Skip to content

VALIDATION.md

Latest commit

 

History

History
36 lines (27 loc) · 4.84 KB

VALIDATION.md

File metadata and controls

36 lines (27 loc) · 4.84 KB
Raw

LoopGrid LangGraph v0.1.0 — Validation and release ledger (2026-10-12)

Verified on user's Windows system for earlier candidates

  • Config-fixed pre-RC2: 55 passed, real installed LangGraph StateGraph, InMemorySaver, interrupt / Command(resume) passed. Both local Core 0.8.1 E2Es passed with signed verification valid. These results do not count as RC3 Core validation.
  • Hardened RC2: Windows Python 3.12, LangGraph 1.2.14, langgraph-checkpoint-sqlite 3.1.1. 67 passed, langgraph_runtime_e2e.py PASS, persistent_langgraph_e2e.py PASS (two restarts, one sandbox tool call, no duplicate args in loopgrid scope).
  • RC2 Core FAIL: both real_core_e2e.py and native_core_langgraph_e2e.py raised EvidenceInvalid: Signed decision identity, authority or context changed. RC2's new signed-metadata comparison was comparing Core's redacted typed SHA-256 fields with original strings. This is an adapter comparison bug, not evidence of a Core integrity failure.

RC3 — verified locally (isolated Python 3.13.5, no installed LangGraph runtime)

  • 77/77 tests passed with PYTHONPATH=src:tests python -m pytest -q. Added ten regression cases (compared with RC2's 67) for typed redacted commitments, replay identity validation and tampering/mismatch.
  • Coverage via --cov=loopgrid_langgraph --cov-fail-under=90: 90.36%, CI gate passed.
  • The compatibility matcher was checked against a historical exported real-Core decision_created evidence record; agent/authority/model/context values and canonical-JSON SHA-256 redaction commitments all matched. This does not constitute a fresh live-Core run.
  • Python wheel built via pip wheel --no-deps --no-build-isolation in this isolated environment.
  • Fix is limited to signed metadata comparison; policy/review/tool dispatch/core event wire contract remain unchanged.

RC3 — verified on user Windows system (2026-10-12)

  • python -m pip install -e ".[test,persistence-test]": installed successfully under Python 3.12.
  • python -m pytest -q: 77 passed in 1.46s.
  • Real LangGraph StateGraph + InMemorySaver + interrupt + Command(resume) script: PASS; one synthetic tool execution, with observed checkpoint interruption and resumption.
  • SqliteSaver persistent restart/resume script: PASS; two restarts, one synthetic tool execution, no duplicate raw arguments in LoopGrid scope. Core in this test is explicitly synthetic.
  • real_core_e2e.py: PASS, against local LoopGrid Core 0.8.1-design-partner, workspace verification true, failures [], two synthetic executions, lifecycle evidence_complete, applicable coverage 100, no real money. This script does not invoke the LangGraph SDK.
  • native_core_langgraph_e2e.py: PASS, actual installed LangGraph StateGraph + local LoopGrid Core, policy, signed human review, interrupt/resume and replay protection; signed workspace verification true, failures [], two synthetic executions, lifecycle evidence_complete, applicable coverage 100, no real money.
  • The earlier RC2 signed-redaction mismatch did not recur.

Final release gates — verified on Windows and GitHub (2026-10-12)

  • Package: python -m build produced wheel and sdist; scripts/check_dist.py passed and twine check passed on both artifacts. Clean non-editable wheel installation, public-class import and pip check passed in a separate fresh local venv.
  • Expanded suite: 12 additional Core, lease and error-path regression tests increased the suite from 77 to 89/89 passed. Windows Python 3.12 statement coverage 95.78%, above the 90% minimum. No runtime Python source changed for this coverage fix.
  • Hosted GitHub CI: 9/9 OS/Python matrix jobs passed across Ubuntu/Windows/macOS on Python 3.10/3.12/3.13; the separate clean-wheel consumer-install job passed (source commit 24327c4).
  • Public release: GitHub release v0.1.0 points to commit 24327c4 and is marked Latest. The pypi GitHub environment, PyPI Pending Trusted Publisher (loopgridio/loopgrid-langgraph, release.yml, pypi) and explicit repository variable enabled publishing.
  • PyPI: The GitHub OIDC publishing job completed successfully and loopgrid-langgraph appears on the publisher's PyPI projects list. A separate new consumer installation using pip install --no-cache-dir --index-url https://pypi.org/simple/ loopgrid-langgraph==0.1.0 remains to be recorded.
  • Evidence interpretation: All local Core/E2E actions were synthetic; signed workspace verification passed. Production host authenticated reviewer, distributed lock, state privacy and real business-outcome proof remain application responsibilities.

Release decision: v0.1.0 is publicly released; do not describe it as a pending release candidate. Publication and independent public-index installation are distinct milestones. Official LangGraph ecosystem listing is not yet established.