Open-source AI Security Operations Center: alert fusion, LLM-agent triage, MITRE ATT&CK investigation, and a replayable decision ledger for every agent step. Self-hostable, runs with no API keys, MIT licensed. Ships an MCP server for Claude, Cursor and Continue.
Python agentic-aiai-securityalert-triageblue-teamcybersecuritydetection-engineeringincident-responselanggraphllm-agentsmcpmitre-attackmodel-context-protocolpurple-teamsecurity-operationsself-hostedsiemsoarsocthreat-detectionthreat-intelligence
/**
* `Alert.confidenceScore` has to mean one thing.
*
* Confidence reaches the console on two keys at two different scales: the API
* surfaces `confidence` as an integer 0-100, and fusion's `confidence_score`
* is the raw [0.0, 1.0] float the band was derived from. `normalizeAlert`
* accepted whichever appeared first and passed it through unchanged, so the
* field's scale depended on the payload — and its consumers disagreed about
* what they were reading. `AlertDetailView` multiplied by 100 and rendered a
* real confidence of 21 as **2100%**; `AttackStory` divided and rendered the
* same value as **21/100**. Each was correct for one payload shape, and each
* had a passing test because its own mock used the scale it assumed.
*
* These assertions are on the boundary rather than on either view, because
* that is the only place the ambiguity can be removed once.
*
* Against the pre-change tree, `normalises fusion's raw float` fails
* (0.21 arrives as 0.21) and `prefers the canonical integer` fails
* (`confidence_score` won).
*/
import { afterEach, beforeEach, describe, expect, it, vi } from 'vitest';
import { alertsApi } from '@/lib/api';
function respondWith(alert: Record<string, unknown>) {
return vi.fn(async () =>
new Response(JSON.stringify({ id: 'a1', title: 't', severity: 'medium', status: 'new', ...alert }), {
status: 200,
headers: { 'Content-Type': 'application/json' },
}),
);
}
beforeEach(() => {
vi.stubGlobal('fetch', respondWith({}));
});
afterEach(() => {
vi.unstubAllGlobals();
});
describe('confidence normalises to one scale at the API boundary', () => {
it('passes the API\u2019s canonical 0-100 integer through unchanged', async () => {
vi.stubGlobal('fetch', respondWith({ confidence: 21, confidence_label: 'low' }));
const alert = await alertsApi.get('a1');
expect(alert.confidenceScore).toBe(21);
});
it("normalises fusion's raw [0,1] float to the same 0-100 scale", async () => {
vi.stubGlobal('fetch', respondWith({ confidence_score: 0.21 }));
const alert = await alertsApi.get('a1');
expect(alert.confidenceScore).toBe(21);
});
it('prefers the canonical integer when a payload carries both keys', async () => {
// A fused alert read back through the API carries both. Taking
// `confidence_score` first meant the row and the detail view could show
// 0.21 and 21 for the same alert.
vi.stubGlobal('fetch', respondWith({ confidence: 21, confidence_score: 0.21 }));
const alert = await alertsApi.get('a1');
expect(alert.confidenceScore).toBe(21);
});
it('decides the scale from the key, never from the magnitude', async () => {
// The tempting shortcut is `v <= 1 ? v * 100 : v`. A genuine confidence
// of 1/100 is then indistinguishable from a raw score of 1.0, and the
// least-confident alert in the estate renders as the most confident.
vi.stubGlobal('fetch', respondWith({ confidence: 1 }));
expect((await alertsApi.get('a1')).confidenceScore).toBe(1);
vi.stubGlobal('fetch', respondWith({ confidence_score: 1.0 }));
expect((await alertsApi.get('a1')).confidenceScore).toBe(100);
});
it('leaves confidence absent when the payload carries neither key', async () => {
// Legacy alerts predate the fusion confidence columns. Absent has to stay
// absent so the views can omit the section rather than print a zero.
vi.stubGlobal('fetch', respondWith({}));
expect((await alertsApi.get('a1')).confidenceScore).toBeUndefined();
});
});