Skip to content

v0.1.0rc3 — stale-context protection baseline

Pre-release
Pre-release

Choose a tag to compare

View all tags
@kyle641320 kyle641320 released this 05 Sep 17:26
· 32 commits to master since this release
v0.1.0rc3
be85e04

Included

  • Source-aware stale-context protection for AI coding agents
  • Offline stale-gating demo
  • Guava M10 scoped case study
  • Open-source minimum checklist and external feedback issue templates

Evidence scope

This release documents validated source-analysis mechanics and scoped stale-context evidence. It does not claim general agent productivity, speed, token savings, or production readiness.

Verification

  • Full test suite: 618 passed, 5 skipped
  • Offline stale-gate demo: PASS

Install and try

Python 3.10+; this is a prerelease pinned to rc3:

python -m pip install --pre "true-memory-fragments==0.1.0rc3"
# Optional Java parsing support:
python -m pip install --pre "true-memory-fragments[java]==0.1.0rc3"

PyPI rc3 · README install guide

For the 30-second deterministic demo, run a source checkout (Git required):

git clone --branch v0.1.0rc3 --depth 1 https://github.com/kyle641320/true-memory-fragments.git
cd true-memory-fragments
python3 scripts/demo_stale_gate.py

Expected markers: STALE CLAIM BLOCKED: PASS, SOURCE FALLBACK PROVIDED: PASS, REREAD REQUIRED: PASS.

The demo imports the checkout's source; it is not a standalone PyPI wheel verification. Installing the package alone does not download repository demo scripts. The commands above check out the rc3 tag; the linked README and evidence pages track current documentation on master.

Results and feedback

In this experiment, semantically evaluable runs passed 34/34 for TMF_STALE_GATED and 20/20 for SOURCE_ONLY. This is an observed semantic-boundary result, not a 50/50 task-success or statistical equivalence claim.

Dated rc3 re-verification (2026-09-06)

At rc3 commit be85e04670ece701838f5206beb417e2efbd6c64, both ordinary and -Werror unittest runs executed 617 tests: 612 passed, 5 skipped, 0 failures per run. This is source-checkout verification, not standalone wheel verification. Historical audit counts above are preserved as historical records, not current totals. Java qualification counts use a separate unit and were not rerun in this verification.

Version-pinned results, commands, environment and raw logs. The separate master snapshot had an intermittent test failure, explicitly retained in that report; its results do not replace rc3 evidence.

Assets 5
Loading