Changelog

All notable changes to equilibria are documented in this file.

The format follows Keep a Changelog and the project adheres to Semantic Versioning. Entries are curated milestones (user-visible capability), not an exhaustive commit log.

[Unreleased]

Fixed

  • GTAP7 pure 5x5 ifSUB=1 MCP shock parity closed: 92.35% → 100% (MCP gate 18/18 rows green, floor restored 92 → 99). The “multi-root pet/pe export block” was a spurious complementarity corner FABRICATED by Python: GAMS declares every quantity as a free variable (its MCP rows must hold with F=0 exactly), while Python’s NonNegativeReals box let PATH park xw=xet=0 and read the violated pet==ps row as bound slack. xw/xet are now free (Reals); the June 15x10 “autarky corner / crash jumps basins” episode was the same class. PATH options now default per mode to the options each reference family was actually solved with (pure bundles carry no path.opt → PATH defaults; altertax bundles → the tight inlined set), which also keeps 15x10 pure ifSUB=0 at 100%.

Added

  • F-docs: restructured documentation site (six sections), architecture overview, templates overview, artifact-style parity matrices (NLP vs NLP / MCP vs MCP for GTAP and PEP), expanded API reference, this changelog, and a docs-build CI job.

Fixed

  • GTAP7 15x10 altertax shock parity closed: 94.5/95.8% → 99.5/99.5% @1% (gate floors 94/93 → 99/99), no regression on any other dataset. Three stacked causes: (1) the June Armington-shares fix was orphaned on an unmerged branch — recovered by cherry-pick; (2) the reference GDX violated its own xfeq (negative factor demands) — regenerated one-shot on NEOS with tight PATH options (convergence_tolerance 1e-10, inlined path.opt); (3) under altertax-CD the GAMS pxeq/pvaeq/pndeq rows are tautologies (proved under-determined GAMS-vs-GAMS: three runs, three parkings), while Python added its own CD-dual pxeq determinant — the full degenerate triple (pva/pnd/px) is now pinned at the reference’s shock slice, replicating gtap.holdfixed=1.

  • Removed the unfaithful 1e-8 lower-bound floor on quantity variables (GAMS bounds prices only): microscopic cells below the floor made the GAMS point infeasible in Python and blew up their paired CES prices.

  • Multi-period gate solves now default to the same PATH options the GAMS reference bundles solve with (crash_method none, nms_searchtype line, convergence_tolerance 1e-10; PATH_CAPI_OPTIONS overrides win). With default options PATH’s crash phase jumps basins even from an exact solution — this closed the long-open 15x10 pure “wrong-branch” debt (shock 89.8 → 100.0% @1%).

  • ytax[ft]/[fs] variable inits aligned with the live eq_ytax (Σftrv / −Σfbep; fs was hardcoded 0).

[0.5.1] — 2026-07-18

Added

  • PEP-1-1 v2.1 ported to Pyomo (equilibria.templates.pep_pyomo): NLP (IPOPT) and MCP (PATH) forms at 100% cell parity vs the GAMS CNS and GAMS-native MCP references, base + SIM1 export-tax shock (PR #25).

  • PEP parity coverage matrix (scripts/pep/pep_coverage_matrix.py) with its generated docs page and CI sync gate.

  • GTAP7 NLP-vs-NLP and MCP-vs-MCP per-stage fidelity gates (test_gtap7_nlp_parity.py, test_gtap7_mcp_parity.py) with NEOS-regenerated references (PR #24).

  • GTAP altertax multi-period pipeline (base → check → shock) with the CI .nl parity gate extended to the check phase (PR #20).

  • GTAP7 parity coverage matrix as single source of truth — declarative ROWS driving the pytest gates and the generated docs page (PR #21).

  • GTAPAgg datasets registered (gtap7_3x3 … gtap7_15x10, incl. the 10r×15c consolidated GDX).

  • equilibria-1.0 roadmap reconciled with real state (PR #27).

Fixed

  • Closed the GTAP7 shock-parity gap (PR #19) and reached 9x10 full NEOS parity (sluggish factors + NEOS compile fixes).

  • Reverted the unfaithful WCO/RCO→1.0 normalization in the PEP calibration; closed xmodel phase-2/3 (PR #26).

[0.5.0] — 2026-05-20

Added

  • RunGTAP welfare-parity engineering: shadow demand integrator + babel HAR writer wired into the GTAP welfare pipeline (PR #10).

[0.4.0] — 2026-05-20

Added

  • Clean-room HAR writer (babel.har): HarWriter builder with L3/L5/L7 record validation (PR #11).

  • GTAP welfare decomposition + per-OS benchmarks page (PR #6).

  • CGEBox altertax + welfare-decomposition port plans (PR #7).

[0.3.0] — 2026-05-12

Added

  • GTAP Standard 7 template at 100% parity (base + shock) vs the GAMS NEOS references for the 9x10 and NUS333 datasets.

  • Native pure-Python HAR reader in babel.har (drops the harpy3 dependency); bundled 9x10/NUS333 HAR datasets behind load_bundled.

  • Public shock API: apply_shock parent + apply_tariff_shock.

  • Sphinx + MyST + sphinx-gallery documentation site for Read the Docs, with MIP→SAM, PEP and GTAP quickstarts and a benchmarks page rendered from committed parity CSVs (dual NEOS/local reference + wall-time).

  • ytax(r,gy) emitted with the 10 canonical GAMS tax streams (PR #3); postsim pdp/pmp recalc for alpha=0 cells.

  • MIP→SAM pipeline closure in sam_tools (balanced simple_mip without xfail).

[0.2.0] — 2026-03-08

Added

  • simulations runtime contract: mapping adapters, ieem/gtap/icio model adapters, multi-model wrappers, and CLI parity-runner coverage.