robbond/confidence-engine
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

feat: advance reasoning after comparability is resolved

Browse Source
This commit is contained in:
robbond
2026-08-02 17:06:34 +01:00
parent 7d408701b5
commit 25a989450c
9 changed files with 649 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tests/graph/apply-proposal.test.js
+174
View File
@@ -3,6 +3,120 @@ import { applyValidatedProposal } from "@/lib/graph/apply-proposal.js";
import { makeEdge, makeGraph, makeNode } from "@/lib/graph/schema.js";
import { validateGraphReferences } from "@/lib/graph/utils.js";
function makeComparabilityUpdateFixture() {
const comparabilityUnknown = makeNode({
id: "n-comparability-unknown",
label: "Whether the figures are comparable",
description:
"Need to know whether the figures use the same period, basis, and scale before comparing them.",
kind: "unknown",
status: "unknown",
confidence: "high",
});
const revenueObservation = makeNode({
id: "n-revenue-observation",
label: "Revenue increased by 18%.",
description: "Revenue increased by 18%.",
kind: "observation",
status: "supported",
confidence: "high",
});
const cashObservation = makeNode({
id: "n-cash-observation",
label: "Cash in the bank decreased over the same period.",
description: "Cash in the bank decreased over the same period.",
kind: "observation",
status: "supported",
confidence: "high",
});
const unrelatedNode = makeNode({
id: "n-unrelated",
label: "Board update",
description: "A separate unchanged note.",
kind: "state",
status: "known",
confidence: "low",
});
const graph = makeGraph({
centralStatement:
"Revenue increased by 18%, but cash in the bank fell over the same period.",
nodes: [
comparabilityUnknown,
revenueObservation,
cashObservation,
unrelatedNode,
],
edges: [
makeEdge({
id: "e-revenue-comparability",
fromNodeId: revenueObservation.id,
toNodeId: comparabilityUnknown.id,
relationship: "supports",
confidence: "medium",
description: "Revenue observation requires comparability confirmation.",
}),
makeEdge({
id: "e-cash-comparability",
fromNodeId: cashObservation.id,
toNodeId: comparabilityUnknown.id,
relationship: "supports",
confidence: "medium",
description: "Cash observation requires comparability confirmation.",
}),
],
activeUnknownNodeId: comparabilityUnknown.id,
resolvedNodeIds: [],
currentSummary: "Initial comparability fixture",
reasoningState: {
comparabilityStatus: "uncertain",
comparabilityReason:
"Comparability between the observations is not yet established across period, scale, or measurement basis.",
comparabilityEvidence: [],
relationshipStatus: "insufficient_information",
relationshipReason:
"Relationship classification is deferred until comparability is established.",
relationshipAssessed: false,
contradictionReasoningAllowed: false,
reasoningStages: [
{
stage: "comparability",
status: "uncertain",
outcome:
"Comparability between the observations is not yet established across period, scale, or measurement basis.",
},
{
stage: "relationship",
status: "insufficient_information",
outcome: "not assessed until comparability is established",
},
],
},
});
const proposal = {
addedNodes: [],
updatedNodes: [
{
nodeId: comparabilityUnknown.id,
previousStatus: "unknown",
newStatus: "resolved",
previousValue: null,
newValue:
"Both figures cover the same accounting period and are taken from the same management accounts.",
reason: "The answer confirms the figures are comparable.",
},
],
addedEdges: [],
removedEdgeIds: [],
resolvedUnknownNodeIds: [comparabilityUnknown.id],
affectedNodeIds: [],
selectedQuestion: null,
};
return { graph, proposal, comparabilityUnknownId: comparabilityUnknown.id };
}
function makeApplicationFixture() {
const complaintRateUnknown = makeNode({
id: "n-complaint-rate-unknown",
@@ -997,4 +1111,64 @@ describe("applyValidatedProposal", () => {
"price",
);
});
it("resolves the existing comparability unknown and advances reasoning after the answer", () => {
const { graph, proposal, comparabilityUnknownId } =
makeComparabilityUpdateFixture();
const originalUnrelatedNode = JSON.stringify(
graph.nodes.find((node) => node.id === "n-unrelated"),
);
const result = applyValidatedProposal({
situationGraph: graph,
proposal,
previousQuestion:
"Were these figures measured on the same basis and at the same scale?",
answer:
"Yes. Both figures cover the same accounting period and are taken from the same management accounts.",
});
expect(result.success).toBe(true);
expect(result.resolvedUnknownNodeIds).toContain(comparabilityUnknownId);
expect(result.resolvedReasoningNodeIds).toEqual([
"reasoning:comparability",
]);
expect(result.previousReasoningState.comparabilityStatus).toBe("uncertain");
expect(result.reasoningState).toMatchObject({
comparabilityStatus: "confirmed",
relationshipStatus: "potentially_related",
relationshipAssessed: true,
});
expect(result.reasoningState.comparabilityEvidence).toEqual([
comparabilityUnknownId,
]);
expect(result.selectedQuestion?.question).toMatch(
/^What changed during that period that could help explain why /,
);
expect(result.selectedQuestion?.question).not.toContain("same basis");
expect(result.selectedQuestion?.question.toLowerCase()).not.toMatch(
/dso|debtor days|receivables turnover|working capital|receivables/,
);
expect(result.reasoningState.reasoningStages).toEqual([
{
stage: "comparability",
status: "confirmed",
outcome:
"Comparability was confirmed by the user answer covering the same period and source basis.",
},
{
stage: "relationship",
status: "potentially_related",
outcome:
"The observations concern connected business signals but do not establish a direct contradiction or cause.",
},
]);
expect(
JSON.stringify(
result.updatedSituationGraph.nodes.find(
(node) => node.id === "n-unrelated",
),
),
).toBe(originalUnrelatedNode);
});
});
tests/graph/orchestrator.test.js
+152
View File
@@ -118,6 +118,90 @@ function makeProposal(overrides = {}) {
};
}
function makeComparabilityScenarioGraph() {
return makeGraph({
centralStatement:
"Revenue increased by 18%, but cash in the bank fell over the same period.",
nodes: [
makeNode({
id: "n-comparability-unknown",
label: "Whether the figures are comparable",
description:
"Need to know whether the figures use the same period, basis, and scale before comparing them.",
kind: "unknown",
status: "unknown",
confidence: "high",
}),
makeNode({
id: "n-revenue-observation",
label: "Revenue increased by 18%.",
description: "Revenue increased by 18%.",
kind: "observation",
status: "supported",
confidence: "high",
}),
makeNode({
id: "n-cash-observation",
label: "Cash in the bank decreased over the same period.",
description: "Cash in the bank decreased over the same period.",
kind: "observation",
status: "supported",
confidence: "high",
}),
],
edges: [],
activeUnknownNodeId: "n-comparability-unknown",
resolvedNodeIds: [],
currentSummary: "Comparability scenario",
reasoningState: {
comparabilityStatus: "uncertain",
comparabilityReason:
"Comparability between the observations is not yet established across period, scale, or measurement basis.",
comparabilityEvidence: [],
relationshipStatus: "insufficient_information",
relationshipReason:
"Relationship classification is deferred until comparability is established.",
relationshipAssessed: false,
contradictionReasoningAllowed: false,
reasoningStages: [
{
stage: "comparability",
status: "uncertain",
outcome:
"Comparability between the observations is not yet established across period, scale, or measurement basis.",
},
{
stage: "relationship",
status: "insufficient_information",
outcome: "not assessed until comparability is established",
},
],
},
});
}
function makeComparabilityProposal() {
return {
addedNodes: [],
updatedNodes: [
{
nodeId: "n-comparability-unknown",
previousStatus: "unknown",
newStatus: "resolved",
previousValue: null,
newValue:
"Both figures cover the same accounting period and are taken from the same management accounts.",
reason: "The answer confirms comparability.",
},
],
addedEdges: [],
removedEdgeIds: [],
resolvedUnknownNodeIds: ["n-comparability-unknown"],
affectedNodeIds: [],
selectedQuestion: null,
};
}
describe("lib/graph/orchestrator startCase", () => {
beforeEach(() => {
vi.resetModules();
@@ -930,6 +1014,74 @@ describe("lib/graph/orchestrator startCase", () => {
});
});
it("advances reasoning after comparability is resolved by the update answer", async () => {
const { updateCase } = await import("@/lib/graph/orchestrator.js");
const provider = {
generateReconstruction: vi
.fn()
.mockResolvedValue(makeComparabilityProposal()),
};
const result = await updateCase(
{
situationGraph: makeComparabilityScenarioGraph(),
previousQuestion:
"Were these figures measured on the same basis and at the same scale?",
answer:
"Yes. Both figures cover the same accounting period and are taken from the same management accounts.",
promptVersion: "v0.4",
},
{
provider,
config: MOCK_CONFIG,
applyProposal: true,
},
);
expect(result.success).toBe(true);
expect(result.resolvedUnknownNodeIds).toEqual(["n-comparability-unknown"]);
expect(result.diagnostics).toMatchObject({
previousComparabilityStatus: "uncertain",
comparabilityStatus: "confirmed",
relationshipStatus: "potentially_related",
relationshipAssessed: true,
resolvedReasoningNodeIds: ["reasoning:comparability"],
});
expect(result.diagnostics.reasoningStagesBefore).toEqual([
{
stage: "comparability",
status: "uncertain",
outcome:
"Comparability between the observations is not yet established across period, scale, or measurement basis.",
},
{
stage: "relationship",
status: "insufficient_information",
outcome: "not assessed until comparability is established",
},
]);
expect(result.diagnostics.reasoningStagesAfter).toEqual([
{
stage: "comparability",
status: "confirmed",
outcome:
"Comparability was confirmed by the user answer covering the same period and source basis.",
},
{
stage: "relationship",
status: "potentially_related",
outcome:
"The observations concern connected business signals but do not establish a direct contradiction or cause.",
},
]);
expect(result.selectedQuestion?.question).toMatch(
/^What changed during that period that could help explain why /,
);
expect(result.selectedQuestion?.question.toLowerCase()).not.toMatch(
/same basis|dso|receivables|debtor days|working capital/,
);
});
it("startCase behaviour remains unchanged", async () => {
mockAnalyseScenario.mockResolvedValue(makeAnalysisResult());
const { startCase } = await import("@/lib/graph/orchestrator.js");