Add classifier hits and alerts pipeline

Add NATS subjects + ClickHouse tables for classifier hits/alerts, evaluate sweep/spike rules in compute, expose API/WS endpoints, and cover storage helpers with tests.
This commit is contained in:
dirtydishes 2025-12-29 15:59:37 -05:00
parent ad58c62c37
commit 58485b4d97
11 changed files with 861 additions and 8 deletions

View file

@ -4,3 +4,7 @@ export const STREAM_EQUITY_PRINTS = "EQUITY_PRINTS";
export const SUBJECT_EQUITY_PRINTS = "equities.prints";
export const STREAM_FLOW_PACKETS = "FLOW_PACKETS";
export const SUBJECT_FLOW_PACKETS = "flow.packets";
export const STREAM_CLASSIFIER_HITS = "CLASSIFIER_HITS";
export const SUBJECT_CLASSIFIER_HITS = "flow.classifier_hits";
export const STREAM_ALERTS = "ALERTS";
export const SUBJECT_ALERTS = "flow.alerts";

View file

@ -0,0 +1,93 @@
import type { AlertEvent, ClassifierHit } from "@islandflow/types";
export const ALERTS_TABLE = "alerts";
export type AlertRecord = {
source_ts: number;
ingest_ts: number;
seq: number;
trace_id: string;
score: number;
severity: string;
hits_json: string;
evidence_refs_json: string;
};
export const alertsTableDDL = (): string => {
return `
CREATE TABLE IF NOT EXISTS ${ALERTS_TABLE} (
source_ts UInt64,
ingest_ts UInt64,
seq UInt64,
trace_id String,
score Float64,
severity String,
hits_json String,
evidence_refs_json String
)
ENGINE = MergeTree
ORDER BY (source_ts, seq)
`;
};
export const toAlertRecord = (alert: AlertEvent): AlertRecord => {
return {
source_ts: alert.source_ts,
ingest_ts: alert.ingest_ts,
seq: alert.seq,
trace_id: alert.trace_id,
score: alert.score,
severity: alert.severity,
hits_json: JSON.stringify(alert.hits),
evidence_refs_json: JSON.stringify(alert.evidence_refs)
};
};
const safeHitArray = (value: string): ClassifierHit[] => {
try {
const parsed = JSON.parse(value);
if (Array.isArray(parsed)) {
return parsed.map((entry) => {
const record = entry as Partial<ClassifierHit>;
return {
classifier_id: String(record.classifier_id ?? ""),
confidence: Number(record.confidence ?? 0),
direction: String(record.direction ?? ""),
explanations: Array.isArray(record.explanations)
? record.explanations.map((item) => String(item))
: []
};
});
}
} catch {
// ignore
}
return [];
};
const safeStringArray = (value: string): string[] => {
try {
const parsed = JSON.parse(value);
if (Array.isArray(parsed)) {
return parsed.map((entry) => String(entry));
}
} catch {
// ignore
}
return [];
};
export const fromAlertRecord = (record: AlertRecord): AlertEvent => {
return {
source_ts: record.source_ts,
ingest_ts: record.ingest_ts,
seq: record.seq,
trace_id: record.trace_id,
score: record.score,
severity: record.severity,
hits: safeHitArray(record.hits_json),
evidence_refs: safeStringArray(record.evidence_refs_json)
};
};

View file

@ -0,0 +1,70 @@
import type { ClassifierHitEvent } from "@islandflow/types";
export const CLASSIFIER_HITS_TABLE = "classifier_hits";
export type ClassifierHitRecord = {
source_ts: number;
ingest_ts: number;
seq: number;
trace_id: string;
classifier_id: string;
confidence: number;
direction: string;
explanations_json: string;
};
export const classifierHitsTableDDL = (): string => {
return `
CREATE TABLE IF NOT EXISTS ${CLASSIFIER_HITS_TABLE} (
source_ts UInt64,
ingest_ts UInt64,
seq UInt64,
trace_id String,
classifier_id String,
confidence Float64,
direction String,
explanations_json String
)
ENGINE = MergeTree
ORDER BY (source_ts, seq)
`;
};
export const toClassifierHitRecord = (hit: ClassifierHitEvent): ClassifierHitRecord => {
return {
source_ts: hit.source_ts,
ingest_ts: hit.ingest_ts,
seq: hit.seq,
trace_id: hit.trace_id,
classifier_id: hit.classifier_id,
confidence: hit.confidence,
direction: hit.direction,
explanations_json: JSON.stringify(hit.explanations)
};
};
const safeJsonArray = (value: string): string[] => {
try {
const parsed = JSON.parse(value);
if (Array.isArray(parsed)) {
return parsed.map((entry) => String(entry));
}
} catch {
// ignore
}
return [];
};
export const fromClassifierHitRecord = (record: ClassifierHitRecord): ClassifierHitEvent => {
return {
source_ts: record.source_ts,
ingest_ts: record.ingest_ts,
seq: record.seq,
trace_id: record.trace_id,
classifier_id: record.classifier_id,
confidence: record.confidence,
direction: record.direction,
explanations: safeJsonArray(record.explanations_json)
};
};

View file

@ -1,6 +1,18 @@
import { createClient, type ClickHouseClient } from "@clickhouse/client";
import { EquityPrintSchema, FlowPacketSchema, OptionPrintSchema } from "@islandflow/types";
import type { EquityPrint, FlowPacket, OptionPrint } from "@islandflow/types";
import {
AlertEventSchema,
ClassifierHitEventSchema,
EquityPrintSchema,
FlowPacketSchema,
OptionPrintSchema
} from "@islandflow/types";
import type {
AlertEvent,
ClassifierHitEvent,
EquityPrint,
FlowPacket,
OptionPrint
} from "@islandflow/types";
import {
normalizeOptionPrint,
optionPrintsTableDDL,
@ -18,6 +30,20 @@ import {
toFlowPacketRecord,
type FlowPacketRecord
} from "./flow-packets";
import {
CLASSIFIER_HITS_TABLE,
classifierHitsTableDDL,
fromClassifierHitRecord,
toClassifierHitRecord,
type ClassifierHitRecord
} from "./classifier-hits";
import {
ALERTS_TABLE,
alertsTableDDL,
fromAlertRecord,
toAlertRecord,
type AlertRecord
} from "./alerts";
export type ClickHouseOptions = {
url: string;
@ -59,6 +85,20 @@ export const ensureFlowPacketsTable = async (
});
};
export const ensureClassifierHitsTable = async (
client: ClickHouseClient
): Promise<void> => {
await client.exec({
query: classifierHitsTableDDL()
});
};
export const ensureAlertsTable = async (client: ClickHouseClient): Promise<void> => {
await client.exec({
query: alertsTableDDL()
});
};
export const insertOptionPrint = async (
client: ClickHouseClient,
print: OptionPrint
@ -95,6 +135,27 @@ export const insertFlowPacket = async (
});
};
export const insertClassifierHit = async (
client: ClickHouseClient,
hit: ClassifierHitEvent
): Promise<void> => {
const record = toClassifierHitRecord(hit);
await client.insert({
table: CLASSIFIER_HITS_TABLE,
values: [record],
format: "JSONEachRow"
});
};
export const insertAlert = async (client: ClickHouseClient, alert: AlertEvent): Promise<void> => {
const record = toAlertRecord(alert);
await client.insert({
table: ALERTS_TABLE,
values: [record],
format: "JSONEachRow"
});
};
const clampLimit = (limit: number): number => {
if (!Number.isFinite(limit)) {
return 100;
@ -196,6 +257,42 @@ const normalizeFlowPacketRow = (row: unknown): FlowPacketRecord | null => {
};
};
const normalizeClassifierHitRow = (row: unknown): ClassifierHitRecord | null => {
if (!row || typeof row !== "object") {
return null;
}
const record = row as Record<string, unknown>;
return {
source_ts: coerceNumber(record.source_ts) as number,
ingest_ts: coerceNumber(record.ingest_ts) as number,
seq: coerceNumber(record.seq) as number,
trace_id: String(record.trace_id ?? ""),
classifier_id: String(record.classifier_id ?? ""),
confidence: Number(coerceNumber(record.confidence) ?? 0),
direction: String(record.direction ?? ""),
explanations_json: String(record.explanations_json ?? "[]")
};
};
const normalizeAlertRow = (row: unknown): AlertRecord | null => {
if (!row || typeof row !== "object") {
return null;
}
const record = row as Record<string, unknown>;
return {
source_ts: coerceNumber(record.source_ts) as number,
ingest_ts: coerceNumber(record.ingest_ts) as number,
seq: coerceNumber(record.seq) as number,
trace_id: String(record.trace_id ?? ""),
score: Number(coerceNumber(record.score) ?? 0),
severity: String(record.severity ?? ""),
hits_json: String(record.hits_json ?? "[]"),
evidence_refs_json: String(record.evidence_refs_json ?? "[]")
};
};
export const fetchRecentOptionPrints = async (
client: ClickHouseClient,
limit: number
@ -242,6 +339,42 @@ export const fetchRecentFlowPackets = async (
return FlowPacketSchema.array().parse(packets);
};
export const fetchRecentClassifierHits = async (
client: ClickHouseClient,
limit: number
): Promise<ClassifierHitEvent[]> => {
const safeLimit = clampLimit(limit);
const result = await client.query({
query: `SELECT * FROM ${CLASSIFIER_HITS_TABLE} ORDER BY source_ts DESC, seq DESC LIMIT ${safeLimit}`,
format: "JSONEachRow"
});
const rows = await result.json<unknown[]>();
const records = rows
.map(normalizeClassifierHitRow)
.filter((record): record is ClassifierHitRecord => record !== null);
const hits = records.map(fromClassifierHitRecord);
return ClassifierHitEventSchema.array().parse(hits);
};
export const fetchRecentAlerts = async (
client: ClickHouseClient,
limit: number
): Promise<AlertEvent[]> => {
const safeLimit = clampLimit(limit);
const result = await client.query({
query: `SELECT * FROM ${ALERTS_TABLE} ORDER BY source_ts DESC, seq DESC LIMIT ${safeLimit}`,
format: "JSONEachRow"
});
const rows = await result.json<unknown[]>();
const records = rows
.map(normalizeAlertRow)
.filter((record): record is AlertRecord => record !== null);
const alerts = records.map(fromAlertRecord);
return AlertEventSchema.array().parse(alerts);
};
export const fetchOptionPrintsAfter = async (
client: ClickHouseClient,
afterTs: number,

View file

@ -1,4 +1,6 @@
export * from "./clickhouse";
export * from "./classifier-hits";
export * from "./alerts";
export * from "./flow-packets";
export * from "./equity-prints";
export * from "./option-prints";

View file

@ -0,0 +1,36 @@
import { describe, expect, it } from "bun:test";
import { alertsTableDDL, ALERTS_TABLE, fromAlertRecord, toAlertRecord } from "../src/alerts";
const alert = {
source_ts: 10,
ingest_ts: 20,
seq: 1,
trace_id: "alert:fp-1",
score: 78,
severity: "medium",
hits: [
{
classifier_id: "large_bullish_call_sweep",
confidence: 0.72,
direction: "bullish",
explanations: ["Likely call sweep.", "Premium $50000."]
}
],
evidence_refs: ["flowpacket:1", "print:1"]
};
describe("alerts storage helpers", () => {
it("includes the correct table name in the DDL", () => {
const ddl = alertsTableDDL();
expect(ddl).toContain(ALERTS_TABLE);
expect(ddl).toContain("CREATE TABLE IF NOT EXISTS");
});
it("round-trips alert records", () => {
const record = toAlertRecord(alert);
const restored = fromAlertRecord(record);
expect(restored.hits).toEqual(alert.hits);
expect(restored.evidence_refs).toEqual(alert.evidence_refs);
expect(restored.severity).toBe(alert.severity);
});
});

View file

@ -0,0 +1,34 @@
import { describe, expect, it } from "bun:test";
import {
classifierHitsTableDDL,
CLASSIFIER_HITS_TABLE,
fromClassifierHitRecord,
toClassifierHitRecord
} from "../src/classifier-hits";
const hit = {
source_ts: 10,
ingest_ts: 20,
seq: 1,
trace_id: "classifier:large_bullish_call_sweep:fp-1",
classifier_id: "large_bullish_call_sweep",
confidence: 0.72,
direction: "bullish",
explanations: ["Likely call sweep.", "Premium $50000."]
};
describe("classifier hits storage helpers", () => {
it("includes the correct table name in the DDL", () => {
const ddl = classifierHitsTableDDL();
expect(ddl).toContain(CLASSIFIER_HITS_TABLE);
expect(ddl).toContain("CREATE TABLE IF NOT EXISTS");
});
it("round-trips classifier hit records", () => {
const record = toClassifierHitRecord(hit);
const restored = fromClassifierHitRecord(record);
expect(restored.explanations).toEqual(hit.explanations);
expect(restored.classifier_id).toBe(hit.classifier_id);
expect(restored.direction).toBe(hit.direction);
});
});