DR original Change Management process/configuration lab, 2026-10-04

Run from project root:
python3 content/labs/change-runtime/run.py --output /tmp/dr-change-runtime.json
Choose a new output filename. Standard library only, no network or existing service
access. The script creates a private temporary tree, original worker copies, JSON
configuration, local symlinks and one invented receipt. CPython 3.13.1 on Darwin
arm64 was recorded. Rolling Python 3.13 documentation currently labels 3.13.16;
that source label does not replace the executed interpreter version.

Thirteen groups passed twice. Require exit 0, passed=13, failed=0, failure=null,
matching scriptSha256 and workerSha256, and expected checks. Every worker loads
configuration once. v1 returns 12; selecting v2 leaves the old worker at 12 while
a new v2 worker returns 22. Incomplete JSON exits 2; a wrong field type exits 3.
These codes are application choices, not standard incident priorities. Hash/target
comparison detects scope changes; adding a newline preserves parsing but changes
byte identity. This comparison is not real identity verification or authorization.

The deliberately faulty candidate passes preflight and announces ready. Ordinary
requests yield 22; critical requests incorrectly yield 23 instead of required 22.
A synthetic receipt records the incorrect result. Selecting v1 again leaves the
already-running faulty process unchanged; starting v1 restores subsequent result
12. The prior receipt remains unchanged and needs a separate decision. Deliberately
bypassing preflight selects invalid JSON: the new process exits 2 while an existing
v1 process continues working. All five owned child processes terminate and all
owned temporary files are removed. The script does not stop any existing service.

os.replace operates within the same filesystem here. It does not atomically change
process memory, routing, several machines or receipt history. Resolving selection
once pairs code and configuration paths, but does not make their contents immutable.
No protection from concurrent deployers is implemented. No crash-durability, power
failure, throughput or production latency claim. Requests use direct stdin/stdout
pipes, not HTTP, load balancing or service-manager activation. No real service
rollout, bank data, external messages or human workshop is executed. The arbitrary
calculation is an original fixture, not a financial formula. Worker code is for
teaching and is not a production deployment library. Historical SQLite/migration
exercises remain separate. Target-runtime and human acceptance remain outstanding.
