Corporate pulse: semiconductor capacity capex growth

Corporate pulse: semiconductor capacity capex growth, version 1.0

Classification. Derived: deterministic arithmetic over stored values; every number is reproducible from the stored ids named in the observation's metadata.

Formula. Per company, capital expenditure over the trailing four quarters ending at its latest reported period end E, and the same construction ending one year before E, both from stored 10-K and 10-Q facts in USD (pulses.v2: the annual value when a fiscal year ends at E, otherwise year-to-date to E plus the prior fiscal year minus the year-to-date for the same span one year earlier). Value = 100 x (sum of latest over contributing companies / sum of prior over the same companies - 1). The period is the latest E among contributors. Custom function corporate_pulse over the dataset built by loadCorporatePulseDataset.

Engine rendering: The custom function corporate_pulse over the whole component dataset, for each geography and period. A period with any component missing produces no value. Rounded to 2 decimals.

Components.

• No component metrics: the inputs are stored rows outside the observations table, named below and recorded on every row.

Windows and alignment. Read from the stored corporate_pulses rows for group semiconductor_capacity (written weekly by corporate.recompute_pulses): one observation per distinct as_of, from the most recently computed row for that as_of. The row id is on every observation.

Missing data. A period is produced only when every component has a value for it; nothing is interpolated, carried forward or estimated.

Rationale. Corporate Physical Economy Pulse (docs/upgrade/20) aggregates operational signals from public companies as sensors for physical systems. Capital expenditure is the component with the cleanest XBRL tagging and the most direct link to capacity being built, so version 1.0 publishes it alone; inventory and PP&E components stay in the pulse rows for a later version.

Limitations. A matched sample of large SEC registrants, not the economy: the companies are chosen as sensors for physical systems and the aggregate is dominated by the largest spenders. Capital expenditure is global for most filers, so this is a US-listed sample, not US investment. Fiscal years differ, so the period is the latest period end among contributors and companies contribute values ending up to a year apart. Nominal dollars. At least two contributors are required for a value; the sample size is on every row. The inputs are pulse rows and XBRL facts, not observations, so the reproducibility harness cannot rebuild the value from observation ids; the pulse row id and the contributors' values are the provenance.

Revision. Recomputed at every derived pass; a changed component value produces a new observation row and earlier rows are kept. A change to anything that affects the numbers is a new formula version (the spec hash is on every row and a changed spec under a used version is refused).