docs(agents): separate gap diagnosis from admission

Author: morluto

.agents/skills/learn-from-math-agent-trajectories/SKILL.md

@@ -124,14 +124,27 @@ postcondition with typed inputs and an exact, incomplete, or unknown result. It
 must not encode the motivating conjecture, prescribe a proof strategy, or
 claim an asymptotic theorem that still requires model reasoning.
 
+Separate gap diagnosis from public-operation admission. First decide whether
+the trajectory establishes a reusable missing mathematical postcondition on
+the inspected surface. Recording that gap does not assert that the result
+belongs in the agent-visible catalog. Then report the admission evidence
+separately: it may support public consideration, suggest a native-only helper,
+or leave the disposition unresolved. Public-admission concerns such as weak
+leverage over ordinary Python or a cheap projection can change the eventual
+disposition without erasing an otherwise well-evidenced gap. Conversely, an
+absent convenience is not an operation gap unless the missing postcondition is
+independently canonical or reusable.
+
 ## Turn observations into repository learning
 
 Choose the smallest durable action supported by the evidence:
 
 - update a repo-local skill for a reusable agent decision rule;
 - improve discovery metadata when an existing operation was hard to find;
 - repair an operation contract when a selected tool was hard to call or use;
-- propose a new operation for a repeated, stable bounded postcondition;
+- record an operation gap for a repeated, stable bounded postcondition, and
+  propose public admission only when the separate admission evidence supports
+  it;
 - update product documentation only for a public contract or durable product
   behavior that users need outside agent instructions;
 - preserve a trajectory as an evaluation when it has a frozen input,
@@ -163,6 +176,9 @@ Lead with what the trajectory teaches, not a chronological replay. Include:
   actions; and
 - unresolved mathematical work that still requires proof or human judgment.
 
+For each operation-gap finding, state the gap verdict and the admission posture
+separately. Do not make a reader infer one from the other.
+
 Keep confirmed facts, plausible hypotheses, and open questions visibly
 separate. The retrospective succeeds when another agent or maintainer can
 verify the lesson from the original artifacts and decide on a bounded next

.github/ISSUE_TEMPLATE/math-gap.md

@@ -15,7 +15,7 @@ assignees: []
 <!-- What custom code, external library call, solver encoding, or manual work was needed? -->
 
 ## Gap classification
-<!-- Check one. Only a genuine operation gap normally proceeds to admission. -->
+<!-- Check one. An operation gap records a missing postcondition; it does not pre-admit a public operation. -->
 - [ ] representation
 - [ ] interoperability
 - [ ] discovery
@@ -24,6 +24,9 @@ assignees: []
 - [ ] operation
 - [ ] reasoning
 
+## Admission posture (operation gaps only)
+<!-- State whether the evidence supports public-catalog consideration, suggests native-only support, or leaves disposition unresolved. Public admission is a later decision. -->
+
 ## Existing math.find queries
 <!-- Queries already tried and what they returned. -->
 

docs/explanation/executable-mathematical-vocabulary.md

@@ -102,6 +102,14 @@ Not every failed attempt reveals a missing operation.
 
 Only a genuine operation gap normally motivates a new public operation.
 
+Gap diagnosis and public-operation admission are separate decisions. A focused
+issue may record a reusable missing postcondition even when the available
+evidence does not yet establish enough leverage for the agent-visible catalog,
+or when the eventual disposition may be `NATIVE_ONLY`, `SPLIT`, or `DROP`.
+Admission gates decide that later disposition; they should not suppress an
+evidence-backed gap record. Conversely, recording an operation gap does not
+pre-admit a catalog operation.
+
 ## What tends to be useful
 
 Useful operations usually turn substantial computation or mathematical