Financial reasoning data

Where our modelers come from

Pool 01 — Current MBA candidates Enrolled now

Harvard
Business School

MBA candidates

The Wharton
School

MBA candidates

Stanford Graduate
School of Business

MBA candidates

Pool 02 — Practising bankers & investors On the desk today

Investment banking

Private equity & asset management

The people who build the Street’s models now build your training data.

FinCUDA employs current students at Harvard Business School, Wharton and Stanford GSB — alongside active investment banking and private equity associates — to build institutional-grade financial models as training data for foundation labs. Every model is checked, line by line, by our own human inspectors.

Request a data sample Apply as a modeler

§I — The pools

Two pools.
One standard.

The Academy

Current MBA candidates

Modelers enrolled right now at the three schools that feed the Street. Most arrive having already spent two to four years on a live deal desk, and return to one on graduation.

  • Two to four years on a deal desk before matriculating
  • Returning to banking or investing on graduation
  • Screened on modeling, never on résumé

The Street

Practising analysts & associates

Bankers and investors building models for real transactions this quarter. They write the same conventions into our training data that they defend in a live committee.

  • Bulge bracket & elite boutique M&A
  • Megafund & middle-market private equity
  • Credit, infrastructure & real assets

§II — Admission & retention

The highest bar in the market.
The highest pay in the market.

We do not hire on résumé. Every modeler builds under observation before they build for a client, and the work is graded by someone who has shipped the same model on a live deal.

01 Sourced Referred from campus finance clubs and active deal desks
02 Modeling test Three-statement build from a raw filing, untimed
03 Timed live build LBO from a blind CIM, against the clock, screen shared
04 Reviewer calibration Candidate marks a seeded model; we check what they catch
05 Admitted Onto the bench

Retention

Which is how the bench stays full.

A high bar only works if the best people say yes to it. We pay more per model than our modelers can earn anywhere else — including the desks they came from — so the people with the most options are the ones choosing to build here, and to stay.

  • Top of market

    Above every competing rate we have found Benchmarked against the other side of the market, not against freelance data work.

  • Per model

    Paid on delivery, never by the hour Nobody earns more for taking longer, so there is no reward for padding a build.

  • Track record

    Rate bands are climbed, not granted Every modeler starts at the base band. Clean inspections accumulate across many models to move someone up — and a run of rejects moves them back down.

§III — Inspection

Every model is checked
by a person.

Not a linter, not a spot check. A second modeler — senior to the builder — walks every formula in the workbook and signs their name to it. Nothing reaches a client corpus unsigned.

  1. 01 — Formula trace

    Every cell walked to its source

    Precedents traced to a hardcode or a stated assumption. No orphan values, no pasted constants hiding inside a formula.

  2. 02 — Assumption audit

    Defensible, sourced, and labelled

    Drivers checked against the filing or comp set they claim to come from, and flagged where a modeler exercised judgement.

  3. 03 — Integrity & circularity

    Balance sheet balances

    Statements tie, cash flows reconcile, circular references resolve cleanly and iteratively rather than being switched off.

  4. 04 — Sign-off

    Attributed to a named inspector

    The reviewer is recorded against the model. Rejected work returns to the builder with annotations and is rebuilt, not patched.

§IV — Delivery

From workbook
to corpus.

A finished model ships as more than a file. We serialise the full reasoning chain — every assumption, every dependency, every intermediate step the modeler took — into records your training pipeline can consume directly.

Scroll to convert — 0% Workbook → reasoning records

§V — Coverage

The full modeling
surface area.

LBO

Leveraged buyout

Sources and uses, debt schedules, cash sweeps, returns attribution and exit waterfalls.

DCF

Discounted cash flow

Unlevered free cash flow builds, WACC derivation, terminal value under both methods.

3-STMT

Three-statement

Fully linked operating models built from filings, with working capital and circular interest.

M&A

Merger & accretion/dilution

Purchase price allocation, synergy phasing, pro forma EPS bridges and sensitivity grids.

PROJ FIN

Project finance

Construction draws, DSCR and LLCR covenant tests, cash traps and long-dated concessions.

SOTP

Sum of the parts

Segment-level valuation, holding company discounts, stub equity and cross-holdings.

Two doors

Start here.

For foundation labs

See the data before you commit

We’ll send a representative sample — a complete inspected model with its full reasoning chain serialised — so your team can evaluate fidelity against whatever you are training on today.

Request a data sample

For modelers

Build on the bench

Currently at HBS, Wharton or Stanford GSB, or on a live banking or private equity desk? The work is remote, paid per model, and graded by people who have built the same thing for real.

Apply as a modeler