ENGINEERING PROOF 020 / CROSS-RUNTIME COMPARISON

Two languages.
Same 13 decisions.

The same proposed-action cases go through separate JavaScript and Python code paths. Inspect each plan, dispatch receipt, callback count, and executed proposal—not just a green checkmark.

Both implementations are maintained by this project. Agreement on these published fixtures is interoperability evidence for a narrow JSON subset, not an outside witness, security certification, or general cross-language canonicalization.

Recorded project-maintained comparison. A mismatch in a future run will make this claim fail in CI. An adversarial numeric-key case exposed a disagreement outside these 13 fixtures; inspect the repair and broader regression.

Loading recorded comparison…

Choose a case. The comparison below is populated from the checked-in machine report, regenerated and checked in CI.

13 / PUBLISHED CASES

Loading…
SELECTED CASE / —

Loading…

CHECKING

Reading both recorded outputs.

01 / JAVASCRIPT
—
Policy decision—
Receipt reason—
Tool callbacks—
PROPOSAL SHA-256
—
02 / PYTHON
—
Policy decision—
Receipt reason—
Tool callbacks—
PROPOSAL SHA-256
—
Waiting for report.
Inspect both exact JSON outputs
—

Run the black-box challenge

From the repository root, invoke the Python adapter as an executable process against all 13 public cases:

node scripts/run-action-boundary-challenge.mjs -- python3 pilots/action-boundary/challenge-python-adapter.py

The adapter imports no JavaScript boundary code and runs only simulated callbacks.

Check the visual report

Regenerate and compare the actual outputs of both implementations to this page’s checked-in data:

node scripts/build-action-cross-runtime-report.mjs --check

Supported wire subset: JSON integers in JavaScript’s safe range and BMP object keys. Other values fail closed in Python.