Add dark inference pipeline

This commit is contained in:
dirtydishes 2026-01-04 17:29:21 -05:00
parent 3164167bee
commit ea61c3b013
9 changed files with 699 additions and 1 deletions

View file

@ -5,6 +5,7 @@ import {
EquityPrintSchema,
EquityQuoteSchema,
EquityPrintJoinSchema,
InferredDarkEventSchema,
FlowPacketSchema,
OptionNBBOSchema,
OptionPrintSchema
@ -15,6 +16,7 @@ import type {
EquityPrint,
EquityQuote,
EquityPrintJoin,
InferredDarkEvent,
FlowPacket,
OptionNBBO,
OptionPrint
@ -42,6 +44,13 @@ import {
toEquityPrintJoinRecord,
type EquityPrintJoinRecord
} from "./equity-print-joins";
import {
inferredDarkTableDDL,
INFERRED_DARK_TABLE,
fromInferredDarkRecord,
toInferredDarkRecord,
type InferredDarkRecord
} from "./inferred-dark";
import {
FLOW_PACKETS_TABLE,
flowPacketsTableDDL,
@ -120,6 +129,14 @@ export const ensureEquityPrintJoinsTable = async (
});
};
export const ensureInferredDarkTable = async (
client: ClickHouseClient
): Promise<void> => {
await client.exec({
query: inferredDarkTableDDL()
});
};
export const ensureFlowPacketsTable = async (
client: ClickHouseClient
): Promise<void> => {
@ -202,6 +219,18 @@ export const insertEquityPrintJoin = async (
});
};
export const insertInferredDark = async (
client: ClickHouseClient,
event: InferredDarkEvent
): Promise<void> => {
const record = toInferredDarkRecord(event);
await client.insert({
table: INFERRED_DARK_TABLE,
values: [record],
format: "JSONEachRow"
});
};
export const insertFlowPacket = async (
client: ClickHouseClient,
packet: FlowPacket
@ -367,6 +396,23 @@ const normalizeEquityPrintJoinRow = (row: unknown): EquityPrintJoinRecord | null
};
};
const normalizeInferredDarkRow = (row: unknown): InferredDarkRecord | 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 ?? ""),
type: String(record.type ?? ""),
confidence: Number(coerceNumber(record.confidence) ?? 0),
evidence_refs_json: String(record.evidence_refs_json ?? "[]")
};
};
const normalizeFlowPacketRow = (row: unknown): FlowPacketRecord | null => {
if (!row || typeof row !== "object") {
return null;
@ -497,6 +543,24 @@ export const fetchRecentEquityPrintJoins = async (
return EquityPrintJoinSchema.array().parse(joins);
};
export const fetchRecentInferredDark = async (
client: ClickHouseClient,
limit: number
): Promise<InferredDarkEvent[]> => {
const safeLimit = clampLimit(limit);
const result = await client.query({
query: `SELECT * FROM ${INFERRED_DARK_TABLE} ORDER BY source_ts DESC, seq DESC LIMIT ${safeLimit}`,
format: "JSONEachRow"
});
const rows = await result.json<unknown[]>();
const records = rows
.map(normalizeInferredDarkRow)
.filter((record): record is InferredDarkRecord => record !== null);
const events = records.map(fromInferredDarkRecord);
return InferredDarkEventSchema.array().parse(events);
};
export const fetchRecentFlowPackets = async (
client: ClickHouseClient,
limit: number
@ -649,3 +713,26 @@ export const fetchEquityPrintJoinsAfter = async (
const joins = records.map(fromEquityPrintJoinRecord);
return EquityPrintJoinSchema.array().parse(joins);
};
export const fetchInferredDarkAfter = async (
client: ClickHouseClient,
afterTs: number,
afterSeq: number,
limit: number
): Promise<InferredDarkEvent[]> => {
const safeLimit = clampLimit(limit);
const safeAfterTs = clampCursor(afterTs);
const safeAfterSeq = clampCursor(afterSeq);
const result = await client.query({
query: `SELECT * FROM ${INFERRED_DARK_TABLE} WHERE (source_ts, seq) > (${safeAfterTs}, ${safeAfterSeq}) ORDER BY source_ts ASC, seq ASC LIMIT ${safeLimit}`,
format: "JSONEachRow"
});
const rows = await result.json<unknown[]>();
const records = rows
.map(normalizeInferredDarkRow)
.filter((record): record is InferredDarkRecord => record !== null);
const events = records.map(fromInferredDarkRecord);
return InferredDarkEventSchema.array().parse(events);
};

View file

@ -5,5 +5,6 @@ export * from "./flow-packets";
export * from "./equity-prints";
export * from "./equity-quotes";
export * from "./equity-print-joins";
export * from "./inferred-dark";
export * from "./option-prints";
export * from "./option-nbbo";

View file

@ -0,0 +1,66 @@
import type { InferredDarkEvent } from "@islandflow/types";
export const INFERRED_DARK_TABLE = "inferred_dark";
export type InferredDarkRecord = {
source_ts: number;
ingest_ts: number;
seq: number;
trace_id: string;
type: string;
confidence: number;
evidence_refs_json: string;
};
export const inferredDarkTableDDL = (): string => {
return `
CREATE TABLE IF NOT EXISTS ${INFERRED_DARK_TABLE} (
source_ts UInt64,
ingest_ts UInt64,
seq UInt64,
trace_id String,
type String,
confidence Float64,
evidence_refs_json String
)
ENGINE = MergeTree
ORDER BY (source_ts, seq)
`;
};
export const toInferredDarkRecord = (event: InferredDarkEvent): InferredDarkRecord => {
return {
source_ts: event.source_ts,
ingest_ts: event.ingest_ts,
seq: event.seq,
trace_id: event.trace_id,
type: event.type,
confidence: event.confidence,
evidence_refs_json: JSON.stringify(event.evidence_refs)
};
};
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 fromInferredDarkRecord = (record: InferredDarkRecord): InferredDarkEvent => {
return {
source_ts: record.source_ts,
ingest_ts: record.ingest_ts,
seq: record.seq,
trace_id: record.trace_id,
type: record.type,
confidence: record.confidence,
evidence_refs: safeStringArray(record.evidence_refs_json)
};
};