On this page

mathlib-fp 1.9.8 qualification

Qualification statement

Version 1.9.8 is qualified only when the exact candidate source archive passes the existing target-tier gates plus the representative workflow qualification. This document distinguishes completed local evidence from mandatory CI evidence; it does not infer Linux or Win32 results from a Windows x86-64 run.

Workflow completion gate

Gate Evidence Status
At least three workflows Sensor, modelling, and probability/finance programs Passed locally, 2026-08-16
Multi-domain and realistic 4 / 3 / 4 domains per workflow; load, validate, transform, fit/solve, diagnostic, export Passed locally
Diagnostic paths Non-finite rejection, iteration-limit root, invalid-fit exception, invalid-probability exception Passed locally
Deterministic bounded output Two runs byte-identical; numerical bounds verified Passed locally
No private API / network / runtime Public APIs only; local fixture and seeded data; no new dependency Passed
Clean-archive journey Checker runs from an isolated work directory with copied fixtures Passed locally

Local environment

Item Value
Date 2026-08-16
OS / CPU Windows x86-64
FPC 3.2.2
Python 3.13.5
Runtime dependencies added None

Full local release qualification

Local workflow qualification passed for all three workflows; the retained result is build-temp/workflow-qualification/results.json. The complete release gate (python tools/qualify_release.py --release 1.9.8 --compiler fpc --lazbuild lazbuild) covers compiler identity, normal/optimized/checked-heap test modes, all examples and output contracts, documentation tests/fragments/ site/offline archive, API decisions, numerical mutation/evidence, migration rehearsal, workflow qualification, portability, Lazarus package, and benchmark/performance evidence.

Required exact-candidate evidence

Before tagging, the exact commit must still pass in CI:

  • Linux x86-64 and Windows x86-64 checksummed, network-isolated clean-archive qualification, including workflow results as retained artifacts;
  • Windows i386 secondary tests, native portability probe, source/package audit, and Lazarus package build;
  • normal, optimized, checked/heap test modes, every top-level example and output contract, documentation/site/offline archive checks, numerical evidence and mutation gates, performance evidence, portability evidence, migration rehearsal, workflow qualification, package build, and final API snapshot check.

Candidate workflow results are intentionally not pre-recorded as passed in the repository. They belong to the exact candidate commit's CI artifacts.

Known limits

  • Local evidence covers Windows x86-64 only. Supported target claims remain bounded by SUPPORT.md and exact candidate CI.
  • The sensor workflow reads a bundled CSV relative to the working directory; the documented run instruction is to execute from the repository or extracted archive root, and the checker enforces an isolated copy of the fixture.
  • Seeded workflows are reproducible within a single build and platform; the checker verifies same-platform repeatability rather than cross-platform bitwise equality of floating-point results.