DR original incident-analysis workshop, 2026-10-04

Run from the project directory:
python3 content/labs/incident-analysis/run.py --output /tmp/dr-incident-analysis.json
Use a new output filename to preserve previous evidence. Python standard library
only; no networking, production data or platform database access. The SQLite
connection is in memory and closed before output. Recorded CPython 3.13.1 and
SQLite 3.51.2 on Darwin arm64; actual versions appear in each report. Rolling
SQLite documentation and older conceptual AWS compatibility guidance are
versioned separately from these observations.

Thirteen groups passed twice. Require exit 0, passed=13, failed=0, matching
scriptSha256 and expected checks. The report retains exact SQL, parameters and
returned synthetic rows. Expected: 49/1000 errors overall; east 40/100 and west
9/900; weighted 4.9%, unweighted mean 20.5%; successful east p95 120 ms and all
attempts p95 6000 ms using nearest rank in Python (not a SQLite percentile
extension); absent east traffic produces n=0 and NULL, not observed health;
region/version cells are confounded; event B belongs to the first event window
but is only known later; adjacent half-open windows do not duplicate D at 300;
tag join returns 1100 rows and 89 error rows for 1000 requests and 49 failures.

The fictional gate rejects missing archive checks, sample n=3, one error in 50,
and observation age 75 seconds. It allows a next stage for n=80, zero errors,
age=20 seconds and three passing probes. Values 50 and 60 seconds are authored
teaching choices, not an SRE standard, SLA, statistical proof or universal
production policy. allowNextStage does not equal realBusinessAcceptance.

These are actual SQLite queries over synthetic records and deterministic rule
checks. No incident, HTTP traffic, application deployment, schema migration,
rollback, restore, causal experiment or human coordination is executed here.
The historical incident-runtime HTTP experiment has separate artifacts.
A guide can be performed by people, but no human performance or independent
specialist approval has been recorded. Team practice, service-representative
acceptance and independent review remain outstanding.
