Getting Started · Interactive
Getting started with the API
Four steps from API key to measured alpha: discover the catalogue your key can read, diagnose the signal, pull its weekly decisions, and read the books that traded them. Everything runs against the live API from this page, and the code shown regenerates from whatever you enter — paste it straight into a notebook.
A key you create for yourself starts with /v1/health, /v1/me and /v1/me/keys only. The interactive panels below need catalogue, signals, portfolio and performance on the key — verify with /v1/me and ask an administrator if callable_endpoints is still the three meta ids. Once catalogue access is granted, the alpha allow-list defaults to every published alpha; an administrator can instead pin a key to a named list, in which case GET /v1/alphas returns only those, and every signal, portfolio and metrics endpoint applies the same restriction before it looks anything up, so a barred alpha is a 403 whether or not it exists (alpha_not_allowed on the catalogue route, run_not_allowed on the data routes). If an alpha you expect is absent from a restricted key, the remedy is a change to that list, not a different query.
Credentials
Calls are forwarded by a same-origin proxy allow-listed to the documented API host.
The key travels only as an Authorization: Bearer header to the API and is never logged. The platform keeps only a one-way hash of it, so it cannot be read back from the server — paste it in here each time, or tick "Remember on this device" to keep it in this browser's localStorage; otherwise it is held in this tab's memory and cleared on reload. A freshly minted key can call only the three meta routes until an administrator grants catalogue and data endpoints — if Verify reports three endpoints callable, the sections below will return 403 endpoint_not_allowed.
Discover your alphas
/v1/alphas. Each row is deliberately thin: the alpha_id — the address every other endpoint takes — plus name and tldr. Call /v1/alphas/{alpha} for the written description (thesis, signal, parameters, papers). Seals still being written stay on the list; their data endpoints return empty results with warnings rather than failing. When a package is ready, /v1/alphas/{alpha}/source/download returns a one-hour signed URL for the cleaned source ZIP. Run the catalogue call, then choose the working alpha the rest of this tutorial operates on.Run step 1 to load the alphas your key can read, then pick one here.
Diagnose an alpha
/metrics reports the rank IC and its information ratio, the IC-decile ladder, Newey–West significance and the length of the track record, all measured on the evaluation window only. The score panel underneath lets you rebuild the weekly IC series those summaries rest on, from this page. The glossary at the foot of the page defines each term precisely.No working alpha selected yet — choose one in step 1; the generated code below uses a placeholder until then.
2aCore statistics
The scorecard: rank IC and its information ratio, the long/short book's information ratio and excess Sharpe, net P&L, worst drawdown and turnover, plus how much track record it all rests on. Measured on the evaluation window only.
The scorecard lands here — IC, ICIR, information ratio, excess Sharpe, P&L, drawdown, turnover and tenure for the working alpha.
2bIC decile monotonicity
The weekly cross-section in ten buckets of the raw, unsmoothed score — Q1 lowest scores, Q10 highest. It carries signal: "raw", so it is not a ladder of the smoothed score the book traded. A good ranker climbs steadily from Q1 to Q10; the spread Q10 − Q1 is the tradeable edge.
Reads the ic_deciles block of the 2a payload — fetch the metrics above first.
2cWeekly IC series & rolling ICIR
/metrics reports the summary. You can rebuild the series behind it yourself: page the score panel, rank-correlate each week, then smooth with 13, 26 and 52-week windows.
Derives from the score panel of the working alpha — fetch the metrics above first.
2dNewey–West significance
Is the mean far enough from zero to believe once autocorrelation is accounted for? Three HAC t-statistics: the weekly IC of the smoothed score (tests core.ic), the weekly IC of the raw score from the published panel (does not test core.ic), and the long/short book's mean return.
Reads the nwt block of the 2a payload — fetch the metrics above first.
Rebalance data
direction × confidence), rank, target weights, tradeability and the decision layer's reasoning. The same score column appears on /scores; confidence is signal strength in the alpha's units, not a 0–1 probability. The index endpoint lists the rebalance calendar; a dated call returns one week in full — page it with limit/offset.No working alpha selected yet — choose one in step 1; the generated code below uses a placeholder until then.
3a · List the rebalance weeks. One row per weekly decision — the table shows the most recent weeks (the API indexes oldest-first, so this panel requests the trailing page). Click a row to load that week below; the newest implemented week on the page (traded or pending) loads automatically.
Pick a working alpha in step 1 to run this call.
The rebalance calendar appears here once you run the call.
3b · Read one week in full. The table keeps every name carrying a target weight — the long tail on top, the short tail below, each sorted by score; the histogram shows the cohort behind them, captioned with how much of it the two pages reached.
Pick a week above (or run 3a) to load its cross-section.
Portfolio & performance
No working alpha selected yet — choose one in step 1; the generated code below uses a placeholder until then.
4a · List the books. One row per (book, kind) pair; the book name is the book= parameter on every portfolio endpoint.
Pick a working alpha in step 1 to run this call.
The books table appears here once you run the call.
4b · Chart the book's equity curve. One observation per trading session, paged 500 rows per call until the window closes.
Run 4a (or pick a book) to chart the equity curve here.
Glossary
The terms step 2's scorecard reports, defined precisely.
- IC (rank information coefficient)
- The Spearman correlation, computed each rebalance week, between the cross-sectional score ordering and the following week's return ordering; the reported value is the mean across weeks. Positive values indicate a predictive ordering — roughly 0.03–0.05 weekly denotes a strong cross-sectional signal.
- ICIR (IC information ratio)
- The mean of the weekly IC series divided by its standard deviation — the signal-level analogue of a Sharpe ratio, rewarding a consistent IC over a high but erratic one.
- Information ratio
- Risk-adjusted return of the long/short book, gross of the risk-free rate — mean net return over its volatility, annualised. Net denotes after modelled transaction costs, borrow and carry; nothing is deducted for the risk-free hurdle. Served as core.information_ratio (core.sharpe is a deprecated alias of the same figure). Per-year values are stored as long_short_information_ratio and served under the unprefixed public field.
- Excess Sharpe
- The only true Sharpe the API publishes: the same long/short book net of the recorded risk-free rate. Served as core.excess_sharpe, with the rate deducted stated beside it. It is lower than the information ratio by exactly that hurdle.
- Turnover
- The fraction of gross notional traded, as the engine recorded it — annualized_turnover where the run has one, otherwise its plain, mean or (per period) total_turnover. Read annualisation and unit on the value: overall is usually per year; yearly and monthly periods serve the stored long_short_total_turnover sum and are not annualisable. It is the principal driver of transaction costs and the first constraint on capacity.
- Tenure
- The length of the evaluation track record — the count of rebalance weeks from first to last, with the calendar span alongside. This is the sample every other statistic on the alpha is measured over.
- Decile monotonicity
- The degree to which mean forward returns rise across decile portfolios sorted on the raw, unsmoothed score, Q1 to Q10 — the ic_deciles block, which always carries signal: "raw". A ladder that rises steadily means the edge runs through the whole cross-section rather than sitting in one tail; the Q10 − Q1 spread is the tradeable edge.
- NWT (Newey–West t-statistic)
- Significance of a mean, with heteroskedasticity- and autocorrelation-consistent standard errors. nwt.ic tests the mean weekly IC of the smoothed score the books trade — the same figure as core.ic. nwt.ic_score_panel tests the mean weekly IC of the unsmoothed raw_score column in the published score panel; it describes that panel and does not test core.ic. nwt.ir tests the mean long/short book return. Overlapping horizons autocorrelate the weekly series, so a plain t-statistic would overstate confidence.