Financial Analysis

• What this produces: written analysis. Tables carry the numbers, prose carries the judgement, and the deliverable ends in a position — a valuation range, a recommended debt level, an invest/pass call — that a reviewer can attack.
• What this is NOT: a spreadsheet build. If the user wants a .xlsx with live formulas, named ranges, and a Checks tab they can open in Excel, that is financial-modeling. If they want a number and the reasoning behind it, it is this skill. When genuinely ambiguous, ask which they want before starting.
• Not a data source. These playbooks structure analysis over figures the user supplies or that are verifiably retrieved. Never invent a comparable company, a transaction multiple, a market benchmark, or a source citation.

1. Identify the analysis from the catalog below — match intent, not the literal model name. "Can it carry another turn of debt" is credit capacity; "should we pay 12x" is comps plus precedents.
2. Load the one reference file for that family.
3. Gather the inputs that playbook lists. Ask for what is missing. Proceed with at most a handful of assumed inputs, and only when each is defensible — if the core drivers are absent, say so and ask rather than manufacturing a model.
4. Run Phase 1 before any number. Every playbook opens with a framing phase that fixes methodology and names the gaps. Skipping it produces output that looks finished and cannot be defended.
5. Deliver against the playbook's required tables and its challenge test.

These compose. Sum-of-the-parts consumes comps and DCF per segment. An IC memo consumes an LBO or DCF plus comps and precedents. Sensitivity attaches to any model with a single key output. When a playbook needs another, run the dependency first and carry its output forward rather than re-deriving it inline — and where the dependency is missing, build it rather than assembling over the gap.

Phase 1 before numbers. Each playbook opens by fixing methodology: which definition, how many stages, which screen, what is excluded and why. State the data gaps here and what you will assume to close them. This phase is part of the deliverable, not scratch work.

Label every assumption. Any input not supplied by the user is marked [ASSUMED] at the point of use with a one-line reason. A reader must be able to separate what they gave you from what you invented on their behalf. Do not use round numbers unless the source data is round.

Report checks as figures, not ticks. Where a playbook specifies an integrity or quality check — statements tying, terminal-value methods reconciling, peer set depth, addback scrutiny — report the actual number, including when it is uncomfortable. A checkmark hides a failure; a figure does not.

Scenarios are narratives. Bear, base and bull describe coherent states of the world with a cause, not percentage haircuts applied to the base case. If the bear case is base minus ten percent across every line, it is not a scenario.

Tables carry the analysis. One short paragraph of commentary per section at most. No opening summary of what you are about to do and no closing restatement of what you did. The exception is the IC memo, whose prose sections are specified in its own playbook.

Hold the scope. Deliver the analysis asked for. A DCF request does not get an LBO attached, a comps request does not get a credit view. Note in one line where an adjacent analysis would change the conclusion, and stop there.

Close by challenging the work. Every playbook ends with the assumptions that would most change the output if wrong, and the level at which each one flips the answer. Where a number is genuinely unknowable, say so rather than picking a plausible-looking value.

These are long-reasoning tasks. Where the environment exposes an effort or thinking-budget control, the model-building and returns playbooks — DCF, three-statement, accretion/dilution, LBO, credit capacity, sum-of-the-parts, operating model, IC memo — want the highest setting available. Comps, precedent transactions, IPO pricing and sensitivity analysis run well one step below. This is a setting the user controls, not something the skill can set; mention it only if output quality is being limited by it.

Before delivering, confirm:

1. Phase 1 is present and states methodology, exclusions, and data gaps.
2. Every assumed input is marked [ASSUMED] with its reason, at the point of use.
3. The playbook's check is reported as figures for every period or line it covers.
4. The bridge ties where one exists — enterprise to equity, sources to uses, statements to each other — and the residual is shown, not asserted to be zero.
5. No fabricated externals. Every peer, transaction, benchmark and rate is either user-supplied, retrieved, or explicitly flagged as an illustrative placeholder.
6. The recommendation is stated, with the condition that would reverse it.
7. Scope held — nothing delivered beyond what was asked.