prompt: add existing-first uncertainty fallback

Add one explicit action-order rule in Additional Guidance for when
rule #6 applies to explicitly unresolved uncertainty:

1. First check whether an existing unresolved node already represents
   the same uncertainty.
2. If so, update/refine that existing structure rather than creating
   a duplicate.
3. If no such node exists, add a new unknown that directly represents
   the unresolved uncertainty.
4. Do not use an edge alone to represent a previously unrepresented
   uncertainty.

14 focused prompt tests verify: existing-first ordering, reuse path,
fallback-to-add, related-node-insufficient, edge-only-prohibited,
possibleInference separation, resolution path preserved, duplicate
contract preserved, scope uncertainty-only, fidelity/traceability
preserved, noop validator untouched, no semantic classifier added.
This commit is contained in:
2026-08-11 14:12:42 +01:00
parent 96b855b8e7
commit a5dd9d3f1a
2 changed files with 153 additions and 0 deletions
+1
View File
@@ -123,6 +123,7 @@ The JSON object must contain exactly these top-level fields:
## Additional Guidance
- If the answer only clarifies an existing unknown, prefer updatedNodes and resolvedUnknownNodeIds over creating duplicate nodes.
- When rule #6 applies to explicitly unresolved uncertainty: first check whether an existing unresolved node already represents the same uncertainty; if so, update/refine that existing structure rather than adding a duplicate; if no such node exists, add a new unknown that directly represents the unresolved uncertainty; do not use an edge alone to represent a previously unrepresented uncertainty.
- When an answer resolves an existing unknown, include that existing node ID in resolvedUnknownNodeIds and update that node rather than creating only a parallel observation.
- If the answer creates a more specific decision situation, add the smallest set of new nodes and edges needed to represent that situation and only its most consequential unknowns.
- If you add a new unknown, do not leave it floating: connect it with an added edge to the relevant decision/context node created or updated from the answer.
+152
View File
@@ -281,3 +281,155 @@ describe("buildGraphUpdatePrompt — semantic-to-mutation MUST rule", () => {
expect(fullPrompt).not.toContain("must add a new node");
});
});
// ── Experiment 57J.46 — existing-first uncertainty fallback ───
describe("buildGraphUpdatePrompt — 57J.46 existing-first uncertainty fallback", () => {
function getAdditionalGuidance(prompt) {
return prompt.split("## Additional Guidance")[1];
}
// Test 1 — existing-first ordering exists
it("test 1: assembled prompt explicitly says to check for an equivalent unresolved unknown first", () => {
const prompt = buildGraphUpdatePrompt(makeContext());
const guidance = getAdditionalGuidance(prompt);
expect(guidance).toContain("first check whether an existing unresolved node");
expect(guidance).toContain("represents the same uncertainty");
});
// Test 2 — reuse path explicit
it("test 2: prompt says update/refine existing structure rather than add a duplicate when equivalent exists", () => {
const prompt = buildGraphUpdatePrompt(makeContext());
const guidance = getAdditionalGuidance(prompt);
expect(guidance).toContain("if so, update/refine that existing structure");
expect(guidance).toContain("rather than adding a duplicate");
});
// Test 3 — fallback-to-add explicit
it("test 3: prompt explicitly requires adding a new unknown when no equivalent unresolved unknown exists", () => {
const prompt = buildGraphUpdatePrompt(makeContext());
const guidance = getAdditionalGuidance(prompt);
expect(guidance).toContain("if no such node exists");
expect(guidance).toContain("add a new unknown that directly represents the unresolved uncertainty");
});
// Test 4 — ordered fallback means: existing first, otherwise add
it("test 4: full ordered fallback is present in the assembled prompt", () => {
const prompt = buildGraphUpdatePrompt(makeContext());
const guidance = getAdditionalGuidance(prompt);
// The entire rule must be present as a single coherent instruction:
expect(guidance).toContain(
"When rule #6 applies to explicitly unresolved uncertainty: first check whether an existing unresolved node already represents the same uncertainty; if so, update/refine that existing structure rather than adding a duplicate; if no such node exists, add a new unknown that directly represents the unresolved uncertainty; do not use an edge alone to represent a previously unrepresented uncertainty.",
);
});
// Test 5 — merely related node is insufficient
it("test 5: prompt does not imply a general related state/cost node counts as representing the same uncertainty", () => {
const prompt = buildGraphUpdatePrompt(makeContext());
const guidance = getAdditionalGuidance(prompt);
// The rule requires "same uncertainty" — not "related" or "similar":
expect(guidance).toContain("same uncertainty");
// Must not use weaker criteria:
expect(guidance).not.toContain("similar");
expect(guidance).not.toContain("related to");
});
// Test 6 — edge-only insufficient
it("test 6: prompt explicitly prevents using an edge alone to represent previously unrepresented uncertainty", () => {
const prompt = buildGraphUpdatePrompt(makeContext());
const guidance = getAdditionalGuidance(prompt);
expect(guidance).toContain(
"do not use an edge alone to represent a previously unrepresented uncertainty",
);
});
// Test 7 — possibleInference does not create unknowns
it("test 7: existing possibleInference separation remains intact — rule #27 still separates interpretation from userSupportedMeaning", () => {
const prompt = buildGraphUpdatePrompt(makeContext());
expect(prompt).toContain(
"Put any stronger interpretation in answerMeaning.possibleInference, not in userSupportedMeaning",
);
// The 57J.46 rule must NOT reference possibleInference as a trigger:
const guidance = getAdditionalGuidance(prompt);
expect(guidance).not.toContain("possibleInference");
});
// Test 8 — actual resolution path preserved
it("test 8: prompt still allows resolution when the user's answer genuinely resolves an existing unknown", () => {
const prompt = buildGraphUpdatePrompt(makeContext());
// The "resolve existing" guidance must remain intact:
expect(prompt).toContain(
"include that existing node ID in resolvedUnknownNodeIds",
);
expect(prompt).toContain("update that node rather than creating only a parallel observation");
// Rule #5 (resolve answered unknown first) must still exist:
expect(prompt).toContain("Resolve the answered unknown first when the answer supports it");
});
// Test 9 — duplicate validator/contract preserved
it("test 9: existing duplicate-avoidance wording remains unchanged", () => {
const prompt = buildGraphUpdatePrompt(makeContext());
expect(prompt).toContain("Do not add duplicate unknowns");
expect(prompt).toContain("genuinely new concepts");
const guidance = getAdditionalGuidance(prompt);
expect(guidance).toContain("prefer updatedNodes and resolvedUnknownNodeIds over creating duplicate nodes");
});
// Test 10 — scope remains uncertainty-only (not universal to all categories)
it("test 10: the new action-order rule does not apply universally to facts, constraints, decisions, or other meaning categories", () => {
const prompt = buildGraphUpdatePrompt(makeContext());
const guidance = getAdditionalGuidance(prompt);
// The rule must be scoped to "unresolved uncertainty" specifically:
expect(guidance).toContain("explicitly unresolved uncertainty");
// It must not say "any answer" or "all categories":
expect(guidance).not.toContain("any answer");
expect(guidance).not.toContain("every change");
expect(guidance).not.toContain("all categories");
// Rule #7 (new unknown conditions) and other category rules must be untouched:
expect(prompt).toContain(
"Add new unknown nodes only when the answer introduces a new decision, claim, object, measure, dependency, or unresolved term directly relevant to the case",
);
});
// Test 11 — fidelity / possibleInference separation preserved
it("test 11: fidelity rule separating userSupportedMeaning from possibleInference is untouched", () => {
const prompt = buildGraphUpdatePrompt(makeContext());
expect(prompt).toContain(
"answerMeaning.userSupportedMeaning must state only what the user's answer directly supports",
);
});
// Test 12 — traceability preserved
it("test 12: traceability requirement for new unknowns remains intact", () => {
const prompt = buildGraphUpdatePrompt(makeContext());
expect(prompt).toContain(
"Every new unknown must be directly traceable to the user's answer and its description must state why that uncertainty matters",
);
});
// Test 13 — no noop validator changed
it("test 13: the 'rule #6 does not apply → empty arrays' permission is preserved unchanged", () => {
const prompt = buildGraphUpdatePrompt(makeContext());
const guidance = getAdditionalGuidance(prompt);
expect(guidance).toContain(
"If rule #6 does not apply (the answer contains no user-supported meaning that requires graph progress) and there is no other justification for change, return empty arrays for every category",
);
});
// Test 14 — no semantic classifier or keyword logic added
it("test 14: no new deterministic semantic matcher or keyword matching was added", () => {
const prompt = buildGraphUpdatePrompt(makeContext());
expect(prompt).not.toContain("threshold");
expect(prompt).not.toContain("synonym");
expect(prompt).not.toContain("keyword match");
});
// Test 15 — provider-agnostic preserved
it("test 15: no provider-specific wording added", () => {
const prompt = buildGraphUpdatePrompt(makeContext());
expect(prompt).not.toContain("qwen");
expect(prompt).not.toContain("claude");
expect(prompt).not.toContain("gpt");
expect(prompt).not.toContain("ollama");
});
});