Valuation Reporting

Valuation reporting

The analysis is done. Every number sits in an artifact on disk. What remains is the part the decision-maker actually reads, and it is the part most often written badly.

Two failures dominate. The first is the data dump: forty pages of tables with the answer buried on page thirty-one. The second is false confidence: one number to the cent, no range, no disclosure of what the critic found. This skill exists to prevent both.

The report is written by the investment-reconciler, which owns 11-verdict/REPORT.md and 11-verdict/verdict.json. It reads every other artifact and writes no other file. That ownership rule has a direct consequence for style, covered under "Pull, never recompute" below.

The templates

One template per analysis mode. Read the one that matches mandate.json.mode before drafting. Each gives the sections in order, what belongs in each, and which artifact supplies each number.

Two method files support every template:

• resources/value-bridge-and-range.md — laying out the bridge, choosing and presenting the range, value against price.
• resources/disclosure-and-vintage.md — the pivotal assumption ledger, unresolved findings, data gaps, source vintages.

Six disciplines, every mode

1. Lead with the answer

The first screen carries the recommendation, the number behind it, the range around it, and the one sentence that says why. Assume the reader stops there. Everything after the first screen exists to let a sceptic audit the claim, not to build up to it.

A report that opens with company history has buried its own finding.

2. Show the bridge

A single value per share tells nobody where the value came from. The bridge does. In an intrinsic valuation it is the ladder from operating assets to value per share. In a restructuring or acquisition it is the value stack: status quo, plus control, plus synergy. Both are laid out in resources/value-bridge-and-range.md.

Show the line items with their bases. Most disputes between competent analysts happen in the bridge, not in the growth forecast, and a bridge on the page is what lets a reader find the line they disagree with.

3. Name the two or three assumptions the answer turns on

Every valuation has dozens of inputs and two or three that decide it. Find them in the sensitivity block of dcf-result.json: they are the inputs whose plausible range moves value enough to change the recommendation. Name them, give each a value, a source and a range, and say which way the answer flips.

Everything else goes in an appendix table. Listing forty assumptions with equal weight tells the reader you have not worked out which ones matter.

4. Present a range, not a point

The payoff from a valuation is largest exactly where precision is lowest. A point estimate to the cent claims a precision the inputs cannot support, and a reader who spots that distrusts the rest.

Report a base case with a range around it, and label the range with what produced it — a two-input sensitivity grid, a named scenario grid, or simulation percentiles. Rule out the cells the narrative cannot support and say why you ruled them out. A grid with impossible cells left in it is not a range, and neither is a wish list of scenarios without likelihood labels.

Round to the precision the inputs justify. Value per share to the cent is defensible only when the share count and the bridge are exact to the cent, which is rare.

5. Disclose unresolved critic findings

G7_challenged passes when every high-severity finding in challenge.json is resolved or explicitly disclosed. The loopback rule caps reruns at two per stage. A finding that survives the third attempt is disclosed in the report as an unresolved risk.

Disclosure is a named section, not a footnote. Give the finding's id, its claim, the stage it targets, what was attempted, and how the value would move if the critic is right. A report that quietly drops a finding the critic raised is the one failure that destroys the reader's trust in every other number.

6. Record the data vintage

Every external input has a date. The riskfree rate, the equity risk premium, the default spread table, the country risk premium table, the industry averages, the market price. Mixing vintages is the most common silent error in this domain, and the report is where it becomes visible or stays hidden.

Carry one vintage table. Name the as_of field of every bundled reference table used, and the valuation_date from mandate.json. Note anything more than a year stale and say what it would change. List the entries from gaps.json with the fallback used for each.

Value against price, without pretending certainty

The gap between value and price is the point of a valuation mandate, and it is the easiest place to overclaim. Report it in four steps, in this order.

1. The gap. Value per share, price per share, the difference, and price as a percent of value. Below 50% or above 200% is a prompt to re-examine your own inputs. Say that you did, and what you found.
2. The implied input. Solve for the growth rate or margin that makes model value equal the market price. That number is what the market is assuming. Then ask whether it is probable, not merely possible. Some scenario justifies any price.
3. The closing mechanism. A gap with no mechanism and no horizon is an opinion, not a trade. Name what closes it: an earnings report, a product milestone, an activist, an acquirer, an index event, a refinancing. If you cannot name one, say so plainly and let the recommendation carry that weight.
4. The margin of safety. The gap net of the range. If the conservative end of your range still sits above the price, say it — that is a far stronger argument than the base case alone. If it does not, the recommendation rests on the base case and the reader deserves to know.

Never average a DCF value with a multiple-based price into one number. Report both, explain the difference, and say which one carries the decision and why. Where the intrinsic and relative answers disagree, the explanation is the finding.

Say plainly that the value estimate moves too. Intrinsic value changes as information arrives; today's gap is not tomorrow's.

Pull, never recompute

Every number in the report comes from the artifact that owns it. The report reports; it does not calculate. A figure in the executive summary that disagrees with the same figure in the body is a validation failure, not a rounding difference.

This applies to derived figures as well. If the gap between value and price appears twice, it is computed once and quoted twice. When a number needs recomputing, the owning stage reruns and rewrites its artifact; the report then picks up the new value.

Arithmetic that genuinely belongs to the report — a percentage gap, a per-share conversion, a sum of disclosed line items — runs through the computation skills rather than in your head. valuation-consistency-checks validates the assembled set before the report ships.

What the verdict carries

verdict.json is the machine-readable half of the same deliverable, and G8_reconciled passes only when it states value against price with a margin of safety. Full schemas live in contracts/. The fields the gate and the report both need:

• mode, company, currency, valuation_date — copied from mandate.json.
• recommendation — the call, in the vocabulary of the mode.
• value and price — with the unit stated, and the range around the value.
• margin_of_safety — the gap net of the range.
• key_assumptions — the two or three, each with value, source and range.
• catalysts and risks — each with a horizon.
• unresolved_findings — the id and claim of every disclosed finding.
• data_vintage — the as_of of every reference table used.
• confidence — high, medium or low, with the reason.

Keep verdict.json and REPORT.md in agreement figure by figure. They are one artifact in two formats.

Writing the prose

State the finding, then the evidence. Not the reverse. A paragraph that walks through the method before arriving at the conclusion makes the reader do work you should have done.

Give every number its unit and currency on first use in a section. A table of figures with no currency header is unusable in a multi-currency mandate, and the mandate currency is binding on every rate and cash flow in the analysis.

Hedge once or not at all. "Roughly 12%" is honest. "Approximately around 12% or so, subject to considerable uncertainty" is noise, and it reads as an attempt to avoid being wrong rather than an attempt to be useful.

Attribute judgment calls to the judgment, not to the machinery. Write "we set the terminal return on capital equal to the cost of capital because no moat survives ten years in this business", not "the model assumes convergence". The second sentence hides a decision somebody made.

State your own bias exposure in one line where it exists: who commissioned the work, what answer they want, whether the market price was seen before the qualitative work was done.

What sinks a report

• The answer arrives on page thirty, after the methodology.
• A point estimate with no range, or a range with no likelihood labels.
• Forty assumptions listed with equal weight and no statement of which two decide it.
• A high-severity critic finding that appears nowhere.
• A riskfree rate from this year paired with a country risk table from four years ago.
• The summary and the body disagreeing because the summary was recomputed.
• A DCF value and a peer multiple averaged into one number.
• A buy call with no catalyst and no horizon.
• Value per share quoted to four decimals off a share count rounded to millions.
• Control value, synergy value and a strategic premium stacked on one DCF.
• The offer price presented as the value, or an exit multiple presented as intrinsic.

Where the detail lives

• Report structure per mode: the six template files above.
• Pipeline stages and their outputs: knowledge/frameworks/intrinsic-valuation-playbook.md (S15–S18) and knowledge/frameworks/corporate-finance-playbook.md (S12).
• Gap, expected return and the three investor questions: knowledge/concepts/narrative-numbers/value-vs-price-gap.md.
• Precision, bias and the small-model rule: knowledge/concepts/narrative-numbers/valuation-misconceptions.md.
• Range construction: knowledge/concepts/deliverables-worked-examples/dcf-sensitivity-analysis.md, knowledge/concepts/narrative-numbers/narrative-scenario-grids.md, knowledge/concepts/narrative-numbers/monte-carlo-valuation-simulation.md.
• Reconciling conflicting estimates: knowledge/concepts/deliverables-worked-examples/valuation-triangulation-and-recommendation.md.
• The one-page scorecard: knowledge/concepts/deliverables-worked-examples/project-executive-summary-scorecard.md.