Skip to content

Proposal 2: load control with an energy reading #5

Open

Description

@abu-aska

Closes: interface L5 of docs/STRUCTURE.md (in #2).

The gap

Neither Weft nor Exo can shed or defer compute on request, and neither measures the energy it uses. A search of Weft's runtime, documents and papers, and of weft-loom's experiments and notes, finds no figure in watts, joules or kWh. weft-loom lists "energy per adapter write" as a measurement still to make.

Without this, the scheduler side of the contract has nothing to act on, and nothing like energy per update or tokens per joule can be computed from what the four organisations hold.

The system

Two calls on the compute side:

  • shed(fraction, seconds) returns what was actually shed;
  • energy() returns joules since the last call, read from the platform's power interface (RAPL on x86, the power metrics interface on Apple Silicon).

Both report into the load slot of the plant frame.

First slice

One Weft experiment run twice, with and without a 50% shed request, with measured joules for each. The swarm benchmark already recorded in weft-loom is a candidate.

Done when

  • the measured energy falls by a stated amount under the shed request;
  • the record is committed with its preregistration, under weft-loom's own rules;
  • a null or negative result is recorded the same way.

Rests on

Weft's plasm and telemetry, weft-loom's experiment format.

This would be the first energy measurement anywhere in the four organisations.

Filed by an AI agent (Claude Code) working for the owner of aska-physical.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions