Alexander Hines · Systematic Market-Intelligence Platform

Scattered market feeds in. One reliable, leak-proof decision surface out.

An independent, production-grade market-data platform that owns the whole loop — automated multi-vendor ingestion → point-in-time-correct normalization → reliability engineering that fails loud → a governed decision surface where proprietary data cannot leak by construction. The goal is time-to-insight: replace manual refresh with monitored, push-based state. This page is that surface, running live.

Paper only · no live capital / feeds OK As of

The loop — feed to decision, owned end to end

Ingest10+ vendors: filings, macro, news-sentiment, prices, alt-data
Normalize · point-in-timebitemporal contracts, leak-tested; no hindsight reaches a past decision
Reliability · freshnessreconnect-hardened, resumable, staleness-guarded — silence is a failure
Decision surfacethis dashboard — monitored, human-in-the-loop, leak-proof at publish

Pipeline scale — verified, point-in-time, multi-vendor

4.31B

rows ingested with 0-delta vs source (reconnect-hardened, 139 GB)

291.4M

intraday auction prints, field-validated (2003–2026)

867

point-in-time data contracts, leak-tested in CI

35,021

catalogued data-source families (115,409 relations mapped)

1,710

datasets in the master inventory, freshness-tracked

Pipelines · reliability · security

Own the stack, fail loud, don't leak

Memory-safe, resumable, reconnect-hardened ingestion; a self-generating coverage catalogue; and a redaction chokepoint where proprietary data cannot leak by construction.

Explore the infrastructure →

Execution · measurement

Measure the truth, not the intent

Fills reconciled field-by-field against the official market print; a +116 bp routing defect diagnosed from real data — the discipline of measuring what actually happened.

See the measurement →

Signal vs noise

Surface what matters; kill the rest

Falsification-first research: regime and anomaly detection, event→exposure mapping, and the discipline to suspend contaminated findings rather than ship vanity metrics.

Read the discipline →

The decision surface, monitored live

Feeds healthy

/freshness + staleness monitored per source

Paper NAV

inception · no live capital

Publish integrity

redactedIDs masked · secrets stripped · refuses if live

Refresh

hourlypush-based, not manual F5

The research the pipeline feeds — held to an honest bar

Generated
1.66Sharpe

Candidate book (26.1% CAGR) before modeled costs — the ceiling the search produced.

Audited → suspended

A look-ahead leak was found in the dominant sleeve; the headline was suspended, not adjusted.

Certified floor
1.40 → 1.05

Pre-registered book: ~1.40 before / 1.05 after modeled costs (1.07 holdout).

Forward validation

The stronger candidate is under live paper test. Execution cost is the binding uncertainty.

Sharpe figures are annualized, hypothetical/backtested, before vs after conservative modeled implementation costs. No live-capital track record is claimed. The point isn't the number — it's the discipline: strong generation, adversarial gating, and suspension of anything that doesn't survive.

Paper trading only. No real capital is deployed. The public surface is filtered by a governed redaction chokepoint (dashboard_publish.py): broker account IDs, order IDs, model scores, and secrets are denied at publish time — proprietary data cannot leak by construction.