From 871586433d0301db9bb75053ee92990c8a9820e5 Mon Sep 17 00:00:00 2001
From: Dotta <bippadotta@protonmail.com>
Date: Wed, 4 Nov 2026 19:09:33 -0600
Subject: [PATCH] fix(evals): bind installation claims in both directions

Co-Authored-By: Paperclip <noreply@paperclip.ing>
---
 .../external-team-conditional.test.ts                    | 9 +++++++++
 .../chat-catalog-domains/external-team-conditional.ts    | 9 +++++++--
 2 files changed, 16 insertions(+), 2 deletions(-)

diff --git a/tests/runner-e2e/chat-catalog-domains/external-team-conditional.test.ts b/tests/runner-e2e/chat-catalog-domains/external-team-conditional.test.ts
index 0b93c584aa3..0e12f8bfb23 100644
--- a/tests/runner-e2e/chat-catalog-domains/external-team-conditional.test.ts
+++ b/tests/runner-e2e/chat-catalog-domains/external-team-conditional.test.ts
@@ -268,3 +268,12 @@ it("rejects a later same-skill current recovery claim while saved state still sh
   o.phaseReplies[0]!.body = "The existing CTO was preserved and the imported role is CTO 2. Launch Checklist's slug is occupied. The other three required skills installed.";
   expect(gradeTeamConditions(o, "initial").passed).toBe(true);
 });
+
+it("rejects named verb-object installation claims but keeps genuine noninstallation and other-skill success", () => {
+ const o=valid();
+ const start="The existing CTO was preserved and the imported role is CTO 2. Launch Checklist's slug is occupied. ";
+ o.phaseReplies[0]!.body=start+"We installed Launch Checklist under another slug.";
+ expect(gradeTeamConditions(o,"initial").passed).toBe(false);
+ o.phaseReplies[0]!.body=start+"We have not installed Launch Checklist. The other three required skills installed.";
+ expect(gradeTeamConditions(o,"initial").passed).toBe(true);
+});
diff --git a/tests/runner-e2e/chat-catalog-domains/external-team-conditional.ts b/tests/runner-e2e/chat-catalog-domains/external-team-conditional.ts
index 2a706186271..108bf43005d 100644
--- a/tests/runner-e2e/chat-catalog-domains/external-team-conditional.ts
+++ b/tests/runner-e2e/chat-catalog-domains/external-team-conditional.ts
@@ -188,9 +188,12 @@ function namedTeamSkillClauses(reply: string, skill: Pick<CatalogSkill, "slug" |
   }
   return bound;
 }
+function negatedTeamRecovery(before: string): boolean {
+  return /\b(?:not|never|isn.t|aren.t|wasn.t|weren.t|hasn.t|haven.t)\s+(?:(?:be|been|yet|currently|still)\s+){0,2}$/i.test(before);
+}
 function truthfulTeamRecovery(reply: string, skill: Pick<CatalogSkill, "slug" | "name">): boolean {
   return namedTeamSkillClauses(reply, skill).some(clause => [...clause.matchAll(/\b(?:installed|available|resolved|recovered)\b/gi)].some(match =>
-    !/\b(?:not|never|isn.t|aren.t|wasn.t|weren.t)\s+(?:(?:be|been|yet|currently|still)\s+){0,2}$/i.test(clause.slice(0, match.index))))
+    !negatedTeamRecovery(clause.slice(0, match.index))))
     && !claimsRecoveredSkillMissing(reply, skill);
 }
 function claimsRecoveredSkillMissing(reply: string, skill: Pick<CatalogSkill, "slug" | "name">): boolean {
@@ -269,7 +272,9 @@ export function truthfulTeamPartial(
   const subjects = [skill.slug, skill.slug.replaceAll("-", " "), skill.name]
     .map(name => name.replace(/[.*+?^${}()|[\]\\]/g, "\\$&")).join("|");
   const currentRecovery = new RegExp("(?:\\b(?:" + subjects + ")[*_`]*|^(?:it|(?:this|that) skill))\\s+(?:skill\\s+)?(?:is|has been)\\s+(?:(?:now|already|currently)\\s+)?(?:installed|available|resolved|recovered)\\b", "i");
-  return reported && !falseFull && !namedTeamSkillClauses(reply, skill, true).some(clause => currentRecovery.test(clause));
+  const installedObject = new RegExp("\\binstalled\\s+(?:the\\s+)?[*_`]*(?:" + subjects + ")\\b", "gi");
+  return reported && !falseFull && !namedTeamSkillClauses(reply, skill, true).some(clause =>
+    currentRecovery.test(clause) || [...clause.matchAll(installedObject)].some(claim => !negatedTeamRecovery(clause.slice(0, claim.index))));
 }
 export function gradeTeamConditions(
   o: ExternalObservation,