# Praecise Metal

> Praecise Metal is the bare-metal business of Praecise: the software and
> infrastructure to run AI on metal you
> own: measurement, forecasting, placement, power and thermal control,
> connectivity and cost. The operating capability that otherwise takes years to
> build, so the hardware you bought does the work you bought it for.
> The software is Praecise Furcate (https://furcate.xyz), built on the open
> Furcate OS (https://furcate.org), which Praecise develops and stewards.

Every watt is a decision made in software. It starts at the instrument, because
optimising a number nobody has read is theatre.

## Measurement

Processor counters are free and cover the package only — not disks, fans or
conversion loss. On the first fleet we instrumented, the declared multiplier
used to convert processor draw into wall draw was wrong by four to nine times,
in the direction that flattered the report. A metered PDU or a UPS with power
reporting is the highest-value hardware on an efficiency programme.

Measured on a running node: 13 W idle, 29 W under inference, against a package
cap of 140 W that has never been approached, and 1.08 J per token. What binds
is idle, at 228 kWh a year for two always-on machines before a request arrives.

## What we build

Telemetry · Forecasting · Placement and consolidation · Power control ·
Thermal and cooling · Carbon and tariff-aware scheduling · Connectivity ·
Reporting · Site design.

Software and infrastructure sit in one place because most of the efficiency
lives in the seam between them. A control loop is only as good as the meter it
reads, and a meter is only worth installing if something acts on it.

## Ownership and edge cells

One operator can own core clusters and edge cells at home, business and city
scale. Praecise Metal is the operator's full ownership view of all of it:
inventory, health, power and cost for every machine and every cell, read from
the operator's own chain of keys. It reads ownership and never controls a cell.

- A cell is a complete cluster with its own head and its own control, and keeps
  running when its link is cut.
- Cells nest: a city cell can hold district and business cells, each in charge
  of itself. Ownership rolls up; control stays local.
- Scoped: a cell and its managers see only that cell and the cells inside it,
  never the rest of the infrastructure, enforced by the machines. The same
  holds for cells that sit on Praecise.
- Status: the first edge cell runs today (gateway key held in its own silicon,
  the cell's authority in a controller's TPM). Nested cells, scoped managers
  and cell roll-ups in the ownership view are coming.

## Refusals

Every system returns nothing where it does not know, and says which kind of
nothing it is.

- No facility meter means no PUE. The overhead is recorded as unaccounted; a
  published 1.0 claims the overhead was measured and was nil.
- Where imports are material and their origin undeclared, no carbon intensity
  is returned. Treating unknown imports as zero-carbon is how a coal-backed
  third of a grid disappears from a report while the arithmetic stays correct.
- Below a 25 per cent spread between cheap and peak hours, deferral is refused
  rather than an arbitrary best slot being picked.
- A minute not covered by any tariff period is not priced. Charging zero for
  uncovered hours makes them the cheapest of the day.
- On-site and source water are carried separately and never summed.
- An undeclared limit draws no bar. No denominator is not zero headroom.

## Carbon

The standard behind software carbon intensity is deliberately location-based:
it asks what the wire actually carried and does not accept a national
generation average. A grid generating at 25 gCO2eq/kWh, almost entirely hydro,
can consume at 203 gCO2eq/kWh once a third of its supply is imported from
coal-dependent grids nearby. The gap is widest in the dry season, when the
clean-grid assumption is least true and the reporting looks best.

## Forecasting

Quantile paths rather than point forecasts, with intervals from each method's
own residuals rather than an assumed normal. A foundation model never simply
replaces the baseline: both run, both are backtested on rolling held-out
windows of the site's own data, and the model is used only while it is
measurably ahead — a 10 per cent margin to be adopted, merely not losing to be
kept. Scale-free error for the median, pinball loss for the quantiles, eight
windows minimum before anything is acted on.

## Reporting

Data centres above a few hundred kW of IT power increasingly report annually on
energy and sustainability indicators, with rating schemes and minimum standards
following. We instrument the site so the same readings drive the control loop
and populate the return. Where a figure cannot be produced, we install the
instrument or report the indicator as unaccounted.

## Engagements

- Survey — instrument what exists, report draw, cost, emissions, and which
  figures cannot yet be produced. Ends with a metered baseline.
- Programme — efficiency levers priced individually against that baseline,
  applied in order of return, each one reversible.
- Software — control loops, forecasting, placement, telemetry and reporting
  built into operations rather than sold as a dashboard.
- Design — new sites and retrofits: layout, power distribution, failure
  domains, cooling strategy and growth increments.

operations@praecise.com