Build 13 · Open
Measure whether a harness change actually helped
Instrument the-machine so a single harness change — a tool, a retry policy, a context strategy — produces a before-and-after number rather than an impression.
Open
State
Off route
Route step
0
Notes
Done when
A state a stranger could checkOne harness change has a measured effect on a task suite, with the null result reported as readily as a positive one.
Why it is harder — The only build whose subject is a system already running in production — yours.
What it reads
Just enough to unblock the next step- AI Agents (11-768) The rung material, once it opens. Not a blocker — the-machine already runs.
- The Modern Software Developer (CS146S) A1 · A2A1’s observed-versus-inferred discipline is the measurement habit this build needs; A2’s braked write tool and structured errors are the bar the-machine’s own tools get held to. Nothing past week 3.
- Modern AI Systems: Agents and System Optimizations (CS2680) A1 · A3A1’s stream-json viewer is the instrument: it gives every run its tool calls and its cost. A3’s rule (hold task success fixed, then cut cost and latency) is the before-and-after this build is asking for.
What it ships
- NotebookThe harness flagship — the earned version of the 98% claim
- BuildHarness instrumentation and its first measured change
Teach it back
If it cannot be explained plainly, it was not learned- 1Explain itExplain what you built to a friend who has never seen it. Plain words, no jargon, five sentences at most.
- 2Predicted vs realWhat you expected to happen, what actually happened, and why the two differ.
- 3Where it breaksThe question about it you still cannot answer, and the point where your explanation stops working.
- 4KeepThe one number, rule or picture worth keeping. It becomes a bottle and gets re-tested.
No notes on this build yet.
How a note lands here
A note is a Learning Log entry with the tag study-harness-instrument. It is written in Connor's words, filed as a draft, and appears here once it is published from the admin queue. One note can carry several build tags.