The setup term
A module often loses its first month to environment setup: library versions, data access, notebook conventions that vary by student. A hosted workspace with a bundled sample project removes that month.
For faculty and students
Spend the term on the research question, not the environment.
For faculty: a module built around one workspace, so grading means reading an evidence pack rather than untangling a dozen notebooks. For students: a project with a chronological split, a purge gap and a method library that cites its sources, from week one.
01A term with Quant Compose
A twelve-week module maps onto the workspace's own eight-stage research journey. Nothing here is a simplified teaching version — it is the same pipeline a researcher runs, with the same refusals.
Week 1
Week 1 — a sample project on synthetic data
Every account can open the bundled reference-data sample project: deterministic, seeded, no network call, and never a basis for a claim about a real security. Students see the full pipeline populated before they touch their own data.
Projects The projects workspace: each project is an isolated environment with its own universe, data, models and results. Week 2
Weeks 2–4 — explore, prepare, engineer features
Students apply a universe, run the fixed cleaning pipeline (duplicates, missing values, calendar alignment, outlier treatment), then build features from the price-and-momentum, statistical and technical-indicator families. The chronological split and purge gap are on by default.
Feature Engineering The feature registry: 57 features grouped by family, 13 selected, each named with its definition. Week 3
Weeks 5–8 — models, then walk-forward validation
Train a classifier or regressor, then run walk-forward validation with a purge gap floored at the label horizon. The fold table reports mean, standard deviation, min and max, so a student cannot quote the best fold and hide the worst.
ML Models Walk-forward validation: five folds with a five-row embargo, mean ROC AUC 0.5546, spread 0.0472, worst fold 0.5034, and the interpretation. Week 4
Weeks 9–10 — portfolio and stress
Optimise a small universe under one of seven construction methods, then replay a real historical episode or a hypothetical shock against it. Infeasible frontier points are recorded with the reason, not dropped.
Scenario & Stress Testing Base versus stress: return, volatility, VaR, CVaR and drawdown before and after the shock. Week 5
Weeks 11–12 — the evidence pack is the submission
Export the fixed PDF pack. Page two carries fold count, mean, standard deviation and the worst fold for out-of-sample validation — a marker reads the retention number before the headline Sharpe.
Evidence pack (PDF) Page 2: model, out-of-sample validation with fold count, mean, standard deviation and worst fold, and the strategy backtest against buy-and-hold.
02Three problems it removes
Named specifically, not as a category.
Unreproducible submissions
A notebook handed in at the end of term rarely re-runs on the marker's machine. An evidence pack is exported from a run record with a content hash, versions and seed already attached.
Leakage caught after grading
A shuffled split or a scaler fit across train and test inflates a metric quietly, and a marker without a leakage checklist can miss it. The pipeline defaults to chronological split, embargo and a purge gap, and states plainly when a student has switched one off.
03Rigour and reproducibility
Method that holds up when someone checks.
What makes a submission defensible is not a grade of confidence — it is a fixed set of defaults a student cannot quietly bypass, and a method library that says where each method comes from.
- Leakage-safe by default
- Chronological split, scaler fit on train only, no shuffling and a purged embargo are all on unless a student explicitly turns them off — and turning one off generates a plain-English warning in the run record.
- 72 of 119 methods checked against a reference
- The method library states validated against a textbook, paper, closed-form identity or independent library, and states plainly which of the 119 are not yet checked.
- A deterministic sample project
- The bundled reference-data provider is seeded and reproducible, so every student's week-one run produces the same numbers, and disagreements are about method, not data drift.
- Deflated Sharpe, not raw Sharpe
- Comparison tables sort by Deflated Sharpe. A student who tried twenty configurations and kept the best cannot present its raw Sharpe as the finding.
04Objections, taken seriously
The questions this audience actually asks.
Is this rigorous enough for academic work?
Purged, embargoed, walk-forward validation is the default rather than an option a student has to know to enable. 72 of the 119 catalogued methods are checked against a published textbook, paper, closed-form identity or independent library, and the method library states which references were used.
Why not just Python and a notebook?
A notebook is still where exploration happens well. What the workspace adds is a run record a marker can trust: chronological split enforced, a content hash on the data, and a worst-fold figure that a notebook does not print unless a student remembers to.
Can students use real data?
Yes — CSV upload, Yahoo Finance and the Ken French factor library need no paid credential. The bundled synthetic sample project exists so week one does not depend on any of them being available.
Is there a cost for students?
Free during the private beta. The planned Researcher plan is designed to remain free, though plans and pricing are not yet billed live.
05In beta, for you
What a department gets access to during the beta, and what is still being built.
- Full access to the research pipeline, the 119-method library and the fixed evidence pack, at no cost
- The bundled synthetic sample project for a dependency-free first week
- Cohort and department-level access, provisioning and seats are design intent, handled individually rather than self-serve today
- Structured feedback from early academic users feeds directly into what gets built next
Other audiences, and further reading
Related capabilities
Bring a module or a project to Quant Compose.
Departments get in touch directly; individual students and faculty can apply for beta access today.