Financial analysis

Growth formulas to make the model use (and show)
 
YoY growth     = this period ÷ same period last year − 1
CAGR           = (ending value ÷ starting value)^(1 ÷ years) − 1
Run-rate       = recent period × periods per year  (not for seasonal)
Organic growth = growth excluding acquisitions and FX effects

DCF formulas to make the model use (and show)
 
FCFF  = EBIT × (1 − tax rate) + D&A − capex − change in NWC
WACC  = E/(D+E) × cost of equity + D/(D+E) × cost of debt × (1 − t)
PV    = FCF in year t ÷ (1 + WACC)^t
TV    = FCF in year n+1 ÷ (WACC − g)          (Gordon growth)
   or = metric in year n × exit multiple      (exit multiple)
EV    = sum of PV of FCF + PV of TV
Equity value = EV − debt + cash (− preferred, − minority interest)
Value per share = equity value ÷ diluted shares

Variance = actual − budget
  favorable: revenue above budget, cost below budget
  unfavorable: the reverse
 
Price/volume split for revenue (sums to the total variance):
  price variance  = (actual price − budget price) × actual volume
  volume variance = (actual volume − budget volume) × budget price
 
Flex budget = budget cost per unit × actual volume (for variable
costs); compare actual with the flex budget to isolate efficiency.

<materials>
[attach the filings, e.g. the last two 10-Ks and the latest 10-Q as
PDF or HTML, or the annual reports of a non-US company]
</materials>
 
<task>
I want an analysis of this company from these filings. Don't analyze
anything yet, and don't state any figure about the company. Interview
me first: at most 10 questions, grouped under the decision, the
documents and periods, definitions and peers, valuation inputs,
tools, and the audience.
</task>
 
<output_format>
For each question: one line on why it matters and 2–3 typical
answers, e.g. decision: "researching a possible investment",
"checking a supplier's health", "a job offer at the company".
</output_format>
 
<process>
Ask all the questions in one message, then stop and wait. Where the
filings answer a question (fiscal year end, units, GAAP or IFRS), say
where, cite the page and ask me to confirm. Never invent numbers,
names, quotes or sources: write [NEEDS SOURCE].
</process>

<task>
From my answers, propose 2–4 materially different ways to analyze
this company, such as: a full three-statement analysis with an
intrinsic DCF; a reverse DCF (what growth today's price implies);
comparables first, with a DCF cross-check; a credit view (liquidity,
leverage, maturities, covenants) with no valuation.
</task>
 
<output_format>
A table: Approach | Question it answers | Data it needs beyond the
filings | Most sensitive input | Effort. Then recommend one for my
decision and say what would change your mind.
</output_format>
 
<process>
Don't compute anything yet; mark guesses [NEEDS SOURCE]. Stop after
the recommendation and wait for my choice.
</process>

<materials>
[attach the filings, and paste the statements as CSV if you have them
(e.g. from EDGAR's XBRL data) with units and period ends; add peers'
figures and a dated share price if the approach needs them]
</materials>
 
<context>
Decision:
[decision: investment research | credit | supplier check | learning].
The company reports under [standard: US GAAP | IFRS] in
[units: thousands | millions] of [currency, e.g. US$]; fiscal year
ends [date, e.g. 30 June]. Approach chosen:
[approach: full DCF | reverse DCF | comps with DCF check | credit focus].
The memo is for [reader, e.g. me, a non-analyst who will spot-check
the code].
</context>
 
<task>
Write an analysis of the company from these filings: tie-out, ratios,
valuation with sensitivity, and red flags.
</task>
 
<requirements>
- Use only <materials>. Recall no figure about this company, its
  peers or the market from memory.
- Cite item, note and page for every figure from a filing, using the
  GAAP or IFRS line item as filed (e.g. "Net cash provided by
  operating activities").
- Compute every derived number in Python and show the code, or as
  spreadsheet formulas with cell references; never type a result.
- Tie out first: net income to the cash flow statement, ending cash to
  the balance sheet, assets = liabilities + equity. Report any
  mismatch before going on.
- Ratios with formulas, using [balances: period-end | average] balances
  where a flow is divided by a stock.
- DCF: FCFF = EBIT × (1 − t) + D&A − capex − ΔNWC, discounted at WACC;
  an inputs table (value, source, rationale); terminal value by
  [terminal value: Gordon growth | exit multiple]; the share of EV from
  terminal value; a 5 × 5 grid of value per share by WACC × g.
- Quality of earnings: CFO vs. net income, DSO and inventory days,
  capitalized costs, non-GAAP measures reconciled to GAAP, stock-based
  compensation, the auditor's report and Item 9A, each cited.
- No buy, sell or hold recommendation.
- Never invent numbers, names, quotes or sources: write [NEEDS SOURCE]
  where support is missing, and [VERIFY] beside any figure read from
  a table that was hard to parse.
</requirements>
 
<quality_bar>
Before answering, check your draft against:
- The statements tie, or the mismatch is reported first.
- Every figure has a citation, and the cited page contains it.
- EBITDA, FCF and debt mean the same thing in every period.
- Terminal growth is below long-run nominal growth of the economy,
  and the terminal share of EV is stated.
- Each red flag cites the note it came from.
</quality_bar>
 
<output_format>
A memo under 1,200 words: a 5-line summary; a tie-out table; a ratio
table (Ratio | Formula | one column per year | Source); DCF inputs,
output and sensitivity grid; red flags (Finding | Numbers | Citation |
Severity: low, medium or high); open questions. Code or spreadsheet
layout in an appendix.
</output_format>
 
<process>
Compute with:
[tool: Python, run by you | a script I run | Excel | Google Sheets].
Ask me up to 3 questions first if anything missing would change your
answer; otherwise go on.
End with a list of the assumptions you made.
</process>

This is a fresh chat. Audit the analysis below against the filings
it cites. Don't rewrite it, and don't add analysis of your own.
 
<materials>
[attach the same filings]
<brief>
[paste the stage 3 prompt]
</brief>
<analysis>
[paste the memo and its code or spreadsheet]
</analysis>
</materials>
 
<task>
Audit the analysis: which numbers, links and conclusions hold up, and
what to fix first.
</task>
 
<requirements>
- Pick the 10 figures that matter most to the conclusion; check each
  citation against the filing and recompute each in code.
- Check the tie-outs, units and periods, and that each definition is
  the same in every period.
- DCF mechanics: FCFF with WACC (not FCFE); terminal growth below
  WACC and long-run growth; terminal share of EV; mid-year convention
  stated; the grid's base cell equals the headline value.
- Red flags the analysis missed, each cited.
- Never invent numbers, names, quotes or sources: write [NEEDS SOURCE]
  where support is missing.
</requirements>
 
<output_format>
A check table (Item | Analysis says | Filing says | Citation | OK?),
then fixes ranked by how much each moves the conclusion, worst first.
</output_format>
 
<process>
Don't rewrite the memo. End with a list of the assumptions you made.
</process>

Attached: [filing, e.g. Acme's 10-K for fiscal 2025]. Answer only from
it: [paste your questions]. Cite item, note and page for each figure,
with a short quote; write "not found" rather than estimate.

<materials>
[attach the filing; for year-over-year questions, also attach the
prior year's]
</materials>
 
<context>
The filing is a [filing: 10-K | 10-Q | annual report | 20-F] for
[period, e.g. fiscal 2025, ending 30 June]. I'm reading it for
[decision: investment research | credit | supplier check | learning]
and I'm not an analyst: define each term once.
</context>
 
<task>
Answer my questions from the filing: [paste your questions, e.g.
revenue by segment and its drivers per MD&A; debt amounts, rates,
maturities and covenants; changes in accounting policies or
estimates; new or expanded risk factors vs. last year; anything in
the auditor's report beyond a clean opinion].
</task>
 
<requirements>
- Use only the attached documents, never memory.
- For every figure or claim: item, note number and page, plus a quote
  of 25 words or fewer.
- Name line items as filed; keep the filing's units and state them.
- Never invent numbers, names, quotes or sources: write [NEEDS SOURCE]
  or "not found" where the filing is silent.
</requirements>
 
<quality_bar>
Before answering, check your draft against:
- Each quote appears word for word on the cited page.
- Every figure carries its units and period.
- Management's views (MD&A, risk factors) are labeled as theirs.
- Anything inferred rather than read is marked as inference.
</quality_bar>
 
<output_format>
One section per question: the answer in 1–3 sentences, then a table
(Figure or claim | Value | Citation | Quote).
</output_format>
 
<process>
Find and quote the relevant passages first, then answer from them.
End with a list of the assumptions you made.
</process>

From these statements [paste 3 years as CSV, with units], compute
margins, ROE, ROIC, liquidity, leverage, DSO/DIO/DPO and CFO ÷ net
income in code, showing each formula and the code. Flag the three
ratios that moved most. Use no figure I didn't give you.

<materials>
[paste the income statement, balance sheet and cash flow statement
for 3–4 years as CSV, one column per year, with units and period ends]
</materials>
 
<context>
Company: [business, e.g. a regional food distributor]. Reports under
[standard: US GAAP | IFRS] in [units: thousands | millions].
I'm checking its health for this decision:
[decision: investment research | credit | supplier check | learning].
</context>
 
<task>
Build a ratio dashboard I can check line by line.
</task>
 
<requirements>
- Margins (gross, operating, EBITDA, net, FCF), ROA, ROE, ROIC
  (NOPAT ÷ average invested capital; state the tax rate), current and
  quick ratios, debt-to-equity, net debt/EBITDA, interest coverage,
  DSO, DIO, DPO, cash conversion cycle, CFO ÷ net income.
- Use [balances: period-end | average] balances where a flow is
  divided by a stock.
- Compute in code and show it, or give spreadsheet formulas.
- Tie the statements first and report any mismatch.
- Never invent numbers, names, quotes or sources: write [NEEDS SOURCE]
  where support is missing.
</requirements>
 
<quality_bar>
Before answering, check your draft against:
- Every ratio recomputes from the CSV.
- Definitions are the same in every year.
- Each reading compares with the company's own history.
- Moves are explained by line items, not adjectives.
</quality_bar>
 
<output_format>
A table: Ratio | Formula | one column per year | Trend | One-line
reading. Then the 3 biggest changes, the line items behind them, and
questions to answer from the notes.
</output_format>
 
<process>
Ask me up to 3 questions first if anything missing would change your
answer; otherwise go on.
End with a list of the assumptions you made.
</process>

Target [target, e.g. a regional food distributor], peers [peers, e.g.
6 listed distributors]; LTM data: [paste CSV: price, shares, debt,
cash, revenue, EBITDA, EBIT, net income]. Build a comps table: EV
(formula shown), EV/Revenue, EV/EBITDA, P/E, median and range. Say
which peers to drop and why. Use only my numbers.

<materials>
[paste one row per company, same period for all: share price and
date, diluted shares, debt, cash, preferred stock, minority interest,
revenue, EBITDA, EBIT, net income, revenue growth, margins, and the
filing each came from]
</materials>
 
<context>
Target: [target, e.g. a regional food distributor]. Peers: [peers,
e.g. 6 listed distributors named in its 10-K]. Period for every
company: [period: LTM | NTM | last fiscal year]. For this decision:
[decision: investment research | credit | supplier check | learning].
</context>
 
<task>
Build a comparable-companies table and a value range for the target.
</task>
 
<requirements>
- EV = market cap + debt + preferred + minority interest − cash, with
  the formula shown per row.
- EV/Revenue, EV/EBITDA, EV/EBIT and P/E; median, mean, low, high.
- Pair EV with pre-interest metrics and price with post-interest ones.
- Judge comparability on business model, growth and margins; drop
  peers that fail, say why, and recompute the medians.
- Apply the medians to the target for a value range by method.
- Compute in code or formulas; don't fetch or recall any number.
- Never invent numbers, names, quotes or sources: write [NEEDS SOURCE]
  where support is missing.
</requirements>
 
<quality_bar>
Before answering, check your draft against:
- Every company uses the same period and the same EBITDA definition.
- Negative or near-zero denominators are excluded, not averaged in.
- The median, not the mean, drives the range.
- The result is a range, with the method behind each end.
</quality_bar>
 
<output_format>
The comps table; the dropped peers with reasons; a range table
(Method | Low | Median | High) for the target.
</output_format>
 
<process>
Ask me up to 3 questions first if anything missing would change your
answer; otherwise go on.
End with a list of the assumptions you made.
</process>

Transcript attached: [company and quarter, e.g. Acme Q3 FY2026].
Summarize results vs. guidance, guidance changes (quoted), new or
dropped metrics, and the analyst questions management dodged. Quote
with speaker names; keep facts apart from forward-looking claims.

<materials>
[attach this quarter's transcript and last quarter's, plus the
earnings release if you have it]
</materials>
 
<context>
Company: [company and quarter, e.g. Acme, Q3 fiscal 2026]. I follow
it for this decision:
[decision: investment research | credit | supplier check | learning].
What I already watch: [metrics, e.g. same-store sales, gross margin,
net debt].
</context>
 
<task>
Write an earnings call read: what changed this quarter, and what
management avoided saying.
</task>
 
<requirements>
- Results vs. prior guidance; guidance changes, quoted.
- New, renamed or dropped metrics vs. last quarter's call.
- The 5 most important analyst questions and whether each was
  answered directly.
- Changes in tone vs. last quarter, each with the quotes that show it.
- Forward-looking statements labeled as management's.
- Never invent numbers, names, quotes or sources: write [NEEDS SOURCE]
  where support is missing.
</requirements>
 
<quality_bar>
Before answering, check your draft against:
- Every quote has a speaker and appears in the transcript.
- Numbers said on the call are marked [VERIFY] until matched to the
  release or the 10-Q.
- "Dodged" is shown by quoting both the question and the answer.
- Facts and management's outlook never share a sentence.
</quality_bar>
 
<output_format>
A 5-bullet summary, then sections for guidance, metrics, Q&A (table:
Question | Asked by | Answered directly? | Quote) and tone.
</output_format>
 
<process>
Go straight to the read.
End with a list of the assumptions you made.
</process>

Budget vs. actual for [period, e.g. October] [paste CSV: line, budget,
actual, prior year; units]. For lines over [threshold, e.g. 5% and
$10k]: the variance, a price/volume split for revenue, and one line of
driver, timing or permanent, and action. Show formulas; tie totals
to my CSV.

<materials>
[paste budget, actual and prior year by account as CSV, with units,
plus volumes and prices if you have them]
</materials>
 
<context>
Business: [business, e.g. a bakery with 3 locations]. Period:
[period, e.g. October 2026]. Readers: [readers, e.g. the owner and two
store managers, who read only the summary].
</context>
 
<task>
Write the monthly variance report with commentary.
</task>
 
<requirements>
- Variance = actual − budget, in amount and %, marked favorable or
  unfavorable.
- Revenue split into price variance (at actual volume) and volume
  variance (at budget price); the two sum to the total.
- Variable costs compared with a flex budget at actual volume.
- Commentary only for lines over [threshold, e.g. 5% and $10,000]:
  the driver, whether it's timing (will reverse) or permanent, the
  owner and the action.
- Formulas shown; totals tie to the CSV.
- Never invent numbers, names, quotes or sources: write [NEEDS SOURCE]
  where support is missing.
</requirements>
 
<quality_bar>
Before answering, check your draft against:
- Price and volume variances sum to the revenue variance.
- Each comment names a driver with a number, not "higher costs".
- Timing differences are labeled so nobody acts on them twice.
- The summary fits its readers: five bullets, no jargon.
</quality_bar>
 
<output_format>
A variance table (Line | Budget | Actual | Variance | % | F/U), the
price/volume and flex-budget tables, the commentary, and a 5-bullet
summary.
</output_format>
 
<process>
Ask me up to 3 questions first if anything missing would change your
answer; otherwise go on.
End with a list of the assumptions you made.
</process>

Project: [project, e.g. a $250k oven, 8-year life, 40k more loaves a
year]. Build an NPV model as spreadsheet formulas with an Inputs block,
a data table of NPV by discount rate and volume, a tornado ranking at
±20% and break-even volume. Every cell a formula, no typed results.

<materials>
[paste the quotes, cost estimates and volume assumptions you have,
with where each came from]
</materials>
 
<context>
Project: [project, e.g. a second bakery oven, $250k, 8-year life].
Drivers: [drivers, e.g. extra loaves a year × price; flour and energy
per loaf; one more baker]. Tax rate: [tax rate, e.g. 25%]. Discount
rate: [discount rate, e.g. 10%, our cost of capital]. The choice:
[decision: go or no-go | choose between options | timing].
</context>
 
<task>
Build an NPV model and sensitivity analysis for this project.
</task>
 
<requirements>
- An Inputs block (value, source, rationale), separate from the
  calculations.
- Incremental after-tax cash flows only, with the depreciation tax
  shield, working capital and salvage value.
- NPV and IRR; XNPV and XIRR if dates are uneven; remember that NPV()
  treats its first value as one period out.
- A data table of NPV across discount rate [range, e.g. 6–12%] and
  volume [range, e.g. −30% to +30%]; a tornado ranking of inputs at
  ±20%; break-even volume.
- Every cell a formula with cell references.
- Never invent numbers, names, quotes or sources: write [NEEDS SOURCE]
  where support is missing.
</requirements>
 
<quality_bar>
Before answering, check your draft against:
- Sunk costs are excluded and opportunity costs included.
- Cash flows and discount rate are both nominal or both real.
- Whether NPV changes sign inside the likely range is stated.
- The input that deserves more research before deciding is named.
</quality_bar>
 
<output_format>
A spreadsheet layout (Inputs, Cash flows, NPV, Sensitivity) as cell
formulas, then a half-page note for [reader, e.g. my business
partner].
</output_format>
 
<process>
Ask me up to 3 questions first if anything missing would change your
answer; otherwise go on.
End with a list of the assumptions you made.
</process>

My debts: [paste name, balance, APR, minimum payment for each]. I can
pay [amount, e.g. $900] a month in total. Compare avalanche and
snowball month by month as spreadsheet formulas: total interest,
months to debt-free, the month each is cleared. Check the math in code.

<materials>
[paste each debt: name, balance, APR, minimum payment, and any
promotional rate with its end date]
</materials>
 
<context>
I can pay [amount, e.g. $900] a month in total, maybe [extra, e.g.
$150] more from [month, e.g. March]. What I care about most:
[priority: least interest | a quick first win | lowest monthly stress].
I'm in the US and use [tool: Excel | Google Sheets].
</context>
 
<task>
Build a debt payoff plan comparing these methods:
[method: avalanche | snowball | both].
</task>
 
<requirements>
- A month-by-month schedule as spreadsheet formulas: interest at
  APR ÷ 12 on each balance, all minimums paid, the rest to the target
  debt (highest APR for avalanche, smallest balance for snowball).
- For each method: total interest, months to debt-free, the month
  each debt is cleared; then the effect of the extra payment.
- Check the totals in code if you can run it.
- No product recommendations (cards, loans, consolidation offers).
- Never invent numbers, names, quotes or sources: write [NEEDS SOURCE]
  where support is missing.
</requirements>
 
<quality_bar>
Before answering, check your draft against:
- Every balance reaches zero and never goes negative.
- Each freed-up minimum rolls to the next debt.
- Promotional rates end on their dates.
- If the minimums exceed what I can pay, the plan says so first.
</quality_bar>
 
<output_format>
A comparison table (Method | Total interest | Months | Order cleared),
the schedule layout as formulas, then 3 questions for a nonprofit
credit counselor if the numbers look unmanageable.
</output_format>
 
<process>
Ask me up to 3 questions first if anything missing would change your
answer; otherwise go on.
End with a list of the assumptions you made.
</process>

Audit this model [attach the file, or paste each sheet with formulas
visible]: hard-coded numbers in formulas, broken links, sign and unit
errors, a balance sheet that doesn't balance, cash that doesn't tie.
List each issue with its cell, severity and fix, worst first. Don't
rebuild it.

<materials>
[attach the spreadsheet, or paste each sheet with formulas visible,
plus the source documents it draws on]
</materials>
 
<context>
The model is a
[model: three-statement | DCF | budget | project NPV | loan schedule]
built by [builder, e.g. a consultant, or a chat assistant last week].
I'll use it to [decision, e.g. set next year's prices], and I can't
audit formulas myself.
</context>
 
<task>
Audit the model and tell me which problems to fix before I rely on
it. Don't rebuild it.
</task>
 
<requirements>
- Structure: inputs separate from calculations and outputs; no
  hard-coded numbers inside formulas; formulas consistent along rows.
- Integrity: statements tie (cash, retained earnings, the balance
  check); no unintended circular references; no #REF! or broken links.
- Logic: signs; units (thousands vs. millions); periods (monthly vs.
  annual rates); NPV() timing; terminal value inputs.
- Inputs traced to the source documents; untraced ones flagged.
- The 5 outputs that matter most, recomputed in code and compared.
- Never invent numbers, names, quotes or sources: write [NEEDS SOURCE]
  where support is missing, and [VERIFY] for anything you couldn't
  check.
</requirements>
 
<quality_bar>
Before answering, check your draft against:
- Each issue names its cell or row and shows the formula.
- Severity reflects the effect on the decision, not the count.
- Recomputed outputs sit beside the model's own.
- Nothing is called fine that wasn't actually checked.
</quality_bar>
 
<output_format>
An issue table (Cell | Issue | Effect on outputs | Severity: high,
medium or low | Fix), worst first; then the recomputation table.
</output_format>
 
<process>
Ask me up to 3 questions first if anything missing would change your
answer; otherwise go on.
End with a list of the assumptions you made.
</process>

Is Acme a good stock to buy right now?

<materials>
Attached: Acme's 10-K filings for fiscal 2024 and fiscal 2025, and
its 10-Q for Q1 fiscal 2026. Share price: $42.10 on 7 November 2026.
Diluted shares: use the EPS note of the latest 10-Q.
</materials>
 
<context>
Acme is a company I'm researching while learning to read filings;
I'm not an analyst, so define each term once. Acme reports under US
GAAP in US$ millions; its fiscal year ends 30 June. I can't judge
code, but I will spot-check numbers against the filing.
</context>
 
<task>
Analyze Acme from these filings and show what the current price
implies, so I can decide whether it deserves more research.
</task>
 
<requirements>
- Use only the attached filings; cite item, note and page for every
  figure.
- Compute everything in Python and show the code.
- Tie out the statements first and report any mismatch.
- A three-year table: revenue growth, gross and operating margin, FCF
  (CFO − capex), cash conversion (CFO ÷ net income), net debt /
  EBITDA, ROIC (NOPAT ÷ average invested capital; state the tax rate).
- Quality of earnings: receivables vs. revenue growth, capitalized
  costs, non-GAAP measures reconciled to GAAP, stock-based
  compensation, the auditor's report.
- Reverse DCF: the 5-year revenue growth the price implies at WACC
  9–11% and terminal growth 2–3%, as a sensitivity grid.
- No recommendation to buy or sell.
- Never invent numbers, names, quotes or sources: write [NEEDS SOURCE]
  where support is missing.
</requirements>
 
<quality_bar>
Before answering, check your draft against:
- Every figure traces to a cited page.
- Definitions are GAAP-based and the same in every year.
- Adjusted metrics appear only beside their GAAP reconciliation.
- Each bull and bear point rests on a cited number.
</quality_bar>
 
<output_format>
Tie-out; the three-year table; quality-of-earnings findings; the
reverse-DCF grid; 3 bull and 3 bear points. Code at the end.
</output_format>
 
<process>
Ask me up to 3 questions first if anything missing would change your
answer; otherwise go on.
End with a list of the assumptions you made.
</process>

TIE-OUT (Python, appendix A)
FY2025 assets 4,812 = liabilities 3,105 + equity 1,707        OK
Ending cash 412 (cash flow statement, p. 64) = balance sheet
cash 412 (p. 62)                                               OK
 
THREE-YEAR TABLE (US$ millions)         FY2023  FY2024  FY2025
Revenue growth                            9.1%    7.4%    5.2%
Gross margin                             41.0%   40.2%   38.9%
Operating margin                         12.3%   11.8%   10.6%
FCF = CFO − capex (p. 64)                  288     301     246
Cash conversion (CFO ÷ net income)        1.31    1.18    0.94
Net debt / EBITDA (EBITDA: p. 41)         1.6x    1.5x    1.4x
ROIC (tax rate 24%, Note 14, p. 91)      11.2%   10.5%    9.1%
 
QUALITY OF EARNINGS
- DSO up from 47 to 58 days while growth slowed (p. 58, p. 62)
- Capitalized software up 38% (Note 7, p. 79)
- Stock-based compensation excluded from adjusted metrics; diluted
  shares up 2.1% (Note 16, p. 95)
 
REVERSE DCF at $42.10: the price implies 6–8% revenue growth a year
for 5 years across the grid (WACC 9–11%, terminal growth 2–3%)...
 
Assumptions: FCF excludes acquisitions; tax rate 24% from Note 14...

Net debt / EBITDA uses adjusted EBITDA from the MD&A table on p. 41.
Recompute it in code for all three years with EBITDA = operating
income (income statement) + D&A (cash flow statement), citing each
input. Show both versions side by side, list the adjustments that
recur every year, and say which conclusions change. Leave everything
else as it is.

Follow-up critique prompts
 
- "Recompute every ratio in code from the raw numbers I gave you
  and list any that differ from your first answer."
- "For each figure, give the source line and page. Mark any you
  can't source."
- "Which three assumptions drive the valuation most? Show value
  per share if each is 20% worse."
- "Argue the bear case as a short seller using only the filing."
- "Which numbers are management's adjusted metrics? Reconcile each
  to GAAP."
- "What would you need to know about my situation before any of
  this applies to my own money?"