Keyboard shortcuts

Press ← or → to navigate between chapters

Press S or / to search in the book

Press ? to show this help

Press Esc to hide this help

0.10 Beta Scope and Known Limits

PowerIO 0.10 is the public beta of the 1.0 API. API corrections may land before 1.0.0 as downstream integrations exercise the new design. The architecture documents under arch-v1/ describe the 1.0 target; 0.10 implements it and asks for real world use before the permanent freeze.

In this beta

  • One parse for every supported source, returning the typed module; twenty built in value kinds across networks, series, scenario sets, instances, and solutions.
  • Byte exact same format writing, diagnosed cross format conversion, and the explicit multiconductor to balanced lowering.
  • Structured diagnostics with stable codes and native record access in every language; the wire form carries span fields end to end, though 0.10 parsers do not yet emit them.
  • Balanced matrices (Y bus, FDPF B’ and B’’, LACPF, incidence, DC operators, AC power flow Jacobians, PTDF and LODF), all carrying element mappings; direct multiconductor admittance assembles in Rust only this release (see Known limits).
  • The stored .pio.json document, version 1, with the one way upgrade from released 0.9 documents.
  • C ABI 6, the Python package, PowerIO.jl, the powerio command line tool, and the MCP server over one set of names.

Known limits

  • Format profiles are bounded. PyPSA support is the CSV electrical profile: multi carrier components, investment periods, and stochastic data are retained and reported, never typed. Egret support is the scalar network profile with time series; unit commitment fields stay outside it. OpenDSS support is the static circuit; load shapes and solve instructions are retained and reported. PyPSA NetCDF does not parse.
  • DOE GO Challenge 3 and DeepMind OPFData are parse only; PowerWorld PWB is a parse only binary.
  • Solving is out of scope permanently: instances feed external solvers.
  • Balanced to multiconductor construction, load linearized multiconductor admittance from an operating point, and a general multi period planning instance wait for after 1.0.
  • Dynamic simulation data has no representation yet; QSTS interchange beyond complete sampled operating point series waits for named instance and solution types.
  • There is no one call facade convert(source, format, destination) yet: conversion is powerio_tx::convert_file/convert_str for the balanced family plus the per family write paths, and the CLI’s powerio convert. The Destination based facade form the architecture record names arrives with 1.0.
  • Classifying an undeclared JSON source is one cheap typed pass, except that a document nesting its payload under a network, grid, solution, or metadata marker key (GO Challenge 3, Surge) still materializes that subtree once during classification; cost is linear in the nested payload and transient.
  • The parser allocation rules in the architecture record (arch-v1/V1_ARCHITECTURE.md) are implemented for MATPOWER, PSS/E, and PowerWorld AUX; PyPSA CSV, PSLF, and OpenDSS still tokenize through owned strings, and several JSON readers (Egret, GO Challenge 3, DeepMind OPFData, pandapower) decode through a serde_json::Value tree. Scheduled work, stated here so the architecture record is not read as already shipped.
  • Multiconductor admittance assembly (powerio_matrix::build_multiconductor_admittance) is Rust only in 0.10: no C entry point, and so no Python or Julia binding yet.
  • The sparse direct DC sensitivity factorization trades memory for speed against the previous conjugate gradient path: dense band peak memory is up about 3x at 2000 to 3000 buses for an 8 to 10x wall time win, measured against the committed allocation baseline in evals/allocation.

Version boundaries

One package version covers the Rust crates, the Python package, and PowerIO.jl. The independently versioned compatibility boundaries:

BoundaryValue at 0.10Checked whereMoves when
package version0.10.0manifests, powerio.versions(), build_infoevery release
C ABI6pio_abi_version handshake at loadan existing C signature or documented behavior changes
.pio.json schema1the stored document headera document version 1 cannot represent is needed
matrix Arrow tablesappend only, no separate numberthe Arrow catalog report, stamped with the package versionan existing table’s identity or column order would change (a removed table’s id is burned, never reused)
MCP tool surfaceno separate number, mirrors the package versionthe schema and powerio_version keys on every tool responsewith the package release

These answer different questions and never race each other: 0.10.0 is what you install, ABI 6 is what a compiled consumer must match, schema 1 is what stored documents declare, the Arrow catalog is the one report a table consumer reads before addressing columns, and a tool response’s own schema/powerio_version keys are what an MCP client reads before trusting its shape.