Skip to content

Lab

Every desk returns to the ORIGIN.

Data, research, validation, portfolio, execution and risk — six desks of one quant lab, each measured from the same point. Pull any of them away; it comes back to where it started.

The desks of the lab

Drag any desk — it returns to the origin.Illustrative activity
  • Data: Ingesting tick feeds, Aligning as-of timestamps, Auditing survivorship.
  • Research: Writing a hypothesis, Naming its conditions, Setting the kill criterion.
  • Validation: Walk-forward window 6 / 8, Counting every variant, Charging real costs.
  • Portfolio: Sizing below f*, Balancing exposures, Holding the risk budget.
  • Execution: Slicing a parent order, Measuring fills vs. arrival, Routing to venues.
  • Risk: Watching drawdown limits, Comparing live vs. backtest, Stress-testing correlation.

IRules of the lab

  1. 01Every run is logged — the rejected ones first.
  2. 02Hurdles and kill criteria are written before the test, not after.
  3. 03Results are reported with the number of trials it took to find them.
  4. 04Survival means re-testing, not deployment.

IIRun log

Updated

A ledger of what was registered, how it was tested, and why it failed.

Example entries — these show the format of the ledger. The live log replaces them when the first runs are published.
Registered
7
Rejected
5
Survived to re-test
1
Running
1
Variants counted
140
Pre-registered
100%
Why runs were rejected83% of concluded runs rejected
  • Look-ahead1
  • Costs1
  • Spurious1
  • Regime1
  • Below hurdle1

Showing 7 of 7 runs

  1. R-0012Overnight gaps in index futures partially reverse during the first trading hour.Index futures · Intraday · exampleVariants 6Running

    Pre-registration

    Mechanism
    Liquidity providers widen quotes at the open; inventory is unwound once depth returns.
    Hurdle
    Deflated t ≥ 3.0 in at least 6 of 8 walk-forward windows
    Kill criterion
    Effect absent after costs in 3 consecutive windows

    Protocol

    Data
    E-mini index futures, 1-minute bars, point-in-time
    Sample
    2016–2025
    Validation
    Walk-forward, 8 windows, 2-day embargo
    Costs
    Half-spread + impact model + exchange fees

    Outcome

    Running

    Windows 1–5 complete. Decision at window 8.

    Registered
    Concluded
    In progress
    • Pre-registered
    • Point-in-time data
    • Costs charged
    • Variants counted
    Link to R-0012
  2. R-0011Funding-rate extremes on perpetual swaps precede short-horizon mean reversion.Crypto perpetuals · Hours to days · exampleVariants 18Survived · re-test

    Pre-registration

    Mechanism
    Crowded leveraged positioning is forced to unwind when funding becomes extreme.
    Hurdle
    Deflated t ≥ 3.0 across all 10 windows, net of costs
    Kill criterion
    Sign reverses in any 2 consecutive windows

    Protocol

    Data
    Top-10 perpetuals by open interest, hourly funding and prices, point-in-time
    Sample
    2020–2026
    Validation
    Walk-forward, 10 windows, 1-day embargo
    Costs
    Taker fees + half-spread + funding paid

    Outcome

    Survived · re-test

    Cleared the pre-registered hurdle after deflating for 18 variants. Moved to paper trading for an independent re-test — not deployed.

    Registered
    Concluded
    · 6 days
    • Pre-registered
    • Point-in-time data
    • Costs charged
    • Variants counted
    Link to R-0011
  3. R-0010Sentiment drift across consecutive earnings calls predicts the next 20 days of relative performance.US equities · 20 trading days · exampleVariants 24Rejected

    Pre-registration

    Mechanism
    Soft information in management language diffuses slowly among investors.
    Hurdle
    Deflated t ≥ 3.0, sector-neutral, net of costs
    Kill criterion
    Effect explained by post-earnings announcement drift

    Protocol

    Data
    Call transcripts aligned to publication time; survivorship-free universe
    Sample
    2012–2025
    Validation
    Walk-forward, 12 windows, 5-day embargo
    Costs
    Spread + impact model at 1% of ADV

    Outcome

    RejectedLook-ahead

    The effect disappears once transcripts are aligned to their true publication time; the raw feed stamped them at the start of the call.

    Used information that was not available at the time.

    Registered
    Concluded
    · 8 days
    • Pre-registered
    • Point-in-time data
    • Costs charged
    • Variants counted
    Link to R-0010
  4. R-0009Abnormal volume in small caps leads price by one day.US small caps · 1 day · exampleVariants 12Rejected

    Pre-registration

    Mechanism
    Informed traders accumulate before information becomes public.
    Hurdle
    Deflated t ≥ 3.0 net of costs at a tradable size
    Kill criterion
    Gross effect smaller than estimated round-trip cost

    Protocol

    Data
    Daily bars and volume, survivorship-free, point-in-time listings
    Sample
    2010–2025
    Validation
    Walk-forward, 10 windows
    Costs
    Spread + impact model at 0.5–2% of ADV

    Outcome

    RejectedCosts

    A gross effect exists, but it is smaller than realistic transaction costs at every tradable size tested.

    Did not survive realistic transaction costs.

    Registered
    Concluded
    · 9 days
    • Pre-registered
    • Point-in-time data
    • Costs charged
    • Variants counted
    Link to R-0009
  5. R-0008Two refiners' equities co-move with the crack spread closely enough to be traded as a pair.Energy equities · Weeks · exampleVariants 9Rejected

    Pre-registration

    Mechanism
    Shared exposure to refining margins.
    Hurdle
    Cointegration in levels and a stable relationship in returns, out of sample
    Kill criterion
    No stable relationship in returns

    Protocol

    Data
    Daily prices and the 3-2-1 crack spread, point-in-time
    Sample
    2011–2025
    Validation
    Rolling estimation, out-of-sample half
    Costs
    Spread + borrow on the short leg

    Outcome

    RejectedSpurious

    The relationship holds in price levels only; in returns it is unstable from window to window.

    A correlation without a stable mechanism.

    Registered
    Concluded
    · 9 days
    • Pre-registered
    • Point-in-time data
    • Costs charged
    • Variants counted
    Link to R-0008
  6. R-0007Short-term momentum in currency pairs persists through central-bank meeting weeks.G10 FX · Days · exampleVariants 30Rejected

    Pre-registration

    Mechanism
    Policy uncertainty delays price discovery around meetings.
    Hurdle
    Deflated t ≥ 3.0 in every volatility regime
    Kill criterion
    Effect confined to a single regime

    Protocol

    Data
    Spot and forwards, meeting calendars as known at the time
    Sample
    2010–2025
    Validation
    Walk-forward, split by volatility regime
    Costs
    Spread + forward points

    Outcome

    RejectedRegime

    Present in the 2016–2019 windows only and absent in every later window; it fails the impermanence re-test.

    Held in one market regime only.

    Registered
    Concluded
    · 11 days
    • Pre-registered
    • Point-in-time data
    • Costs charged
    • Variants counted
    Link to R-0007
  7. R-0006The breadth of analyst estimate revisions predicts next-month sector returns.Global sectors · 1 month · exampleVariants 41Rejected

    Pre-registration

    Mechanism
    Analysts update slowly and in herds, so revisions cluster.
    Hurdle
    Deflated t ≥ 3.0 net of costs
    Kill criterion
    Below hurdle after deflation

    Protocol

    Data
    Point-in-time consensus estimates, sector ETFs
    Sample
    2008–2025
    Validation
    Walk-forward, 14 windows
    Costs
    ETF spread + commissions

    Outcome

    RejectedBelow hurdle

    Promising before deflation; below the pre-registered hurdle once all 41 variants are counted.

    Not significant once every variant is counted.

    Registered
    Concluded
    · 10 days
    • Pre-registered
    • Point-in-time data
    • Costs charged
    • Variants counted
    Link to R-0006

Verdicts

Rejected
Failed its pre-registered hurdle or a validity check. Logged with the reason and never re-tested in disguise.
Survived · re-test
Cleared its hurdle after deflating for every variant. Moves to an independent re-test — surviving is not deployment.
Running
Registered and under test. The hurdle and kill criterion were fixed before the first window ran.

Rejection reasons

Look-ahead
Used information that was not available at the time.
Costs
Did not survive realistic transaction costs.
Spurious
A correlation without a stable mechanism.
Regime
Held in one market regime only.
Below hurdle
Not significant once every variant is counted.

“Deflated t” is the test statistic after adjusting for the number of variants tried. Hurdles are fixed before the first test. The ledger records process, not investment results.

Follow the lab

Start at the origin. Prove it with data.

Leave your email to hear when new runs are published — and when early access to the lab opens. No spam, no predictions, no promises of returns.

We only use your email to contact you about early access. No spam, unsubscribe anytime.

The lab publishes research process, not investment results. Nothing here is a recommendation or an offer.