docs(corpus): standard for diagrams in corpus nodes (#1312)

launchpad/decisions/ADR-0031-knowledge-find-ask-permanently-descoped.md

@@ -10,6 +10,11 @@ supersedes: none
 
 ## Decision
 
+**Option E.** `knowledge.find(query)` and `knowledge.ask(text)` — the two of #211's
+seven methods with an unbounded free-text input domain — are permanently out of scope
+for the shipped knowledge crate. #1400 (the task that would have implemented them)
+closes as won't-do rather than staying open against a design nobody is committing to
+build. `#211`'s existing Python implementation of both methods stays in the tree,
 **Option E.** `knowledge.find(query)` — one of #211's seven methods — and
 `knowledge.ask(text)` — a separate eighth entry point that shares `find`'s routing
 rather than being one of the seven (`CONTRACT.md`: "an eighth public function, not one
@@ -91,6 +96,9 @@ written.
 
 - Ruling 11 and Ruling 12 both continue to hold as ratified by ADR-0027; this decision
   does not reopen or reinterpret either.
+- `find`/`.ask` are the only two of #211's seven methods with an unbounded input
+  domain — the other six are keyed on a symbol, an area, or a task name, and all six
+  already return correct content from the packaged corpus.
 - `find` (one of #211's seven methods) and `ask` (a separate eighth entry point that
   shares `find`'s routing, not one of the seven) are the only two surfaces with an
   unbounded input domain — the other six methods are keyed on a symbol, an area, or a
@@ -105,6 +113,9 @@ written.
   (A) or as a called sidecar (C) — inherits that gap; Option B would too, since it
   reuses the same computation, only relocated.
 - `#533`'s own milestone entry is scoped Effort: Low against a 2026-09-11 date. Six of
+  seven methods already ship real, working capability; the concrete cases an agent
+  actually asks (explain a symbol, its dependencies, impact, setup, conventions,
+  history) are already covered.
   seven methods already work end-to-end in `#211`'s Python prototype — real, tested
   capability, not yet packaged into the shipped crate (`#551`/`#552`); the concrete
   cases an agent actually asks (explain a symbol, its dependencies, impact, setup,
@@ -144,6 +155,8 @@ written.
 ## Consequences
 
 **Good.** #1400 stops carrying an undecided architecture question inside a Task's
+Definition of done. The project board reflects reality: six methods ship, one is
+permanently out of scope, instead of an open task against a design nobody is building.
 Definition of done. The project board reflects reality: six of #211's seven methods are
 implemented and validated in the Python prototype awaiting crate packaging, while `find`
 — the seventh — and the separate `ask` entry point are both permanently out of scope,

launchpad/docs/corpus/standards/confidence.md

@@ -68,6 +68,9 @@ evidence:
   - statement: "An INFERENCE whose citation supports the subject of the claim but not the choice the claim makes is a decision in disguise, and belongs in TEAM_KNOWLEDGE attributed to whoever decided it."
     entry_class: TEAM_KNOWLEDGE
     provided_by: "launchpad-26/buzz#636 cross-model final review, relayed in the #1309 task brief: a second INFERENCE on AGENTS.md was rejected as laundering an unsourced policy choice into a class that made it look derived, and was reclassified to TEAM_KNOWLEDGE attributed to the issue's definition of done"
+---
+
+# Standard: `confidence`
   - statement: "Per Serina's decision on launchpad-26/buzz#1486, this node reconciles to #1313's documentation-standard: the H1 no longer code-formats the topic, and the normative sections are named MUST and SHOULD rather than Requirements and Guidance."
     entry_class: TEAM_KNOWLEDGE
     provided_by: "launchpad-26/buzz#1486, decided 2026-08-27"
@@ -148,6 +151,7 @@ it feels about right, and a reader — or a generated view, or a later agent —
 as calibrated and reasons onward from it. The number then carries more weight than
 anything that produced it. Everything below is aimed at that.
 
+## Requirements
 ## MUST
 
 These are MUSTs. The first two are enforced mechanically. Requirement 3 is a MUST that
@@ -173,6 +177,7 @@ review, and a reviewer who lets one through has approved a defect.
    it to TEAM_KNOWLEDGE and name who decided. See *Reasoning versus deciding*.
 6. **A number MUST NOT be moved because someone pushed back on it.** Re-verify the claim
    or reclassify the entry. Adjusting the number to settle an argument records agreement
+   where there was none. Re-encoding an existing value onto the band values in *Guidance*
    where there was none. Re-encoding an existing value onto the band values in *SHOULD*
    is not a move under this rule, but do not tell yourself nothing changed: snapping to a
    peg preserves the assessment while it can and does alter the entry's **rank** against
@@ -184,6 +189,7 @@ review, and a reviewer who lets one through has approved a defect.
    in the same edit.** A number that outlived the reasoning it rated is worse than no
    number, because it still looks current.
 
+## Guidance
 ## SHOULD
 
 These are SHOULDs. Depart from them with a reason.
@@ -387,6 +393,7 @@ edge to make first.
 
 **Expected but not verified when this node was written:**
 
+- **The bands in *Guidance*, and the values `0.8` / `0.6` / `0.4`, are not derived from
 - **The bands in *SHOULD*, and the values `0.8` / `0.6` / `0.4`, are not derived from
   anything.** They are a proposed convention for keeping values coarse, offered because
   unbounded precision is the observed failure. No study, no sample, and one authored

launchpad/docs/corpus/standards/decision-references.md

@@ -116,6 +116,9 @@ evidence:
   - statement: "Issue #1410 owns encoding ADR-0029's claim-type classification and unestablished or flagged state in the corpus schema and validator."
     entry_class: TEAM_KNOWLEDGE
     provided_by: "launchpad-26/buzz#1410"
+---
+
+# Citing a decision
   - statement: "Per Serina's decision on launchpad-26/buzz#1486, this node reconciles to #1313's documentation-standard: the H1 is now 'Standard: decision references', and the enforcement section's name now uses the word 'Enforcement'."
     entry_class: TEAM_KNOWLEDGE
     provided_by: "launchpad-26/buzz#1486, decided 2026-08-27"
@@ -303,6 +306,7 @@ record and take one of three routes:
 The third route is the only one that leaves a superseded citation in place, and it is only
 available to a claim that is explicitly historical.
 
+## What the checks establish, and what they do not
 ## Enforcement, and what the checks establish
 
 The checker's verdict on a decision citation is **structural**. A repository path — bare,