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

feat: add one-turn situation graph update UI

Browse Source
This commit is contained in:
robbond
2026-08-02 09:43:42 +01:00
parent a948910ba8
commit a9bce79658
11 changed files with 901 additions and 42 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tests/graph/apply-proposal.test.js
+54
View File
@@ -280,6 +280,60 @@ describe("applyValidatedProposal", () => {
expect(result.updatedSituationGraph.resolvedNodeIds).toContain(
ids.complaintRateUnknown,
);
expect(
result.updatedSituationGraph.nodes.find(
(node) => node.id === ids.complaintRateUnknown,
)?.status,
).toBe("resolved");
});
it("rejects resolvedUnknownNodeIds that do not reference actual unknown nodes", () => {
const { graph, proposal, ids } = makeApplicationFixture();
const result = applyValidatedProposal({
situationGraph: graph,
proposal: {
...proposal,
resolvedUnknownNodeIds: [ids.qualityDeterioration],
},
});
expect(result.success).toBe(false);
expect(result.stage).toBe("proposal_compatibility");
expect(result.errors.join(" ")).toContain(
"Resolved unknown must reference an existing unknown node",
);
});
it("rejects a duplicate semantic node without resolution", () => {
const { graph, proposal } = makeApplicationFixture();
const result = applyValidatedProposal({
situationGraph: graph,
proposal: {
...proposal,
resolvedUnknownNodeIds: [],
updatedNodes: proposal.updatedNodes.filter(
(update) => update.nodeId !== "n-complaint-rate-unknown",
),
addedNodes: [
makeNode({
id: "n-parallel-rate",
label: "Complaint rate",
description: "Need the complaint rate per 100 units",
kind: "observation",
status: "supported",
confidence: "medium",
}),
],
},
});
expect(result.success).toBe(false);
expect(result.stage).toBe("proposal_compatibility");
expect(result.errors.join(" ")).toContain(
"duplicating unresolved unknown meaning",
);
});
it("keeps the active unknown when it remains unresolved", () => {
tests/graph/orchestrator.test.js
+7 -6
View File
@@ -13,9 +13,9 @@ function makeAnalysisResult(overrides = {}) {
return {
success: true,
validationStatus: "valid",
modelName: "llama3",
modelName: "configured-model",
responseDurationMs: 321,
rawResponse: "{}",
rawResponse: undefined,
promptVersion: "v0.3",
reconstruction: {
summary: "Revenue and complaints diverge",
@@ -164,7 +164,7 @@ describe("lib/graph/orchestrator startCase", () => {
expect(result.situationGraph.currentSummary).toContain("Nodes:");
expect(result.diagnostics).toMatchObject({
validationStatus: "valid",
modelName: "llama3",
modelName: "configured-model",
graphReferenceValidation: { valid: true, errors: [] },
});
});
@@ -208,7 +208,7 @@ describe("lib/graph/orchestrator startCase", () => {
error: "Provider unavailable",
errors: ["socket hang up"],
rawResponse: null,
modelName: "llama3",
modelName: "configured-model",
responseDurationMs: 99,
promptVersion: "v0.3",
validationStatus: "invalid",
@@ -279,8 +279,9 @@ describe("lib/graph/orchestrator startCase", () => {
diagnostics: {
promptVersion: "v0.4",
modelName: "configured",
graphNodeCount: 2,
graphEdgeCount: 0,
nodeCount: 2,
edgeCount: 0,
validationStatus: "valid",
},
});
expect(provider.generateReconstruction).toHaveBeenCalledTimes(1);
tests/smoke.test.js
+43 -12
View File
@@ -4,7 +4,7 @@ const BASE_URL = process.env.PLAYWRIGHT_BASE_URL || "http://localhost:3000";
test.setTimeout(300000);
test("graph-backed start flow smoke test", async ({ page }) => {
test("graph-backed one-turn update smoke test", async ({ page }) => {
await page.goto(BASE_URL);
// Page should load without error
@@ -26,7 +26,11 @@ test("graph-backed start flow smoke test", async ({ page }) => {
await page.getByRole("button", { name: /Analyse/i }).click();
await expect(
page.getByRole("heading", { name: /Selected Question/i }),
page
.locator("section")
.filter({ hasText: /Selected Question/i })
.last()
.getByRole("heading", { name: /Selected Question/i }),
).toBeVisible({ timeout: 180000 });
await expect(
page.getByRole("heading", { name: /Situation Graph/i }),
@@ -40,13 +44,43 @@ test("graph-backed start flow smoke test", async ({ page }) => {
await rawJsonToggle.click();
await expect(page.getByText(/centralStatement/i)).toBeVisible();
const selectedQuestionSections = page
.locator("section")
.filter({ hasText: "Selected Question" });
await expect(selectedQuestionSections).toHaveCount(1);
const questionText = await selectedQuestionSections.first().innerText();
expect(questionText.length).toBeGreaterThan(25);
const answerTextarea = page.locator(
"textarea[placeholder*='Enter the answer']",
);
await expect(answerTextarea).toBeVisible();
await answerTextarea.fill(
"The complaint rate fell from 2.0 complaints per 100 units to 1.9 complaints per 100 units.",
);
await page.getByRole("button", { name: /Update situation/i }).click();
await expect(page.getByText(/Graph update applied/i)).toBeVisible({
timeout: 240000,
});
await expect(page.getByText(/Resolved unknowns/i)).toBeVisible();
await expect(page.getByText(/Affected nodes/i)).toBeVisible();
await expect(
page.getByText(/No next question selected yet\./i),
).toBeVisible();
await expect(page.getByText(/Error:/i)).toHaveCount(0);
await expect(page.getByText(/Update error:/i)).toHaveCount(0);
await expect(answerTextarea).toHaveValue("");
const proposalToggle = page.getByText(/Proposal details/i);
await expect(proposalToggle).toBeVisible();
await rawJsonToggle.click();
await expect(page.getByText(/resolvedNodeIds/i)).toBeVisible();
// Get full body text for verification
const bodyText = await page.locator("body").innerText();
console.log("\n=== UI Smoke Test Results ===");
console.log("Page title:", await page.title());
console.log("Body content length:", bodyText.length);
// Check key content indicators
const hasSelectedQuestion = bodyText.includes("Selected Question");
const hasComplaints =
@@ -59,12 +93,9 @@ test("graph-backed start flow smoke test", async ({ page }) => {
bodyText.toLowerCase().includes("denominator") ||
bodyText.toLowerCase().includes("per-unit");
console.log("Has selected question heading:", hasSelectedQuestion);
console.log("Has complaints reference:", hasComplaints);
console.log("Has production reference:", hasProduction);
console.log("Has rate context (rate/unit/denominator):", hasRateContext);
// Basic structural checks
expect(bodyText.length).toBeGreaterThan(200);
expect(bodyText.length).toBeGreaterThan(400);
expect(hasSelectedQuestion).toBe(true);
expect(bodyText.includes("Resolved unknowns")).toBe(true);
expect(bodyText.includes("Affected nodes")).toBe(true);
});
tests/ui/scenario-form.test.jsx
+238
View File
@@ -2,9 +2,12 @@ import React from "react";
import { describe, expect, it, vi } from "vitest";
import { renderToStaticMarkup } from "react-dom/server";
import DiagnosticsView from "@/components/diagnostics-view.jsx";
import GraphUpdateView from "@/components/graph-update-view.jsx";
import SituationGraphView from "@/components/situation-graph-view.jsx";
import {
ScenarioResultPanels,
UpdateErrorPanel,
submitAnswerForUpdateCase,
submitScenarioForStartCase,
} from "@/components/scenario-form.jsx";
@@ -70,6 +73,73 @@ function makeGraphResult(overrides = {}) {
};
}
function makeUpdateSuccess(overrides = {}) {
return {
success: true,
stage: "update_applied",
updatedSituationGraph: {
centralStatement: "Complaints increased while production increased.",
currentSummary: "Updated summary",
activeUnknownNodeId: "n-next-unknown",
resolvedNodeIds: ["n-unknown"],
nodes: [
{
id: "n-1",
label: "Complaints up 35%",
description: "Complaints increased by 35%",
kind: "observation",
status: "supported",
confidence: "high",
value: 35,
unit: "%",
},
{
id: "n-conclusion",
label: "Quality deterioration",
description: "Quality deterioration conclusion",
kind: "conclusion",
status: "weakened",
confidence: "medium",
value: null,
unit: null,
},
{
id: "n-unknown",
label: "Complaint rate denominator",
description: "Need the denominator for complaint rate",
kind: "unknown",
status: "resolved",
confidence: "medium",
value: "1.9 complaints per 100 units",
unit: null,
},
],
edges: [],
},
proposal: {
addedNodes: [],
updatedNodes: [
{ nodeId: "n-unknown", newStatus: "resolved", reason: "answered" },
],
addedEdges: [],
removedEdgeIds: [],
resolvedUnknownNodeIds: ["n-unknown"],
affectedNodeIds: ["n-conclusion"],
},
affectedNodeIds: ["n-conclusion"],
resolvedUnknownNodeIds: ["n-unknown"],
previousActiveUnknownNodeId: "n-unknown",
newActiveUnknownNodeId: "n-next-unknown",
changesApplied: {
updatedNodeCount: 2,
resolvedUnknownCount: 1,
affectedNodeCount: 1,
},
diagnostics: { responseDurationMs: 100 },
...overrides,
};
}
describe("scenario-form UI helpers", () => {
it("submits to /api/cases/start", async () => {
const fetchImpl = vi.fn().mockResolvedValue({ ok: true });
@@ -84,6 +154,46 @@ describe("scenario-form UI helpers", () => {
}),
);
});
it("empty answer is rejected without fetch", async () => {
const fetchImpl = vi.fn();
const result = await submitAnswerForUpdateCase(fetchImpl, {
situationGraph: { nodes: [] },
previousQuestion: "What changed?",
answer: " ",
});
expect(result.skipped).toBe(true);
expect(fetchImpl).not.toHaveBeenCalled();
});
it("update request body contains graph, previousQuestion and answer", async () => {
const fetchImpl = vi.fn().mockResolvedValue({
ok: true,
json: async () => ({ success: true }),
});
const graph = { nodes: [{ id: "n1" }], edges: [] };
await submitAnswerForUpdateCase(fetchImpl, {
situationGraph: graph,
previousQuestion: "What changed?",
answer: "The rate fell.",
});
expect(fetchImpl).toHaveBeenCalledWith(
"/api/cases/update",
expect.objectContaining({
method: "POST",
headers: { "Content-Type": "application/json" },
body: JSON.stringify({
situationGraph: graph,
previousQuestion: "What changed?",
answer: "The rate fell.",
}),
}),
);
});
});
describe("graph-backed UI rendering", () => {
@@ -128,6 +238,17 @@ describe("graph-backed UI rendering", () => {
).toHaveLength(1);
});
it("no answer form appears when selectedQuestion is null", () => {
const html = renderToStaticMarkup(
<SituationGraphView
situationGraph={makeGraphResult().situationGraph}
selectedQuestion={null}
/>,
);
expect(html).not.toContain("Update situation");
});
it("renders diagnostics", () => {
const html = renderToStaticMarkup(
<DiagnosticsView result={makeGraphResult()} />,
@@ -187,4 +308,121 @@ describe("graph-backed UI rendering", () => {
expect(html).toContain("Raw graph JSON");
expect(html).toContain("&quot;centralStatement&quot;");
});
it("resolved unknowns render", () => {
const html = renderToStaticMarkup(
<GraphUpdateView
updateResult={{
...makeUpdateSuccess(),
previousSituationGraph: makeGraphResult().situationGraph,
}}
/>,
);
expect(html).toContain("Resolved unknowns");
expect(html).toContain("Complaint rate denominator");
});
it("affected nodes render", () => {
const html = renderToStaticMarkup(
<GraphUpdateView
updateResult={{
...makeUpdateSuccess(),
previousSituationGraph: makeGraphResult().situationGraph,
}}
/>,
);
expect(html).toContain("Affected nodes");
expect(html).toContain("Quality deterioration");
});
it("no fake next question appears", () => {
const html = renderToStaticMarkup(
<GraphUpdateView
updateResult={{
...makeUpdateSuccess({ newActiveUnknownNodeId: null }),
previousSituationGraph: makeGraphResult().situationGraph,
}}
/>,
);
expect(html).toContain("No next question selected yet.");
});
it("previous and new active unknowns render labels", () => {
const html = renderToStaticMarkup(
<GraphUpdateView
updateResult={{
...makeUpdateSuccess(),
previousSituationGraph: makeGraphResult().situationGraph,
}}
/>,
);
expect(html).toContain("Previous active unknown");
expect(html).toContain("Complaint rate denominator");
expect(html).toContain("New active unknown");
expect(html).toContain("Unknown node (ID: n-next-unknown)");
});
it("raw ids remain only in collapsed proposal details", () => {
const html = renderToStaticMarkup(
<GraphUpdateView
updateResult={{
...makeUpdateSuccess(),
previousSituationGraph: makeGraphResult().situationGraph,
}}
/>,
);
expect(html).toContain("Proposal details");
expect(html).toContain("&quot;resolvedUnknownNodeIds&quot;");
});
it("update diagnostics render valid values", () => {
const html = renderToStaticMarkup(
<DiagnosticsView
result={{
diagnostics: {
promptVersion: "v0.4",
modelName: "configured-model",
responseDurationMs: 456,
validationStatus: "valid",
nodeCount: 7,
edgeCount: 3,
graphReferenceValidation: { valid: true, errors: [] },
},
}}
/>,
);
expect(html).toContain("v0.4");
expect(html).toContain("456ms");
expect(html).toContain("7");
expect(html).toContain("3");
expect(html).toContain("✅ valid");
});
it("structured update error renders", () => {
const html = renderToStaticMarkup(
<UpdateErrorPanel
updateError={{
error: "Invalid graph update proposal",
proposalErrors: [{ message: "bad proposal" }],
}}
/>,
);
expect(html).toContain("Update error: Invalid graph update proposal");
expect(html).toContain("bad proposal");
});
it("proposal details remain collapsible", () => {
const html = renderToStaticMarkup(
<GraphUpdateView updateResult={makeUpdateSuccess()} />,
);
expect(html).toContain("Proposal details");
});
});