Graywater

Data

Inputs worth trusting.

Prices are only one part of the record. Timing, limits, costs and missing outcomes often decide whether an edge exists.

What we work with

01

Prices & market state

Timestamped quotes, odds, spreads, limits, liquidity and fee schedules.

Check
Stale quotes, missing venues, timing drift and changes in market rules.
Derive
Executable prices and comparable opportunity windows.
02

Decisions & outcomes

Authorised hand histories, game events, decision logs and settlement records.

Check
Survivorship bias, duplicated events, disputed outcomes and incomplete sessions.
Derive
Decision quality, calibration and outcome distributions.
03

Digital-asset records

Public ledger events, authorised wallet records, exchange fills and funding rates.

Check
Finality, decimals, network costs, asset mapping and reconciliation.
Derive
Cash-flow models, execution costs and exposure.
04

Operational data

Approved API exports, inventory, eligibility rules, incident logs and system telemetry.

Check
Schema drift, permissions, gaps and inconsistent identifiers.
Derive
Monitoring thresholds, failure states and auditable pipelines.
05

Documents & text

Licensed research, contractual terms, public filings and client-authorised documents.

Check
Provenance, effective dates, conflicts, citations and extraction accuracy.
Derive
Structured evidence with a route back to the source.

Keep the evidence chain.

An output should be traceable to the input, the transformation and the version of the model that produced it.

  1. Scope the minimum

    Define the decision and necessary fields before accepting a full dataset. Remove direct identifiers where they add no analytical value.

  2. Inspect before modelling

    Check coverage, timestamps, duplicates and missingness. Separate observations from labels that only become known later.

  3. Preserve the record

    Track source versions, exclusions, feature definitions and evaluation windows alongside the result.

  4. Agree the exit

    Set authorised access, storage location, retention and deletion terms in the engagement scope.

Inspect a sample

GW-S01 / CSV

Baseline and three probability scenarios, with the same payout, cost and turnover assumptions.

Download synthetic inputs

This file contains four scenario rows. The 1.2 million observations used in the illustrative study are a placeholder workload, not a supplied historical dataset.

Examine the method Data handling