Course Info

How the release harness works

A short course on the Tutors release harness: the theory behind it, the design of its containers, the philosophy of A/A and A/B testing, the approach it takes from scripted traffic to a verdict, and the A3 that explains every night whether main is ready to become a release.

The gap, explained

Find The gap, explained. It is a ledger: every failing difference of the run lands in exactly one row, and the rows add up to the total.

Row Read it as
a claim the release said it meant this
a PR, most likely one PR merged the night the difference first appeared, and its files touch what changed
one of several PRs the night’s PRs together; the harness does not guess between them
unexplained no PR merged in between, or more than eight could have made it (suspects are listed, never counted)

Write the ledger as one sentence, in the A3’s own form:

N differences: claims explain C; #PR most likely contributed P; …; U unexplained.

The unexplained number is the work. A large row from one PR (for example a framework upgrade adding many SBOM packages) is one cause, not hundreds of problems.