How SheetRank Grading Works: Cell-Level Verification Explained

Most financial modeling practice tools check whether your final number matches an answer key. SheetRank checks something harder: whether you built the right model. That distinction — value matching vs. formula verification against hidden inputs — is why live graded practice catches mistakes that static case studies miss. This guide explains exactly how SheetRank's grading engine works, what it checks, and how to use the feedback to improve faster.

The problem with traditional answer keys

A static case study has one set of inputs and one correct output. Once the answer is published, practice degrades into memorization:

• Hardcode the exit equity value instead of building the formula
• Copy a template that produces the right number without understanding the logic
• Reverse-engineer the output from a leaked answer key

In a real interview, the interviewer changes an assumption mid-conversation and asks you to recalculate. A model that only produces the right number for one specific input set falls apart immediately. SheetRank's grading is designed to test the same skill interviews test: can you build logic that works under any reasonable input?

Three layers of grading

Layer 1: Formula integrity

Every target cell is checked for a live formula — not a hardcoded number:

Pass:  =B05*(1-B06)*B07          (formula referencing input cells)
Fail:  4250000                     (typed number with no formula)

If a cell matches the expected value but contains a hardcoded number instead of a formula, it fails. This closes the most common shortcut: typing the answer instead of building the relationship.

Why this matters: in an interview, when the partner says "what if vacancy is 8% instead of 5%?" — you need a formula that recalculates. A hardcoded cell can't do that.

Layer 2: Hidden scenario inputs

Each deal includes hidden scenarios — input variations you haven't seen before submission. The grading engine runs your formulas against these hidden inputs and checks whether your outputs are still correct.

Visible inputs:  200 units, 5% vacancy, 5.0% cap rate
Hidden scenario: 185 units, 8% vacancy, 5.5% cap rate

If your model is built with correct logic, it produces the right answer under both visible and hidden inputs. If you hardcoded around the visible scenario, hidden inputs expose the gap.

This is the same principle audit firms use: perturb an input and verify the output moves correctly. It tests whether you understand the math, not whether you memorized a specific answer.

Layer 3: Tolerance bands

Real modeling has legitimate small differences in convention:

• Rounding at different stages
• Beginning vs. average balance interest
• Tax floored at zero vs. negative tax allowed
• Mid-year vs. end-of-year discounting

SheetRank grades with tolerance bands, not exact match:

Dollar amounts: ±$1 tolerance
Percentages/rates: ±0.01% tolerance
Ratios/multiples: ±0.01x tolerance

This checks the relationship between inputs and outputs without punishing reasonable convention differences. A model that's conceptually correct but rounds differently than the answer key still passes.

Partial credit by target

Grading isn't all-or-nothing. Each deal has multiple targets — individual cells or sections that are checked independently:

Deal: Project Sunbelt (Multifamily Pro Forma)
Targets:
  ✓ GPR                    $2,400,000
  ✓ EGI                    $2,160,000
  ✗ NOI                    $1,350,000 (expected $1,296,000)
  ✓ Going-In Cap Rate      6.48%
  ✓ DSCR                   1.32x
  ✗ Exit Value             $18,500,000 (expected $17,200,000)
  ✓ Levered IRR            18.2%

You see exactly which sections passed and which failed — with the expected value for every miss. This tells you where to focus: not "the model is wrong" but "NOI is wrong because you missed property tax reassessment."

What the grading report shows

After submission, you receive:

1. Overall score — percentage of targets passed
2. Cell-level errors — every missed target with your value vs. expected value
3. Formula failures — cells that matched the value but lacked a formula
4. Section grouping — errors organized by model section (revenue, expenses, debt, returns)

Use the report diagnostically:

• One section wrong, rest correct: conceptual gap in that section (read the relevant resource guide, rebuild)
• Everything slightly off: likely a convention difference (check rounding, timing)
• Formula failures on correct values: you hardcoded — rebuild with formulas
• Everything wrong: structural issue (wrong build order, missed a section entirely)

How grading differs by model type

| Model Type | Key Graded Sections | Common Errors Caught | |-----------|----------------------|---------------------| | Multifamily pro forma | GPR, EGI, NOI, DSCR, exit value, IRR | Annualization, tax reassessment, CFBD vs. NOI | | LBO | Sources & uses, debt schedule, MoIC, IRR | Equity as input not plug, missing cash sweep, circularity | | 3-statement | IS/BS/CF linkages, balance check | Retained earnings, cash flow reconciliation | | JV waterfall | Tier 1–4 distributions, GP/LP split | Catch-up formula, pref compounding | | DCF | UFCF, WACC, terminal value, implied price | Terminal value method, WACC weights, net debt bridge | | M&A | Pro forma EPS, accretion/dilution | PPA goodwill, interest on new debt, share count |

How to use grading feedback effectively

The rebuild loop

Don't just fix the wrong cell — rebuild the section:

1. Submit model → receive grading report
2. Read every error, categorize by type (conceptual vs. careless)
3. Read the relevant resource guide for conceptual gaps
4. Rebuild the failed section from scratch (don't patch individual cells)
5. Resubmit → compare score improvement
6. Repeat until 90%+ accuracy

Patching individual cells teaches you to match numbers. Rebuilding sections teaches you the logic.

Speed + accuracy progression

Track both metrics over repetitions on the same deal type:

| Attempt | Score | Time | Notes | |---------|-------|------|-------| | 1 | 62% | 95 min | Missed tax reassessment, wrong DSCR numerator | | 2 | 78% | 72 min | Fixed taxes, still wrong exit cap convention | | 3 | 91% | 58 min | All sections pass, 2 formula failures | | 4 | 97% | 48 min | Clean submission |

The goal isn't just 90%+ — it's 90%+ within the time limit you'd have in a real test.

Why this matters for interview prep

SheetRank's grading tests the same skills interviewers test:

| Interview Test | SheetRank Equivalent | |---------------|---------------------| | "Build from a blank grid" | Deal flow starts from empty workbook | | "Walk me through your assumptions" | Formula integrity proves you built logic, not hardcoded | | "What if I change this input?" | Hidden scenarios test formula flexibility | | "Your DSCR is wrong — what's in the numerator?" | Cell-level error on DSCR target with expected value | | "Present your investment thesis" | Grading report shows which sections you understand vs. don't |

Candidates who practice on SheetRank and consistently score 90%+ on their track's core deals are demonstrating the same proficiency that modeling tests and superday cases require — with objective, repeatable measurement instead of self-assessment against a static answer key.

Getting started

1. Pick your track's recommended first deal (Project Sunbelt for REPE, Project Apollo for PE, Project Catalyst for IB)
2. Build from the blank grid in Beginner mode
3. Submit for grading
4. Read every error in the report
5. Rebuild failed sections
6. Resubmit until 90%+
7. Move to speed sessions for timed practice
8. Progress to advanced deals and new archetypes

The grading engine is the feedback loop that turns practice into proficiency. Use it on every session — not just the first attempt.