We build custom systems for trading and research teams: start with our own market data or swap in your own feeds, run factors and backtests on your definitions, and deliver intraday monitoring with on-premise deployment together. The platform covers A-shares, Hong Kong, US equities, domestic futures, international futures and crypto — one rule set per market, never blended.
The platform already runs the full chain — data, factors, backtesting, decision — so customisation never starts from zero. Data begins on our own market feeds and can be swapped for yours; whether definitions change, how pages are laid out, how workflows run — we implement it for your scenario, or we deliver just one segment of it. Either way it lands, and someone maintains it afterwards.
Our own multi-market data works out of the box, and so does yours. Custom HTTP sources with configurable auth headers and 7 dataset types (daily bars / adjustment factors / realtime / minute / level-5 order book / full tick / fundamentals), switchable across sources, materialised to local files with adjustment and completeness validation.
77 built-in factors ready to use, plus a formula engine for your own. Every factor comes with IC, IR and stratified NAV curves — not a one-word claim of being "effective". The wider workbench also covers correlation heatmaps, out-of-sample mining and composite factor construction.
Strategies are expressed as conditions and signals; backtests are priced with realistic costs: limit-up/limit-down fills blocked, commission and stamp duty, slippage, entry and exit conventions — all tunable, with parameter optimisation, Walk-Forward and robustness assessment to fight overfitting. The robustness grade is derived from real backtest metrics, never assigned by hand.
9 rule families cover strategy signals, price, limit-order queues, sectors, unusual moves, volume surges and date reminders; triggers push to Feishu, WeCom, custom webhooks or email, with cooldowns and severity tiers.
Pages and workflows are developed and delivered by us for your scenario — layout, metric definitions, menu order and permission granularity all follow how your team actually works. You never have to touch the code yourself; accepted versions go straight into the system. Branding — logo, primary and secondary colours — can be switched to your own.
The whole system installs on the server you specify, with data and configuration never leaving your network; multiple accounts or teams sharing one deployment stay isolated; delivery includes deployment scripts and documentation, with upgrades on a release cadence. It can also be packaged as a desktop client for standalone use.
The interface is driven by trading session: overnight and picks before the open, realtime and unusual moves intraday, review after the close. Numbers right-aligned and monospaced, red for up and green for down, so you read the number before you look for where it is.
Multi-account management, limit orders that queue rather than expire at the limit, full position / trade / NAV records, and batch account creation for side-by-side comparison.
Logo, primary and secondary colours written through to the backend, so the whole interface changes — white-label delivery without touching code.
Every market has a different trading regime: price limits, board lots, settlement, trading hours, taxes and costs. We isolate data and rules per market with one profile each, rather than reskinning a single logic set — so "US stocks have no price limits" and "futures need mark-to-market and margin" never get treated as the same thing. A-shares, Hong Kong, US equities, domestic futures, international futures and crypto are all connected — the first five follow equity and futures regimes, crypto trades 7×24, each computed on its own terms.
5,500+ instruments across the full market, with the Main Board, ChiNext / STAR Market and BSE handled separately for price limits and trading rules. A-shares is our deepest market — limit-up ladders, the leaderboard, sentiment cycles and other A-share-only concepts each have their own pages.
A separate build from A-shares: trading hours, board lots, no price limits and T+0 all differ. Hong Kong-specific data lives in dedicated pages in the system, not as a handful of extra metrics.
Runs on Eastern Time: 390 minutes of continuous trading with no midday break, with daylight saving converted through the trading calendar. US-only timing signals, dividends and splits, and corporate actions each have their own pages.
Futures are not "another kind of stock": the underlying is a contract (variety plus expiry month), marked to the daily settlement price, with margin and leverage. Futures-only term structure, basis, open interest and inventory each have their own pages.
A separate regime from domestic futures, so it is a separate market: currency, contract suffix, sessions and calendar all differ.
Continuous 7×24 trading with no trading calendar, matched and settled per exchange rules — a different regime from both stocks and futures, so it is handled on its own.
A market's special data only counts if it is implemented against that market's own rules. Below are the dedicated pages for each of the six markets. The counts differ, but every page answers a question unique to that market.
Limit-up counts stacked by board level, marking first board, second board and promotion rate
Sealed volume and amount, with broken boards and re-seal warnings handled apart
Daily top-trader board and institutional seat activity
Leaderboard, call-auction baseline and an AI daily review — archivable and pushable
Cycle position judged from sentiment metrics, counting sectors held and above historical average
Add the names you want to keep watching; hits route straight to the monitoring centre
Industry and concept analysis on both axes, with member counts and a rotation matrix
Indices run on their own data pipeline and never mix into stock strategy calculations
Scale in over batches with position reminders
Financials loaded at announcement timestamp, with archivable reports
Describe an idea, generate the condition set, then replay it on local minute bars
Estimated net buying and holding distribution, with the estimation basis and known noise labelled on screen
Short-sale turnover as a share of total turnover, from HKEX per-stock official disclosure
Call level, distance to call, street percentage and premium
Call and put, in and out of the money, effective gearing and premium
Dual counter, primary counter and management fee with direct-deduction basis
Mandatory distribution, yield and scale — sensitive to interest rates
Broker and custodian seat-level holdings, the only institutional-level position view in Hong Kong
AH premium, per-lot amount and a valuation cross-section ranking
CBOE VIX / VIX9D / VVIX / SKEW, separating near-term from far-term fear
Full calendar of cash dividends and splits, TTM yield basis, pure price return against total return
The official Nasdaq directory separating ETFs from stocks precisely — the precondition for a screening universe
SEC EDGAR SIC classification with XBRL financials on a TTM basis
Price curves across expiries of the same underlying, with contango/backwardation and annualised roll yield
Basis = spot − futures, a second price for the same thing that only futures have
The only futures metric showing capital actually present; the main contract is identified by open interest
Warehouse receipts and inventory percentile — with roll-related breaks excluded so a fake "all-time low" never shows
Multiplier, margin rate, minimum tick and price limit — what one lot costs and how much it levers
Contract rankings across all sectors, priced in USD and cents
Price curves across expiries of the same underlying, contango and backwardation, and annualised roll yield
The only futures metric showing capital actually present — continuous main contract plus per-contract rankings
CFTC regulatory disclosure — the truth about commodity and rate positioning
Side-by-side detail for the same underlying priced in two venues
Per-contract coverage and data gaps, stating plainly when the source has nothing
7×24 minute bars and tick data organised on a continuous timeline with no daily reset
Exchange conventions for matching, settlement cycle and minimum tick
The same factor library and backtest engine, with rules adapted to the continuous timeline
A factor has to be writable, computable, testable, combinable and minable before it deserves a place in a strategy. Only factors that survive the whole pipeline are worth keeping.
Write factors in a DSL, with dependencies expanded recursively and syntax and field availability validated live. Once written it trial-runs over the most recent 40 sessions on the panel, showing at a glance whether the direction is unknown, high-is-bullish or low-is-bullish.
Set the window, rebalance frequency (daily / weekly / monthly), grouping and cost parameters, then get layered equity curves, IC and IR. Filter results carry a verdict column — not just an equity chart.
Every pair of factors is correlated and rendered as a heatmap. Highly correlated factors side by side look like more signal but actually express the same thing repeatedly — see the overlap first, then decide whether to combine.
Not sure what to test? Start from a preset and batch-test a selected set of factors in one click. The system can also mine candidate factors automatically, validate them out-of-sample with nested folds, and search factor combinations with beam search.
Combine several factors into a new one by weight, with the weighting scheme chosen explicitly. Once combined it is an ordinary factor — backtestable and tradable like any other.
Base, virtual, composite and custom factors are managed together, annotated with warm-up period, applicable asset scope and stability. Any factor can be turned into a strategy or sent for testing in one click.
AI in this system has an explicit boundary: it explains the quantitative results clearly and falsifies the candidates. It does not pick stocks and it does not promise returns — every number comes from local computation, and the model only organises the language.
We do not sell "let AI pick a few stocks for you" — that is not reproducible, not falsifiable, and every outcome can be explained after the fact, which means it carries no information. So in our system AI does exactly two things: explain the quantitative results clearly, and try to falsify the candidates. Selection and scoring are done by a deterministic scorer that is reproducible and auditable; AI takes the local data and looks for reasons to reject.
Reads the strategy development spec, has the model write framework-compliant strategy code, then backtests it for a baseline, diagnoses the problem and revises — a bounded "generate → backtest → diagnose → revise" loop, capped at 4 rounds by default. The discipline is hard-coded: backtest before changing anything, one primary fix per round, and every change must be justified by backtest metrics rather than imagination. The backtest is read-only, and the result never goes live on its own — a human signs off.
The scorer only compares cross-sectional strength, so it is structurally blind to absolute risk — a name can be the most resilient in the universe while sitting on its limit-down. The falsification layer closes that gap: every candidate is treated as a hypothesis awaiting test, checked objectively against local data, and anything that trips a hard veto is removed outright while softer findings stay in the pool with the risk flagged.
The assistant can call system tools directly — list factors, list strategies, query data capabilities, run a backtest — retrieving the catalogue on demand instead of stuffing the entire factor library into the context window. The backtest tool returns only summary metrics (total return, annualised, max drawdown, Sharpe, win rate, trade count), never the equity curve or per-trade log.
Single-name analysis produces reports across four dimensions — technical, market context, financials and news — all sharing one focus instruction. Financial interpretation and end-of-day market commentary are model-written too, but every number they cite comes from local computation: the model organises the language, it does not invent the figures.
The AI layer speaks any OpenAI-compatible service — point it at a cloud API, at your own inference server, or straight at a local Codex CLI. Keys live in the system key store, never in source code, and never leave your network.
The desktop handles research; mobile does three things: watch, receive alerts, and make the call on a single name. Strategy building and factor formulas need a large screen, so we do not force them onto a small one. Mobile can be part of a custom delivery, or built on its own first.
The whole system installs on your own server — market data, factors and strategies never leave your network.
When multiple clients or teams share one deployment, data directories and configuration stay separate and mutually invisible.
Estimates, incomplete data and untradeable signals are clearly labelled in the interface. We never fake precision.
Ships with our own multi-market data and swaps to yours — custom HTTP sources and source switching, with failover if one feed drops.
Private deployment, desktop client, mobile — pick the form that fits your situation. The underlying definitions stay identical across all three.
The whole system runs on a server you specify; data and configuration never leave your network. When several clients or teams share one deployment, data directories and configuration stay separate and mutually invisible.
The most common form, chosen by most teams.
Packaged as a Windows installer — double-click and it runs: a backend service bound to localhost only, with the frontend in its own window. No server to configure, and the full feature set on your own machine.
Suits small teams, single-machine research, and anyone who does not want to run infrastructure.
Mobile is built around your situation and can be part of a full delivery or come first on its own. It is not a shrunken desktop build — watching quotes, receiving alerts and making the call are what genuinely belong on a phone.
Currently in development — please treat the actual shipped feature set as the source of truth.
This system is delivered as custom work: we review your requirements and data conditions first, then agree scope, timeline and price. Below are the three common shapes — they can be combined.
Price and timeline depend on scope, data licensing and whether we need to integrate with your existing systems — so we do not publish a single price. We start with a requirement review, then put a written proposal and scope list in front of you before discussing price and schedule.
We look at your data conditions, current process and the problems that must be solved.
A written scope list, milestones and quote; key screens delivered first as clickable prototypes.
Connect a data source — ours or yours — and get adjustment, backfill and completeness validation working.
We walk the acceptance checklist item by item and keep fixing what is blocked. Nothing gets waved through.
Deployment docs, operating manual and upgrade path, plus support for as long as we agree.
Not investment advice.SuperAlpha is a research and backtesting tool. All scores, recommendations and ranges are model computations based on historical data and are not promises about the future.
Data sources and definitions.Market, fundamental and flow data comes from our own data services, licensed providers or public sources; our own market data is delivered with the system and is not resold or redistributed separately. Estimated metrics (such as net southbound buying) are labelled "estimated" in the interface, and missing data is shown as empty rather than zero.
A backtest is not live trading.Backtest results are simulated using a cost model and the trading rules in force at the time. They do not account for liquidity shocks or failed fills in extreme markets, so they diverge from live results.
Custom delivery and ownership.Code ownership, data ownership, redistribution scope and operational responsibility for each project are written down in the contract; deployment is private by default and data directories are never shared between clients. System pages and business workflows are developed and delivered by us, and extensions follow the same requirement-review process.
Read the full disclaimerLianxin Digital builds tools for quantitative research. The "chain" in our name is the data chain — we link quotes, factors, backtests and decisions into one traceable thread; the "xin" means accumulation, because research has no shortcuts.
We do one thing: turn the research workflow of trading and research teams into a system that runs and stays maintainable — data ingestion, factors and backtesting, intraday monitoring, on-premise deployment. We make no promises about returns; we lay out the definition behind every step and leave the judgement to you.
Leave a way to reach you and what you are after. We start with a requirement review and give you a written proposal and scope list before discussing price and schedule.