How compounders.ai reads a serial acquirer
Every figure on compounders.ai is read from a company's own filing and published with its citation. Normalized EBITA applies one deterministic rulebook to every company: it strips what is an artifact of having acquired and keeps what is a real ongoing cost.
Where does compounders.ai's data come from?
From the companies' own annual filings, fetched from the official source: the 10-K, 40-F or 6-K on SEC EDGAR, the ESEF or UKSEF report package the company publishes, and, where a company files PDFs, its annual report. Each document is stored unchanged and identified by its hash, so every figure traces to the exact document it was read from. compounders.ai does not take financial figures from data vendors, aggregators, press coverage or a model's general knowledge.
How does a filing become a published figure?
- Fetch the filing. The 10-K from EDGAR, the ESEF or UKSEF package from the company, the PDF from SEDAR+ or the company's own site. The bytes are stored and hashed.
- Read the facts. Inline-XBRL facts are read exactly as tagged, with their scale, sign and period. PDF reports are parsed into tables, and a row is mapped to a statement line by a reviewed dictionary; a mapping a model proposes is checked by a person or a deterministic rule before it can publish. Nothing is typed in by hand.
- Cite every number. Each fact becomes a citation: the document, the element or page, and the text as filed. The database refuses a figure without one.
- Apply the ruleset. Deterministic rules classify each item. Where a rule needs a judgement the filing does not support, the item goes to review or the year is withheld with its reason, instead of being guessed.
- Publish through procedures. Database procedures publish a figure only with its citation, its ruleset version and its lineage. A re-run supersedes a figure; it never edits one.
What is normalized EBITA?
Normalized EBITA is reported operating profit adjusted under compounders.ai's published ruleset, so that it measures the operating earnings of the businesses a company owns without the accounting artifacts of having bought them. It adds back the amortization of intangibles created by acquisitions (hence EBITA: earnings before interest, tax and the amortization of acquired intangibles), removes the other purchase-accounting effects in both directions, and removes genuinely one-off charges unless they recur. It keeps every real cost: stock-based compensation, the amortization of software and development a company capitalizes itself, and the costs of making acquisitions.
Each adjustment is a cited row on the company's Normalization tab, so the bridge from reported operating profit to normalized EBITA can be checked line by line.
What are the seven rules?
The rulebook is the same for every company. The current version is 1.4.0.
Rule 1 — Acquired-intangible amortization: added back
Amortization of customer relationships, technology, order books and trade names created by purchase accounting is a non-cash artifact of having acquired, and the largest single reason reported earnings misprice serial acquirers. It is added back. The useful lives the company uses are shown beside the add-back, so a long life cannot hide a small add-back. Where a filing does not split acquired from self-generated amortization, the split is not guessed.
Rule 1b — Other purchase-accounting effects: removed in both directions
An earn-out (contingent consideration) remeasurement is a change in the price paid, not an operating result: a charge is added back and a credit is removed, with no recurrence test. The fair-value step-up of acquired inventory, released through cost of sales, is added back. The release of a loss-contract reserve recorded at acquisition, the same effect with the opposite sign, is removed. A payment to a seller that depends on the seller staying employed is pay for services and is kept.
Rule 2 — Capitalized software and development amortization: kept as a cost
The amortization of software and development a company capitalizes itself is ongoing reinvestment, a real cost of staying in business, and stays in. Where one line mixes it with acquired amortization and the notes do not split them, the split goes to review instead of being guessed.
Rule 3 — Stock-based compensation: kept as a cost
A real cost of paying people and a real source of dilution. Management's adjusted figures that exclude it are never used as the normalized figure.
Rule 4 — One-off items: removed unless they recur; acquisition costs kept
Impairments, restructuring and one-time litigation are removed. Any item type a company labels one-time in four of the last five years is treated as recurring and stays in. Acquisition transaction costs (advisers, due diligence, the integration a deal needs) are kept for every company, with no recurrence test: acquiring is the business model, so its cost is an operating cost, and the amount stays visible on the ledger. A share of an equity-accounted investee's result inside operating profit is carved out so margins stay comparable.
Rule 5 — Growth: organic, acquired, currency and price
Revenue growth is split into its components from the company's own growth disclosure, and the company's definition of organic growth is shown beside the number, because companies define it differently. An organic figure is computed only against a base the filing discloses.
Rule 6 — Returns on capital
Normalized NOPAT over average invested capital, with and without goodwill, both shown. The three-year incremental return (ROIIC) and the return on acquisition capital show whether the acquisition engine still earns its returns at scale. Each is published only when every input is cited.
Rule 7 — Leverage and acquisition multiples
Net debt to EBITDA, with and without lease liabilities, the capital deployed on acquisitions, and the multiples paid where the filings disclose them. Rising multiples and rising leverage with falling returns on acquisition capital is the pattern of a roll-up reaching for growth.
What does compounders.ai never do?
- Estimate, interpolate or model a financial value. A missing input is a missing output, labelled with the reason.
- Use management's adjusted figures as the normalized figure. Where a company prints its own adjusted measure, it is shown separately, as printed, with the gap to normalized EBITA.
- Strip the cost of acquiring.
- Guess a split the notes do not give.
- Cite an aggregator. The source of every number is the filing itself.
- Convert currencies inside a published number.
- Change a published figure in place. New evidence or a new ruleset publishes a superseding figure, and the history is kept.
How are each company's own KPIs captured?
Each archetype is judged on its own operating measures: revenue by the company's own classes for software roll-ups, locations for distributors, contingent commissions for brokers, the company's own margins, returns and cash conversion where it prints them. Each is captured as the filing prints it, cited to its table row, sentence or tagged fact, beside the company's own definition, and published only when the filing's own arithmetic checks it: classes that add up to the printed total, a rollforward that foots, a ratio reproduced from the inputs the same table prints. A measure the filings do not print is shown as not printed, never filled in. Two companies' measures under one name are compared within an archetype, never combined.
How are per-share figures and share counts handled?
Share counts come from each company's own filings and are published only where the filing's arithmetic proves them: a rollforward that foots, or counts that tie to a printed total. No figure is adjusted by a split ratio; a † marks a per-share figure on an earlier share basis than the latest year. Normalized EBITA per diluted share is consolidated, minorities included; it is withheld in a year where minorities take 5% or more of the profit or redeemable minorities 5% or more of equity, and wherever an earlier filing printed the share count differently with no registered split to explain it.
What does the auditor's report section show?
Each company's Normalization tab quotes its independent auditor's report as the filing prints it: the firm as signed, and every critical or key audit matter's title, description and response in full, each with its citation. Auditors often name exactly the acquisition figures that took the most judgement, so the page shows them beside the ledger. Nothing is scored, summarised or machine-translated.
How are restatements and corrections handled?
When a later filing restates a year, the later filing's figure stands and the page marks the year: ↻ where the figure was re-presented, ≈ where the earlier filing printed it at lower precision, † where a per-share figure sits on an earlier share basis. Correcting or re-running a figure publishes a new row that supersedes the old one; nothing is edited in place, and the earlier rows stay in the history.
How are the rules versioned?
A ruleset version is identified by the SHA-256 hash of the document that defines it, and every normalized figure names the version it was made under. When the rules change, every figure is re-published under the new version, and the earlier version's figures stay visible in each company's ruleset history. Version 1.4.0 is current; the live Methodology page lists every version with its hash and change notes.