Anyone can read a derivation.
This proves you can build it.
Eight desk-level quant projects, each with a hidden test suite. You build the model, submit your numbers, and get scored against a reference implementation — the same way a desk would check your work.
Free warm-up available now · Project 01 live · Projects 02–08 being authored.
The gap this fills
Reading a PDF
- You follow the derivation and it makes sense
- You copy the code and it runs
- Nothing tells you your model is wrong
- In the interview, you describe — not demonstrate
Passing the Gauntlet
- You build it from a spec, like real desk work
- Hidden tests catch the errors you cannot see
- A wrong day count fails immediately, with a hint
- You walk in with numbers you can defend
How grading works
Read the spec
Conventions are fixed and stated exactly — day count, compounding, pillars. No ambiguity to argue about later.
Build it
Python, your own way. Public sample tests let you check yourself locally before submitting.
Submit your numbers
A small JSON object with your outputs. Graded against a reference implementation at a calibrated tolerance.
Get a scorecard
Pass or fail per test, a score, and a hint on what broke — never the answer.
Why the tests are hidden
If failures revealed the expected value, the grader becomes an oracle you could brute-force without solving anything. You get the failing test and a hint about the likely cause — enough to debug, not enough to cheat.
Try the loop before you pay
Project 00 is a free warm-up: price a 5-year bond and compute its duration. An hour of work at most. You run the whole mechanism end to end — read the spec, implement, self-check, submit, read a real scorecard — so you can judge the grading before spending anything.
- 1 Read the spec — conventions fixed, nothing to argue about
- 2 Fill in three functions in
starter.py - 3 Run
sample_tests.py— 9 public checks - 4 Email
submission.json, get a scorecard
The projects
One free warm-up, then eight paid projects. Deliberately fewer and harder — each paid one is one to three days of genuine work, not an exercise.
- Free
Warm-up: bond price and duration
An hour of work, graded the same way as everything else. Exists so you can test the loop end to end before paying.
- Available now
OIS / SOFR discount curve bootstrap
Build the curve everything else prices off. Graded on discount factors, zero rates, a forward, and curve monotonicity.
- Authoring
Vol surface arbitrage screen
Detect calendar and butterfly violations. Graded against seeded arbitrage cases.
- Authoring
XVA on a netting set with collateral
CVA and DVA under CSA thresholds. Graded on exposure profile quantiles.
- Authoring
Market-making PnL simulator
Inventory risk and adverse selection. Graded on the moments of the terminal PnL distribution.
- Authoring
Bermudan swaption via Longstaff-Schwartz
Graded on price within tolerance of a reference lattice, plus the early-exercise boundary.
- Authoring
Model validation challenge
You are handed a model with a deliberately seeded bug. Find it, fix it, and match the corrected outputs.
- Authoring
Cross-currency basis hedge optimiser
Graded on hedge ratios under a basis shock.
- Authoring
Intraday liquidity and slippage model
Graded on execution cost estimates against a reference fill simulation.
See a real project spec
Project 01 is finished and its full candidate spec is public, so you can judge the standard before paying anything.
Project 01 — Bootstrap an OIS / SOFR discount curve
Fixed conventions (ACT/365F, continuous compounding, annual pillars), five par OIS quotes in, and a JSON object out. The spec includes the self-check that must reprice every input swap to zero, plus the three ways candidates actually fail it.
Read the full specTolerance is measured, not guessed
Project 01, scored against its own hidden tests:
| Submission | Score | Result |
|---|---|---|
| Correct implementation | 100 | Pass |
| Wrong day count (360 vs 365) | 10 | Fail |
| Discrete instead of continuous compounding | 85 | Fail |
Loose enough that any sane floating-point path passes. Tight enough that the two mistakes candidates actually make are caught immediately.
Questions
How is this different from the ₹799 project pack?
Completely different projects, and a different mechanism. The 45-project pack is reference material you read — derivations, Python and C++ implementations, interview questions. The Gauntlet grades work you produce. Owning one does not overlap the other, and the projects here appear in no other Desk2Quant product.
What language can I use?
Python for the first release. You submit numerical outputs rather than code, so anything that produces the right numbers is fair game — numpy, pure standard library, or your own solver.
What if I think the grader is wrong?
Every expected value is generated from a reference implementation that reprices each input instrument to machine precision, and the tolerances are calibrated against deliberately wrong solutions. If you still believe a test is unfair, email the scorecard and it gets reviewed by a human.
What does it cost, and what if I cannot pass it?
₹999 for Project 01 (list ₹1499), with GAUNTLET20 for 20% off.
There is no time limit and no attempt expiry — if you get stuck, the hints and
the public self-checks are there to get you unstuck, and you can email for help.
How many attempts do I get?
Multiple, with a per-day cap. Debugging against a hidden suite is part of the exercise; unlimited rapid-fire guessing is not.
Is it live yet?
Project 01 is available now at ₹999 and its grading engine is built and unit-tested. The remaining seven are being authored. Submissions are currently graded by email within one working day; instant automated grading is in development and every buyer gets it free when it ships.
Start with Project 01
₹1499 ₹999
The OIS/SOFR curve bootstrap: full spec, runnable starter, eight public
self-checks, and a graded scorecard against six hidden tests.
Use GAUNTLET20 for 20% off.
Submissions are graded by email within one working day. Instant automated grading is in development and every buyer gets it free when it ships.