astroquanta

loading run 04:12 · 0 / 48

astroquanta 04:12:00

astroquanta · run 04:12

6 / 48

A quant research agent that shows every test it ran.

For systematic funds. Give it a trading idea tonight. By morning it has backtested dozens of versions against real trading costs and shows you the few that survived, with a log of every one that failed.

Purchase a PlanRun a sweep of your own
astroquanta · run 04:1204:12:00

One hypothesis, 48 specifications, overnight.

  1. onboard vendor.datasetmapped 4,118 entities; backfilled 2004 to 2026; 2 licence flags raised.DONE
  2. rebuild point in timerestatement lag applied; index membership reconstructed.DONE
  3. sweep 4831 cleared walk forward; 14 cleared costs; 6 cleared capacity.6 KEPT
Nothing allocated.

Illustrative output. Hypothetical results, not actual trading.

  1. point in time → 17 killed
  2. walk forward → 0 killed
  3. costs → 17 killed
  4. capacity → 8 killed 6 KEPT
  5. 6 of 48 kept

How it works: four gates, and every failure logged with its reason.

The sweep does not rank ideas. It kills them, and says why.

Gate 01 · point in time

48 → 31

17 killed. Restated fundamentals leaked into the fit window.

Data is rebuilt as it was reported on each date.

Gate 02 · walk forward

31 → 31

0 killed. Thirty one in, thirty one out.

Fit on one window, test on the next. A gate that kills nothing is still logged.

Gate 03 · costs and slippage

31 → 14

17 killed. Edge under 4 bps against a 7 bp round trip.

Gate 04 · capacity

14 → 6

8 killed. Under $40 million before impact ate the edge.

6 / 48. The denominator stays on screen.

Five things a backtesting library does not hand you

itemwhat you getwithout it
Point in time, rebuilt rather than laggedEvery figure stored with the date it became public.Look ahead you find after allocation.
The trial log is the deliverableEvery specification tried, with the reason for each kill. Each Sharpe is cut for the count.A Sharpe with no denominator.
Costs charged before conclusionsSpread, commission and size-scaled impact on every fill. Every curve is net.Ideas that only work trading free.
Capacity stated with the resultThe dollar size at which impact eats a third of the edge.A $40 million strategy and a $310 million one look the same on a chart.
Vendor onboarding in daysEntity mapping, backfill and licence checks, once per dataset.Most of a researcher’s quarter per new dataset.

What the desk reads in the morning: the log and the survivor

astroquanta · trial log 48 rows04:12:07

The denominator.

skip the grid. Arrow keys move between its cells. →
  1. kept 6each with its own out of sample window; haircut Sharpe beside the raw oneKEPT
  2. killed 17, look aheadrestated fundamentals inside the fit windowKILLED
  3. killed 17, costsedge under 4 bps against a 7 bp round tripKILLED
  4. killed 8, capacityunder 40 million dollars before impactKILLED
Attached to every run.

the run

Ran overnight.

the log

Every attempt, with the reason it died.

the candidate

Costed, capped and haircut.

astroquanta · candidate id 3404:12:09

One survivor, fully costed.

  1. sharpe1.42 in sample; 0.81 walk forward; 0.63 haircut for 48 trials
  2. capacity310 million dollars before impact exceeds a third of the edge
  3. decay watchhalf life estimate 19 months; monitor rebuilt weekly.LIVE
What a portfolio manager asks for.
48specifications run
6still standing
0.63haircut Sharpe for 48 trials
$310Mcapacity before impact

All figures come from one modelled run, id 04:12.

Illustrative output. Backtested, modelled and simulated results are hypothetical. They do not represent actual trading, they carry the benefit of hindsight, and they may differ materially from any result achieved in the market. Past performance, real or simulated, does not indicate future results.

No output of the Services is a recommendation, an offer, a solicitation, or advice about any security, strategy or account. Every investment decision is made by the customer and the customer's own investment professionals.

“A researcher that does not report the number of trials N used to identify the selected backtest configuration makes it impossible to assess the risk of overfitting.”

Bailey, Borwein, Lopez de Prado and Zhu, Notices of the American Mathematical Society, 2014

Plans and prices, per seat

plan · researchMOST POPULAR

$7,500/mo

5 seats · $1,500 per seat

For a research pod of five.

Purchase Research
  1. Desk · $1,800/mo · 1 seatFor one researcher, one dataset at a time.Purchase Desk
  2. Research · $7,500/mo · 5 seatsFor a pod sharing one trial ledger.Purchase Research
  3. Fund · Quoted/yr · firm wideFor the whole firm, on your own infrastructure.See the Fund plan

Prices in USD and proposed. Monthly plans cancel before the next billing date. Compare every inclusion.

Try the sweep simulator. Choose N.

It runs N synthetic specifications through the same gates. No market data.

Run the sweep

About 6 seconds. Nothing is stored.

astroquanta · sms04:12:00

A run that finishes at 04:12 is worth a text. Account notifications only: run finished, log ready, billing and renewal, decay watch changes.

Join Our SMS List

  1. Astroquanta: run 0412 finished. 48 specifications, 6 kept. Log is in your workspace. Reply STOP to cancel.
  2. Astroquanta: your Research plan renews 18 Oct. Invoice sent to your billing email. Reply HELP for help.
  3. Astroquanta: decay watch on candidate 34 revised half life to 16 months. Reply STOP to cancel.