> ## Documentation Index > Fetch the complete documentation index at: https://docs.chicago.global/llms.txt > Use this file to discover all available pages before exploring further. # Data and Model Pipeline > Data foundation, machine-learning models, and evidence checks that scale the Parallax valuation framework across global equity markets The data pipeline supports consistent valuation across securities, currencies and accounting regimes. It prepares the financial inputs, routes companies to suitable methods, and checks the evidence used in published results. **The short version:** automated ingestion keeps the statement, market and estimate data current. A machine-learning layer predicts forward financial-statement line items such as operating margins, produces the factor signals the models consume, and adapts how that evidence is weighted. An automated routing layer sends every company to the valuation method its balance sheet actually justifies. Evidence checks then decide what is publishable. The workflow brings data preparation, model routing and publication checks into a common process. ## Consistency Across Markets Applying a valuation framework across markets requires consistent treatment of data, accounting conventions and business models. Method availability depends on the evidence available for each issuer. * **Alignment.** Price, statements, estimates, share count and exchange rates must describe the same economic moment for every company, not just the one being examined. * **Comparability.** Accounting bases, reporting perimeters and fiscal calendars differ across markets. A peer multiple is meaningless when numerator and denominator sit on different bases. * **Method selection.** A bank, a REIT, a mine and a software company do not share a valuation model. Choosing correctly, every time, is a routing problem. * **Evidence availability.** Publication checks assess whether the available inputs support each method. The pipeline applies these requirements throughout data preparation and valuation. ## The Data Foundation Aligned annual and interim periods across income statement, balance sheet and cash flow, reconstructed onto a common operating and financing basis. Daily prices, volumes, returns and volatility with multi-decade history, plus market-value capital weights and exchange rates. Dated sovereign curves, credit spreads, equity risk premiums and country risk evidence, carried with the date they were observed. Published analyst estimates and summary statistics, used as a market baseline and as one clearly labelled evidence source among several. **Ingestion on natural frequencies.** Every series refreshes on the schedule its source publishes. Market data moves daily. Statements and estimates move as they are filed and revised. The valuation layer reads from that base rather than from a snapshot someone took once. **Point-in-time discipline.** Each production run fixes one retrieval date, so current-snapshot evidence is never mixed across calendar boundaries. A valuation, its peer evidence, its consensus statistics and its volatility measure all refer to the same day. This is what makes a result reproducible and a historical comparison meaningful. ## Where Machine Learning Enters Machine learning does specific jobs in this framework. It does not replace the valuation identities and it does not choose the answer. | Layer | What the models do | What they do not do | | ------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------- | | **Statement forecasting** | Predict forward financial-statement line items, including operating margins, that seed the operating case, so a forecast starts from a data-driven baseline rather than a blank sheet | Set the final operating assumptions, which remain visible and inspectable | | **Factor signals** | Combine many underlying metrics into each factor score through a precision-weighted ensemble, so noisier inputs carry less weight | Forecast company cash flows or produce a price target | | **Score integration** | Adapt the weighting of factor evidence to the prevailing market regime within disclosed bounds | Override the valuation range or the evidence checks | | **Anomaly detection** | Flag trading, flow and data patterns that deserve attention before evidence is used | Delete an inconvenient observation | Machine-learning estimates reflect their training data and objectives. In this framework, they provide forecast baselines, signal weights and review flags. The operating assumptions and valuation calculations remain available for inspection. ## Automated Method Routing Every company is routed to an economic archetype before any model runs. The archetype determines the primary intrinsic lens, the peer-metric family and the companion variable the reader should check. A bank is valued on equity economics because deposits and borrowings are operating raw material. A resource company is valued on a finite reserve life rather than a perpetual annuity. A holding company is valued look-through rather than on consolidated revenue. Automated routing selects metrics appropriate to the business model. Where the available evidence supports a proxy, the output labels it accordingly. See [Archetype Conventions](/methodology/valuation/archetypes) for what each archetype means. ## Publication Checks Before any valuation output is published, its components must survive automated evidence checks. These verify that the security, currency and date are consistent, that a multiple and its target denominator sit on the same accounting and time basis, that the enterprise-to-equity bridge reconciles, and that enough usable observations exist to support the statistic being displayed. The checks determine which components support the published result. Each evidence source is checked on its own. One defective input changes which evidence can determine the result, and its failure reason stays in the audit trail. Every calculated figure is retained even when it is excluded from the published result. Readers can inspect the original calculation alongside the publication decision. When the evidence cannot support a method, the output says the method is unavailable and names what is missing. It does not substitute an economically unrelated metric to fill a panel. Where two qualified sources corroborate an extreme result, or remain irreconcilable, the case goes to human evidence review. The reviewer chooses among existing evidence or declines publication. The reviewer cannot enter a bespoke number. ## The Reconstruction Test The pipeline is built so that a reader can reproduce the result. Given revenue, margin, reinvestment, WACC, terminal growth, net debt and diluted shares, a reader should be able to reproduce the order of magnitude of intrinsic value with a calculator. Given the peer evidence and consensus statistics, a reader should be able to reproduce the market-evidence result and identify every judgment that separates the raw calculation from the published one. Calculations retain full precision internally. Displayed client figures are rounded for readability. The calculation trail helps readers assess the inputs, assumptions and publication decisions behind each result. ## Analytical Roles * The forward forecast sets out the operating case. The reverse DCF provides a separate comparison with price-implied expectations. * Intrinsic valuation and market-evidence targets have distinct roles. A target that combines compatible sources identifies its inputs, weights and exclusions. * Technical interpretation is grounded in the supplied price and volume metrics. * Peer outliers are flagged and explained, with the observations retained in the evidence. ## Related Pages How factor scores are calculated and integrated into an overall recommendation. The statement reconstruction that the pipeline performs before any forecast begins. How a company's balance sheet determines which valuation method applies. What the pipeline finally publishes, and how to read it.