One reconciled book, and the tools to run it.

Everything on this page is written and tested. Where a capability has not yet carried live client volume, the heading says so. Where something is not built at all, it is listed at the bottom rather than written in the future tense up here.

Each line names the file that proves it. This page describes how the platform works, not how it has performed.

The book is assembled from the brokers rather than typed in. The adapters and the reconciliation are written and tested; no desk has connected a live account yet.

• Holdings sync built for eleven Indian and US brokers, plus contract-note and statement import.Broker connections are read-only. Positions come in; no order ever goes out.proof: lib/brokers/registry.ts
• Duplicate positions are prevented by the database, not by a nightly clean-up job — the holdings key is unique per connection, ticker and exchange. Reconciling is reachable on demand from the transactions view, not only when the overnight sync runs.proof: lib/brokers/syncConnection.ts
• A client roster with one screen per client: holdings, exposure, and what changed since the last look.proof: app/(hub)/hub/clients

Once the book is trustworthy, the work is deciding what to change and knowing what it cost. These run against a loaded book; none has been exercised on live client volume.

• Rebalancing against a target model, with the drift and the trades it implies shown before anything is acted on.proof: app/(hub)/hub/rebalance
• Each client's book measured against the mandate they signed — asset-class bands per risk profile, with any trade sized to the nearest band edge rather than the midpoint.The bands are written policy held as a table, not optimizer output, so a client and a regulator can read the same ranges the software applies. A book inside its band shows no action.proof: lib/allocation/bands.ts
• Performance attribution and benchmark comparison per client and across the whole book.proof: app/(hub)/hub/benchmark
• Cost drag: what brokerage, taxes and slippage actually removed before anything reached the client.proof: lib/queries/costDrag.ts
• Rules-based smart-beta sleeves — a universe sorted on measured characteristics like value, quality, momentum and low volatility, then weighted with concentration limits so a tilt does not quietly become a sector bet. The result saves as a model basket, which the rebalancing and mandate checks above already read.Each factor carries its own stated limitation onto the screen. The value sleeve here means cheap on earnings, not cheap on assets, because the balance-sheet history a textbook definition needs is an empty table — the sleeve says so rather than substituting a weaker proxy silently. Only the price-based sleeves can be tested against history; the fundamental ones have no history to test against without looking ahead.proof: lib/factors/definitions.ts
• Client-ready reports and exports.proof: app/(hub)/hub/reports

Fee terms are contractual and client-by-client. Getting them wrong is a billing dispute, not a bug report — so the rules are enforced structurally rather than left to whoever runs the billing.

• Management and performance fees with a high-water mark and an optional hurdle, computed from an append-only invoice ledger.The high-water mark is derived from invoice history, never stored as a column anyone can edit — a client who falls and recovers owes nothing on the recovery. The engine has not yet billed a live client.proof: lib/fees/feeEngine.ts
• Billing and an audit-ready compliance log alongside it.proof: app/(hub)/hub/compliance

The discipline that separates a measured track record from a modelled one. This part is running in production daily.

• Every signal is logged when it fires and settled later against what price actually did — hit, missed, or expired. It is never fitted to the outcome afterwards.proof: app/(hub)/hub/signals
• Capital preservation is measured as maximum adverse excursion: how far a position went against you before it worked, recomputed from raw price history rather than asserted.The figures themselves sit behind sign-in — this page describes the method, not the results.proof: scripts/mae-ledger.ts
• Every signal is measured to where it actually left the market — stop, target or held to the horizon — and the result is expressed in units of the risk taken rather than as a bare count of targets reached.The screen leads with where the headline and the book disagree: signals the source ledger counted as reaching target, which breached their stop first and would have been a loss to anyone holding them. Figures sit behind sign-in.proof: app/(hub)/hub/track-record
• A monitoring fleet runs the data pipeline on an external schedule and quarantines bad batches, so a poisoned ingest stops at the gate instead of entering the record.proof: app/(hub)/hub/analytics

A desk can put its own name on the platform without a separate deployment. Branding is resolved per hostname at the root layout, so it applies across every screen rather than being painted on the login page.

• Own subdomain or custom domain, with the desk's logo, favicon, colour and display name, an optional “powered by” toggle, and its own disclaimer text in place of ours.SEBI registration type and number are carried per tenant, so the right registration appears on the right screens.proof: lib/tenant/resolveTenant.ts
• An embeddable widget locked to the tenant's own domains by a shared secret.proof: app/api/widget/data/route.ts
• Tiered seat and client limits per tenant, billed separately from both the client fee ledger and retail subscriptions — three ledgers that never mix.No revenue share has settled to a partner desk yet.proof: app/(hub)/hub/billing

Talk to us about where your desk fits →