Release 0.2.0 (deployment monitoring; the orchestrate decay stack)

[mspinola]

0.1.0 published 2026-07-23 and main has been 15 commits ahead ever since, carrying the entire half of the framework that runs after a promotion. None of it was installable from PyPI, which is the same shape as the crucible 0.4.0 problem: real work on main, a changelog that reads as if it shipped, and no tag.

It also makes today's floor fix mean something. The pin was raised to crucible>=0.6.0 so a PyPI install could not resolve a crucible without Thresholds.monitor_arl0_years — but PyPI's crucible-stack didn't contain the code that needs it. The built wheel now carries Requires-Dist: crucible>=0.6.0.

Contents

orchestrate.decay + EdgeDecayTrigger, the EdgeBaseline frozen into the DeploymentLedger beside the DriftEnvelope, trade_r / trade_dates through the trigger seam, the --edge-decay and --arl0-years flags, docs/api-stability.md, and the public-API surface pins.

Three changelog problems fixed while consolidating

Rather than published as they were:

• Two separate ### Added blocks, because three PRs each prepended one, so entries were filed under whichever heading happened to be nearest.
• A ### Changed entry still warning that the crucible>=0.5.0 floor was "not yet true" and "must not be released before crucible 0.5.0 is on PyPI". Both conditions have since been met and the floor has moved twice more. The three steps are now one entry that is true as written.
• No ## [0.1.0] heading existed at all, so the published release had no record. Added with its upload date.

__version__ said 0.1.0.dev0 against a published 0.1.0; now 0.2.0, matching pyproject.

Verification

Reproduced what the release workflow does: python -m build then twine check — both artifacts PASSED, and the wheel contains orchestrate/decay.py and orchestrate/__main__.py. 407 tests pass, ruff clean.

Tag v0.2.0 after merge.

🤖 Generated with Claude Code