLECTION.md.template` (the templates ship; the filled `USER.md`/`SOUL.md`/`kb/CONVENTIONS.md`/`kb/<name>/COLLECTION.md`/`types/<page-type>.md` never do - all of them bind their instance and none are the stack's to decide, and `find_leaks` refuses a plan carrying one), and a generated `.wikitool-release.json` stamp (version, export date, origin, and a sha256 per exported file - the base a later upgrade would compare against). The four origin options only fill stamp fields: `export` never calls git and cannot discover them. Refuses a non-empty target, and a tree with no `VERSION`. See `instructions/setup-instance.md`. One-way: there is no command that reconstructs a distributed instance into a dev instance - work on the stack in the origin repo (or a new dev instance exported from it) instead |
+| `dist export <target> [--dry-run] [--source-repo U] [--source-commit SHA] [--release-url U] [--update-url U]` | Write a contentless, distributable copy of this repo's machinery into an empty `<target>` directory: `AGENTS.md`/`README.md`/`EVALS.md` with any `<!-- dist:strip-start -->...<!-- dist:strip-end -->` region removed, `instructions/` (minus `instructions/dev/`), `types/` (the `root: kb` page type-specs and their schemas re-keyed as `.template`, the stack's own verbatim), `docs/` verbatim, `tools/` (no venv/caches), the `.github/hooks/`+`.vibe/` session-tracing config plus `.claude/settings.json`, `kb/CONTRACT.md` (no pages, no areas), empty `raw/{articles,documents,notes,assets}/`, `VERSION`, `USER.md.template`/`SOUL.md.template` plus `kb/CONVENTIONS.md.template` and each collection's contract re-keyed as `kb/<name>/COLLECTION.md.template` (the templates ship; the filled `USER.md`/`SOUL.md`/`kb/CONVENTIONS.md`/`kb/<name>/COLLECTION.md`/`types/<page-type>.md` never do - all of them bind their instance and none are the stack's to decide, and `find_leaks` refuses a plan carrying one), and a generated `.wikitool-release.json` stamp (version, export date, origin, and a sha256 per exported file - the base a later upgrade would compare against). The four origin options only fill stamp fields: `export` never calls git and cannot discover them. Refuses a non-empty target, and a tree with no `VERSION`. See `instructions/setup-instance.md`. One-way: there is no command that reconstructs a distributed instance into a dev instance - work on the stack in the origin repo (or a new dev instance exported from it) instead |
 | `version show [--json]` | Print this instance's stack version and where it came from (development tree, or a distribution with its export date and origin). Bare `wikitool version` is an alias for this. Read-only, offline, and **exempt from the Iteration Budget Gate** |
 | `version check [--url U] [--timeout S] [--json]` | Ask the origin's release feed whether a newer stack exists, and whether the step crosses a compatibility boundary (`state: current\|update\|migration\|ahead`). **The only command in `wikitool` that makes a network call** - never reached implicitly from another command, needs no key, times out, and reports an unreachable feed as an error rather than as "up to date". The feed is `$WIKITOOL_UPDATE_URL`, else the release stamp's, else the built-in origin; `$WIKITOOL_UPDATE_TOKEN` is only needed if that feed is not readable anonymously. Read-only and exempt from the budget gate |
 | `version notes [--version X.Y.Z]` | Print one version's `CHANGES.md` entry, for use as release notes (default: this tree's `VERSION`). Read-only and exempt from the budget gate |

tools/chemenu/commands/confidence_decay.py

@@ -11,6 +11,14 @@ Keeping the undecayed anchor in `confidence_base` is what makes repeated runs
 idempotent - decaying the stored `confidence` in place (the pre-2026-08-13
 behavior) compounded on every run, because the elapsed-months factor kept
 growing while the multiplicand had already shrunk.
+
+Pages with `concept_type: decision` are skipped structurally, not as an
+interim measure. The formula models staleness - a claim that nobody has
+re-checked in a while becomes less trustworthy - and a decision is not a
+claim about the world that time can falsify. What retires a decision is a
+later decision superseding it, never elapsed months on its own; that is a
+category the decay formula does not have a term for, so it does not apply
+one.
 """
 from __future__ import annotations
 
@@ -118,6 +126,8 @@ def confidence_decay(
     missing_base = []
 
     for title, page in sorted(pages.items()):
+        if page.frontmatter.get("concept_type") == "decision":
+            continue
         confidence = page.frontmatter.get("confidence")
         if confidence is None:
             continue

tools/chemenu/commands/dist_cmd.py

@@ -347,6 +347,12 @@ def build_plan(origin: Optional[Origin] = None) -> dict[str, PlannedFile]:
     for hook_dir in HOOK_DIRS:
         plan.update(_copy_tree(config.ROOT / hook_dir, hook_dir, frozenset()))
 
+    # docs/ is stack background - why the stack is built the way it is - and
+    # ships verbatim like instructions/ and types/: it carries no page, no
+    # frontmatter, and (AGENTS.md § File naming) no normative sentence, so
+    # there is nothing instance-owned in it to split off as a .template.
+    plan.update(_copy_tree(config.ROOT / "docs", "docs", frozenset()))
+
     # `kb/CONTRACT.md` is stack-owned and ships verbatim; everything beside it
     # under `kb/` is the instance's own and ships only as a `.template`. That is
     # the personalization split (`USER.md`/`SOUL.md`) one directory down, and
@@ -492,7 +498,8 @@ def export_command(
 ):
     """Export a contentless, distributable copy of this repo's machinery:
     AGENTS.md/README.md (dev-instance-only marker blocks removed),
-    instructions/ (no instructions/dev/), types/, tools/ (no venv/caches),
+    instructions/ (no instructions/dev/), types/, docs/ verbatim,
+    tools/ (no venv/caches),
     the .github/hooks/+.vibe session-tracing config plus .claude/settings.json,
     kb/CONTRACT.md plus a COLLECTION.md.template per collection and
     kb/CONVENTIONS.md.template (no pages, no areas), empty

tools/chemenu/tests/test_confidence_decay.py

@@ -5,7 +5,7 @@ import pytest
 from chemenu import config
 from chemenu.commands import confidence_decay
 from chemenu.commands.confidence_decay import FLOOR, compute_decay
-from chemenu.frontmatter_io import read_page
+from chemenu.frontmatter_io import read_page, write_page
 
 
 def test_no_decay_at_zero_months():
@@ -68,3 +68,22 @@ def test_decay_skips_pages_without_a_base(decay_wiki):
     confidence_decay.confidence_decay(apply=True)
     frontmatter, _ = read_page(decay_wiki / "entities/systems/aurora.md")
     assert frontmatter["confidence"] == 0.9
+
+
+def test_decay_skips_decision_pages(decay_wiki):
+    """A decision is not falsified by elapsed time, only by a later decision
+    superseding it - `concept_type: decision` is a categorical skip, not
+    something an old `modified` date should ever decay (Gitea #38)."""
+    decision_path = decay_wiki / "concepts" / "some-decision.md"
+    write_page(
+        decision_path,
+        {
+            "type": "types/concept.md", "concept_type": "decision",
+            "tags": [], "created": "2015-01-01", "modified": "2015-01-01",
+            "related": [], "sources": [], "confidence": 0.9, "confidence_base": 0.9,
+        },
+        "\n# some-decision\n",
+    )
+    confidence_decay.confidence_decay(apply=True)
+    frontmatter, _ = read_page(decision_path)
+    assert frontmatter["confidence"] == 0.9

tools/chemenu/tests/test_dist_cmd.py

@@ -135,6 +135,10 @@ def repo(tmp_path: Path, monkeypatch: pytest.MonkeyPatch) -> Path:
         f"<!-- {config.TEMPLATE_SENTINEL} -->\n# conventions template\n", encoding="utf-8"
     )
 
+    docs_dir = root / "docs"
+    docs_dir.mkdir()
+    (docs_dir / "why-gates-are-code.md").write_text("# Why gates are code\n", encoding="utf-8")
+
     for relative in ("raw/CONTRACT.md", "reports/CONTRACT.md", "work/CONTRACT.md"):
         path = root / relative
         path.parent.mkdir(parents=True, exist_ok=True)
@@ -156,6 +160,11 @@ def test_plan_never_includes_commonplace(repo):
     assert "commonplace" not in combined
 
 
+def test_plan_ships_docs_verbatim(repo):
+    plan = dist_cmd.build_plan()
+    assert plan["docs/why-gates-are-code.md"].content == "# Why gates are code\n"
+
+
 def test_plan_never_includes_instructions_dev(repo):
     """instructions/dev/ - flat dev-only instructions and the nested skill
     that switches a session into tool-development mode - is pruned