Directory view

Static listing intelligence; search and filters live on the Directory tab.

Built from Erika dataset metadata for browsing, filtering, and static profile lookup.

This directory is a static metadata browser for research and educational use. It is not an offer to buy or sell securities, and it does not replace financial, legal, or tax advice.

Directory search

Explore Train'ing world Market data.

A GitHub Pages-ready catalog for 56,322 global listings across stocks and ETFs, with searchable metadata, exchange filters, and date-coverage signals.

Browse the directory Open featured profile Open raw data page Live analysis desk
Latency topology Live pipeline Complete date coverage ETFs in the US NASDAQ listings Healthcare stocks
Coverage snapshot
High-level shape of the Erika listing universe.
Listings 56,322
Countries 80
Exchanges 71
Company dates 4,555
IPO dates 21,236
Complete coverage 3,028
Stock: 40,971 ETF: 15,351

Coverage profile

Same six dimensions as the grid, normalized to a radar (see ECharts radar).

Trending prediction markets

Clean cards, $100 → payout, Game-console density

Train stays static on GitHub Pages; the companion Game Next app hosts the interactive strip and merged /live (μgrad + Bloomberg). Product reference: Coinbase Predictions. Minors context: Baller League US.

Live wiring on Train

Game 6: Cleveland at Toronto

Pro basketball · single game

Live
CLE 34%
$100 → $294
TOR 66%
$100 → $152

$6,350,233 vol. (24h)

Set PUBLIC_GAME_CONSOLE_URL to link out

Texas vs Detroit

MLB · moneyline

TEX 90%
$100 → $111
DET 10%
$100 → $1,000
Set PUBLIC_GAME_CONSOLE_URL to link out

Baller League US · MD9 headline

Minor league · exhibition ladder

SPD 41%
$100 → $244
SNO 38%
$100 → $263

$842,110 vol. (24h)

Set PUBLIC_GAME_CONSOLE_URL to link out

Custom wind-style field inspired by Apache ECharts custom-wind — static ornament only.

Illustrative numbers, sector commentary, and the rest-of-month expandable sector rows are for education only — not trading advice, not live Kalshi/Coinbase data, not a quote, and not an investability signal. Contracts can expire worthless; venues have eligibility and geographic rules. Set PUBLIC_GAME_CONSOLE_URL at build time for header + outbound Game links.

Live analysis desk

Game /live merges μgrad + Bloomberg with Train context here

Use this page for live pipeline architecture and the Bloomberg relay panel below. Use the Game app for side-by-side iframes: serve sports-field-ugrad.html from uvspeed web/, set NEXT_PUBLIC_UGRAD_SPORTS_URL (or NEXT_PUBLIC_UGRAD_TOOLS_DECK_URL), run browser-page/server.py (default port 8765), set NEXT_PUBLIC_BLOOMBERG_CHAT_URL. HTTPS hosts cannot embed http://127.0.0.1; use a tunnel or same-origin proxy when needed.

Captions desk (paper + relay) Bloomberg relay on this page Live pipeline table Hop topology

What merges where

  • Train (this site) — static Erika catalog, listing profiles, hop/pipeline copy, Bloomberg relay instructions.
  • Game /live — μgrad embed + Bloomberg chat embed + champion / buy-in UI (non-executing demo).
  • Next step — feed real contract rows from data/catalog.json into the Game trending grid when you are ready to wire cross-repo data.

Game env (Next)

NEXT_PUBLIC_UGRAD_SPORTS_URL
Full URL to served μgrad sports field (preferred).
NEXT_PUBLIC_UGRAD_TOOLS_DECK_URL
Fallback tools deck HTML origin.
NEXT_PUBLIC_BLOOMBERG_CHAT_URL
e.g. http://127.0.0.1:8765
Bloomberg live relay

Live captions and Granite Tiny side chat

Companion to the qbitOS Bloomberg tool in bloomberg/live chat/browser-page/: reference stream, then caption-driven chat when the local Python relay is running.

Open relay (127.0.0.1) Open relay (localhost)

This site does not start the relay. In a terminal, run the copied command (or python3 server.py from that folder), wait for “Serving http://…”, then use the buttons. If a new tab does not appear, try Safari or Chrome; the buttons fall back to navigating this tab when pop-ups are blocked.

Reference stream —
Granite Tiny side chat

Captions and model reactions stream from the relay; this directory site stays static on GitHub Pages.

Quick setup

Drop-in relay next to this static Train shell

Change one working directory, run one command, open one port. Works with the bundled Python forwarder, the Bloomberg browser page, Granite Tiny over Ollama (or your host), and any tab that can reach 127.0.0.1:8765.

Train
Python
Granite
Browser
:8765
And more
  1. From this repo root (…/cursor/train), run python3 server.py — it forwards to ../bloomberg/live chat/browser-page (default port 8765). Or cd there and run python3 server.py directly.
  2. Use Open relay once the server prints “Serving…” — for the full dashboard (captions, metrics, chat).
  3. HTTPS production hosts cannot embed http://127.0.0.1; use localhost over HTTP during dev if you need both in one tab.
Relay link targets 127.0.0.1:8765
Top countries
Most represented jurisdictions in the dataset.
United States · 16,586 China · 5,994 Canada · 4,947 South Korea · 3,369 Japan · 3,309 Ireland · 2,578 Taiwan · 2,333 Australia · 1,743
Top exchanges
Quick pivots into the busiest listing venues.
OTC · 8,193 NASDAQ · 5,563 LSE · 3,767 TSE · 3,191 SZSE · 3,083 NYSE · 2,986 SSE · 2,789 NYSEARCA · 2,691
Top sectors
Fast routes into sector-based slices of the directory.
Industrials · 6,882 Information Technology · 4,478 Financials · 4,280 Materials · 4,117 Consumer Discretionary · 3,797 Health Care · 3,553 Consumer Staples · 1,792 Real Estate · 1,627
Priority markets

Put metals and the key market presets up front

The Erika lake is updated through 2026-04-30, but the time-series coverage is not uniform across all symbols. Metals are the strongest local histories, so they now get a dedicated top-level block instead of being buried below the listing directory.

Highest-priority market block

Major metals

These are the strongest, most complete histories visible in the local Erika lake and should sit above the generic listing browse flow.

Macro proxy set

Global benchmarks and futures

The preset exists for major indexes and futures, but the local lake is strongest right now on futures and ETF proxies rather than every benchmark index parquet.

Liquid digital majors

Major crypto

Crypto is present and current, with Bitcoin available locally and the README preset covering the broader major-crypto basket.

Names referenced in the presets

AI / LLM market names

These are the headline equity names from the ai-llm-boom preset. Coverage is uneven locally, so they are useful as a prominence layer but not all are equally complete.

Architecture control panel

Hop, relay, bottleneck, and system-view controls

Open the sections below for the Databento-style plant view: venue hops, live pipeline stages, relay junctions, bottlenecks, and the hardware stack behind them.

Hop and speed analysis

Indicative topology for major exchanges, index flows, and relay points

Static ECharts metaphors — gauge ring , geo-style graph , and a 2D take on flight paths (not WebGL 3D here).

Venue Benchmarks / flows Primary PoP Relay path Hop budget Latency budget Link speed
CME / Aurora futures

Best home for index futures, rates, and metals when futures lead price discovery.

 ES, NQ, RTY, YM, GC, SI, HG Aurora I / CH2 Aurora edge -> Chicago metro relay -> NY4/NY5 fanout 1 in-cage; 3-5 metro; 6-9 to NY metro sub-0.1 ms local; 0.3-0.9 ms Chicago metro; 8-10 ms one-way to NJ 10/25 GbE capture, 25/100 GbE relay
NASDAQ / NYSE / NYSE Arca / OPRA

Use this path for ETF NAV, cash open/close logic, and US equity index arbitration.

 QQQ, SPY, cash equities, ETFs, options fanout Carteret / Mahwah / Secaucus / NY4 Venue edge -> New Jersey relay spine -> Chicago and London distribution 1-2 local; 3-6 metro; 6-10 to Chicago sub-0.15 ms local; 0.2-1.1 ms NJ metro; 8-10 ms one-way to Chicago 10/25 GbE edge, 25/100 GbE relay
LSE / ICE Europe / Euronext

London is the clean aggregation point for European cash, futures, and energy venues.

 FTSE, ICE energy, STOXX-linked flows LD4 / Basildon / Slough London metro edge -> LD4 relay -> Frankfurt and NY distribution 1-2 local; 3-6 metro; 5-7 to Frankfurt sub-0.15 ms local; 0.3-1.2 ms London metro; 3-5 ms one-way to Frankfurt 10/25 GbE capture, 25/100 GbE backbone
Deutsche Boerse / Eurex

Keep Eurex-leading macro logic in Frankfurt if rates or DAX futures are your trigger feed.

 DAX, Euro Stoxx, Bund, Bobl, Schatz FR2 / FR5 Frankfurt edge -> Frankfurt relay -> LD4 and NY fanout 1-2 local; 3-5 metro; 5-7 to London sub-0.12 ms local; 0.2-0.8 ms metro; 3-5 ms one-way to London 10/25 GbE edge, 25/100 GbE relay
JPX / OSE

APAC strategies should stay regional; sending the first decision loop to the US is too slow.

 Nikkei 225, TOPIX, JGB-linked flows TY3 / Tokyo metro Tokyo edge -> Tokyo relay -> Singapore and Hong Kong regional fanout 1-2 local; 3-5 metro; 7-10 regional backbone sub-0.12 ms local; 0.2-0.9 ms metro; 35-70 ms to SG/HK depending path 10/25 GbE edge, 25 GbE regional relay
HKEX / SGX

Pair Hong Kong and Singapore to absorb regional outages and keep APAC fanout tight.

 Hang Seng, CNH, SGX index and derivatives flows HK1 / HK3 / SG1 Local metro edge -> Hong Kong or Singapore relay -> Tokyo and London distribution 1-2 local; 4-8 regional; 8-12 inter-region sub-0.15 ms local; 0.3-1.0 ms metro; 30-40 ms HK<->SG 10/25 GbE edge, 25/100 GbE relay uplinks
Live data architecture

Primary path, redundant path, and the stack between them
Stage Primary path Redundant path Data carried Junction Likely bottleneck Tech stack
Venue ingress Primary exchange handoff / multicast plant / direct cross-connect Secondary venue handoff or alternate A/B line card / network path Raw packets, sequence numbers, heartbeats, venue control messages Exchange demarc -> cage switch -> capture NIC Cross-connect oversubscription, bursty open/close traffic, sequence gaps at the handoff edge Cross-connects, L1/L2 switching, UDP multicast or direct TCP/ITCH-style feeds, optical diversity
Lossless capture Hot capture host pinned to the primary feed leg Parallel hot-capture host or standby NIC path with the same session view Nanosecond timestamps, packet payloads, gap alerts, drop counters NIC -> kernel bypass / capture process -> write-ahead queue RX ring overflow, interrupt storms, NUMA mismatch, disk flush latency during bursts PTP, hardware timestamping, RSS/IRQ pinning, DPDK/AF_XDP-class capture, NVMe journals
Decode and normalize Venue parser and schema-normalization pipeline Secondary parser workers fed from mirrored capture or replay queue Trades, MBO/MBP books, instrument defs, status events, normalized records Capture queue -> parser workers -> normalized event bus Single-thread parser saturation, schema edge cases, slow gap-fill replay after reconnect Feed handlers, schema codecs, DBN-style binary encoding, lock-free queues, replayable logs
Gap fill and reconciliation Real-time sequencer and gap-fill service Replay path from mirrored capture store or alternate venue session Recovery requests, retransmits, sequence windows, continuity markers Normalizer -> sequencer -> recovery service Recovery storms during venue instability and replay contention against live flow Sequence tracking, retransmit requests, bounded in-memory buffers, deterministic replay
Regional relay mesh Chicago, New Jersey, London, Frankfurt, Tokyo, Singapore active fanout Cross-region mirror relay or same-region hot standby Normalized live stream, partial depth, full depth, trades, metadata side channels Metro relay spine -> regional backbone -> local fanout node East-west congestion, route asymmetry, fanout amplification when many clients subscribe at once 25/100 GbE leaf-spine, ECMP, low-jitter metro circuits, internal pub-sub / stream fabric
Hot cache and serving plane In-memory live cache and session gateway Replica cache and second gateway pool in the same or adjacent region Latest books, subscription state, session tokens, entitlement-aware stream state Relay -> cache -> auth/session gateway -> outbound stream Cache stampede on reconnect, gateway fanout pressure, entitlement checks in the hot path In-memory caches, session services, load balancers, service discovery, TLS termination off hot path where possible
Historical persistence Append-only live archive and object-store landing zone Mirrored object storage / second-region archive Raw PCAP-class capture, normalized DBN records, catalog metadata, audit trail Capture + normalized stream -> archive writer -> object store / cold tier Write amplification, checksum/recompression overhead, slow cold-tier recalls during replay NVMe staging, object storage, immutable partitions, compression, checksum verification
Client delivery Live streaming endpoint close to the consuming strategy Secondary endpoint / region plus replay bootstrap path Subscription responses, normalized frames, heartbeat, backpressure / reconnect control Gateway -> internet/private link -> strategy node Last-mile jitter, client-side backpressure, TLS/session churn, slow consumer queues TCP/WebSocket/streaming APIs, client libraries, backpressure handling, reconnect and resume logic
Junction bottlenecks

Primary and redundant paths still fail at the joins
Critical junction

Venue demarcation

Primary → Venue A handoff -> cage switch -> primary capture NIC
Backup → Venue B handoff -> alternate ToR -> redundant capture NIC
Choke
Optics, ToR oversubscription, or microburst loss at the handoff edge
Watch
Drops, burst counters, asymmetric sequence drift
Critical junction

Capture to parser handoff

Primary → Hot capture host -> write-ahead queue -> parser shard
Backup → Mirrored journal -> standby parser worker
Choke
Queue depth and parser lag during open/close bursts
Watch
Backlog, gap-fill requests, replay catch-up time
Critical junction

Relay fanout spine

Primary → Metro relay -> regional spine -> local fanout cache
Backup → Secondary relay region -> mirrored fanout node
Choke
Cross-region congestion and too many subscribers on one spine node
Watch
Jitter, egress imbalance, reconnect fanout spikes
Critical junction

Gateway and entitlement plane

Primary → Hot cache -> session gateway -> client stream
Backup → Replica cache -> standby gateway pool
Choke
Inline auth/entitlement work and reconnect pressure
Watch
Connect latency, CPU hotspots, slow session creation
Critical junction

Archive and replay

Primary → Live journal -> archive writer -> object store
Backup → Second-region archive -> isolated replay path
Choke
Replay I/O contention against live writes
Watch
NVMe saturation, replay start delay, live-write jitter
Relay points

Feed aggregation and redistribution laid out like signal paths
Primary futures and macro relay

Aurora / Chicago relay

Aurora capture → Chicago metro core → US macro + metals fanout
Backup
NJ mirror + replay path
Choke
Open/close burst fanout and gap-fill overlap
US equities, ETFs, and options aggregation

New Jersey / NY4-NY5 relay

Mahwah / Carteret / Secaucus → NJ metro relay → Cash equity + ETF + options fanout
Backup
Chicago cache + second gateway pool
Choke
Cash open/close microbursts and reconnect storms
European core relay

London / LD4 relay

LSE / ICE / Euronext edge → LD4 metro core → EU fanout + transatlantic handoff
Backup
Frankfurt hot standby
Choke
Metro jitter before long-haul replication
Eurex and Xetra recovery / primary node

Frankfurt relay

Eurex / Xetra edge → Frankfurt relay core → Rates + DAX + EU standby fanout
Backup
London mirror relay
Choke
Recovery traffic colliding with live macro flow
APAC relay mesh

Tokyo + Singapore pair

TY3 / HK / SG venue edge → Regional APAC mesh → JPX + HKEX + SGX regional fanout
Backup
Tokyo<->Singapore mutual failover
Choke
Regional path asymmetry and outage reroute bursts
Server and system requirements

Baseline hardware for accessing the feed at speed
System profile

Capture / exchange edge

CPU
16-32 high-clock cores
Memory
64-128 GB ECC
NIC
Dual 25 GbE or single 100 GbE with hardware timestamps
Storage
2-4 TB NVMe scratch plus mirrored archival path

Raw packet capture, gap detection, feed recording, and immediate handoff to relay.

System profile

Relay / normalization node

CPU
24-48 cores
Memory
128-256 GB ECC
NIC
Dual 25/100 GbE
Storage
4-8 TB NVMe RAID1 or RAID10

Protocol normalization, gap-fill, compression, replay, and fanout to multiple consumers.

System profile

Strategy / hot-path node

CPU
8-24 high-frequency cores
Memory
64-128 GB ECC
NIC
10/25 GbE with RSS and IRQ pinning
Storage
2-4 TB NVMe local cache

Feature calc, signal generation, cache-resident books, and order-routing adjacency.

System profile

Research / replay node

CPU
32-64 cores
Memory
128-512 GB ECC
NIC
10/25 GbE
Storage
8-32 TB NVMe or U.2 array

Historical ingest, backtests, full-session replay, and derived dataset generation.

Fast-path checklist

Operating assumptions behind the latency budgets

Linux 6.x LTS with tuned IRQ affinity, CPU isolation, and unnecessary C-states disabled.
PTP or a tightly disciplined clock so exchange timestamps, relay stamps, and strategy clocks line up.
NICs with hardware timestamping, deep RX rings, and driver settings reviewed for burst handling.
NVMe-first storage for capture and replay; avoid SATA or network filesystems on the ingestion hot path.
25 GbE east-west links as the minimum serious relay target; 100 GbE where multi-venue fanout is shared.
Tech stack layer

Time and clocking

PTP, hardware NIC timestamps, disciplined grandmaster, monotonic local clocks

Required to compare venue ingress, relay transit, and client receive times without lying to yourself.

Tech stack layer

Network fabric

Diverse cross-connects, low-latency switches, ECMP, 25/100 GbE spine, metro wavelength diversity

This is where primary/redundant paths stop being theoretical and become physically separate.

Tech stack layer

Capture and feed handling

Kernel-bypass capture, pinned cores, lock-free queues, feed handlers, sequencers, gap detectors

The first place burst traffic and venue pathologies show up.

Tech stack layer

Normalization and stream fabric

Binary schema codecs, replayable logs, in-memory buses, shard-aware parser workers

Normalizing many venue dialects without creating a single chokepoint is the core systems problem.

Tech stack layer

Storage and replay

NVMe journals, immutable partitions, object storage, compression, checksum validation

Live and historical paths should share truth, not share contention.

Tech stack layer

Serving and client edge

Session gateways, load balancers, auth cache, streaming APIs, client resume/backpressure logic

Most visible outages come from connect/reconnect pressure rather than raw market data loss.