What AI Is Made Of

Run 04 modeled engineering scenario

Run 04 calculates equipment material stocks using the Run 03 reference plant, explicit engineering assumptions and reproducible model inputs.

Three distinct records

  1. Run 03 evidence-qualified views: unchanged data, admission rules and historical physical results. Their open material balances remain open under that standard.
  2. Historical Run 04 report: a user-pasted report recovered from the MBI conversation. The available excerpt is capped at 20,000 characters and ends mid-sentence. Its numbers are unverified historical claims. They are stored and compared, not used as model calibration targets.
  3. Modeled Run 04 scenario: explicit physical constructions, proxies and analyst assumptions applied to the frozen equipment populations. This output has its own identity and reproducible inputs. Numerical agreement with the report is not independent empirical validation.

Scenario admission, proposed v0.2 extension

An equipment material estimate may be admitted to a named engineering scenario when its population, units, scope, physical construction or composition, source basis, transfer rationale and limitations are explicit. Admission means the calculation is inspectable, not that the modeled construction is an observed installed product. It must not silently enter a prior evidence-qualified subtotal.

Keep source evidence and target application separate. Manufacturer evidence about a comparison product remains evidence about that product. Scaling or transferring it to another rating or design is a proxy, with its own uncertainty. Analyst selections must not acquire a manufacturer evidence grade by proximity to a citation.

Low/base/high results are engineering scenarios or envelopes, not statistical confidence intervals. Alternative architectures replace affected constructions; they are not additional stock. The output must identify whether cases are jointly buildable states or material-specific marginal envelopes. Do not sum marginal endpoints into a simultaneous plant mass.

Accounting

Normalize to 1 GW commissioned IT electrical capacity, using the inherited fixed plant counts including selected redundancy. Separate incremental B1 compute, incremental B2 facility and incremental modeled B3 enabling equipment. Sum those increments once.

Copper-grade stock and aluminum alloys are industrial material categories. Ferrous material classes must be disjoint; a linked ferrous total is not another material to add. Lithium is a nested chemical view of identified host stocks and must not be added to those stocks. Unknown quantities remain unknown. Partial component estimates do not establish the composition of the remaining equipment mass.

Civil works, building shells, foundations, roads and unallocated offsite infrastructure are excluded. Equipment steel is not total campus steel. Initial fluid stocks are not annual consumption. No direct conversion to mine demand, national supply shares or commissioning dates is supported.

Reconciliation and acceptance

The comparison reports modeled minus historical quantities, including rounding tolerances, missing material scopes and unresolved differences. A discrepancy is a research result to explain. Do not adjust a dimension or coefficient solely to recover a reported aggregate.

Software tests check arithmetic, physical constraints, admission boundaries, preservation and deterministic reproduction. They do not establish representativeness, code compliance or external engineering validation. A completed model development can retain material uncertainties; it must expose them at the point where results are communicated.

The Anatomy app remains a separate branch. Its illustrative Run 04 figures should be promoted only after an explicit mapping to the modeled identity and revised results. This research update does not deploy or relabel the app.

Source details

Run 04 research. Equipment material estimates with documented assumptions and scope. Web prose uses current site terminology; downloadable records preserve original wording. Hashes identify the downloadable records.

Source: docs/RUN_04_METHOD.md
SHA-256: ee086608d9ed64194a707f613e6defe0674b455170c682a1ba2d03f38d8e2bde
Download research file

Optional: tell Sid you’re reading
The document is open; sharing your name/email is optional.