Skip to content

Data & Methodology

Audience: Users · Version 0.2 (draft) · Updated 2026-06-11

Where Capital Workbench's numbers come from, how current they are, and why some values are blank. For exact metric definitions, see the KPI Glossary.

Where the data comes from

All financial data is derived from publicly available company filings — there are no private or estimated figures. CW combines two layers of that public data:

  • Structured financial statements — the standardized numbers companies file (income statement, balance sheet, cash flow).
  • The underlying filing documents — the filings themselves and their tagged facts, used for filing lookups and to broaden coverage.

Data source

Every figure in CW is derived from publicly available company filings — there are no private or estimated numbers. CW presents public data only and is not investment advice.

How the data is processed

In plain terms, raw filings become the numbers on the page in a few stages:

  1. Import — raw filing data is ingested from the public sources.
  2. Normalize — each company's filing tags are mapped to a consistent set of concepts, so "revenue" means the same thing across companies that tag it differently.
  3. Chart of accounts (COA) — normalized facts are organized into CW's standard line-item structure that the financial statements and KPIs read from.
  4. KPI computation — KPIs (margins, working-capital days, cash ratios, etc.) are computed from those line items.
  5. Serving — pages and the AI assistant read the computed values.

Because filers tag their statements differently, some line items require mapping work before they appear. When a company files something CW hasn't mapped yet, the value can show as pending rather than wrong — see below.

How current is the data?

Filings are published on the companies' own schedules and then processed, so values can lag the latest calendar quarter, sometimes by one to two quarters. A few practical consequences:

  • The "latest" figure for a metric may be from an earlier period than you expect; CW labels the period it's showing.
  • In comparisons and reports, two companies may have different most-recent periods (and different fiscal year-ends), so "latest vs latest" isn't always the same calendar date.
  • Reports note when a metric's latest available period is older than the report's reference period ("sources can lag by one or more quarters").

Admins: freshness also depends on the data pipeline and the wallpc→Render sync having run after the latest backfill. A target staleness SLA to promise users is still an open product question (TBD).

Why is a value blank?

A missing or "Not reported" value almost always means the company doesn't disclose the underlying line item — not that CW failed. Every KPI carries a plain-language reason that the app and the AI assistant quote verbatim. The common cases:

What you see What it means
Not reported ("Not reported by issuer") The filer doesn't disclose any tag CW can use for this metric.
Mapping pending The filer discloses it, but under a tag CW hasn't activated yet — coverage will improve as mappings are added.
Not applicable ("Excluded by industry rule") The metric isn't meaningful for this company's industry (e.g. inventory-days metrics for a bank).
Pending The value hasn't been computed yet.
error ("Formula error") A computation problem; rare, and surfaced rather than hidden.

When a metric is computed, the reason instead explains its basis (e.g. "Reported from …", "Computed from …", or "Computed via fallback path").

You can ask the AI Agent "why is X blank for this company?" and it will quote the specific reason.

Restatements & data nuances

  • Some line items can be flagged as divergent when filings disagree across sources; CW surfaces this rather than silently picking one.
  • Working-capital day metrics (DSO/DIO/DPO/CCC) use period-length conventions; the exact basis is noted in the KPI Glossary.
  • EBITDA is computed (operating income plus depreciation & amortization, with a fallback) and is therefore unavailable when its inputs aren't disclosed — which is why some EBITDA-based ratios on Cash Analysis can be blank.

How peers are chosen

Peer Benchmarking selects comparison companies in this order:

  1. Curated peers for the company, if defined.
  2. Same-sector companies (up to a handful), otherwise.
  3. A general fallback if no sector is available.

Important caveat (shown in-app): the current peer selection is sector-based and is acknowledged as a placeholder; a curated peer framework is in development. Treat the specific peer set as illustrative, not authoritative. When the sector is missing, peers fall back to a non-industry-matched set and are flagged as low quality — you can supply your own tickers for a better comparison (including by asking the AI Agent to compare specific companies).