Observation — Navigation Drills (query log)
The queries are the deliverable. Log failures.
Scorecard
| Drill | Predicted subsystem | Correct? | Queries to converge | Time | Technique that worked |
|---|---|---|---|---|---|
| 1 specificity | |||||
| 2 xref | |||||
| 3 naming | |||||
| 4 generated | |||||
| 5 history | |||||
| 6 tests | |||||
| 7 invariants | |||||
| 8 layering |
Prediction accuracy: __ / 8
Drill 1 — specificity
Queries tried:
Answer: Where my model was wrong:
Drill 2 — xref
Queries tried:
Caller count: Callers from different subsystems: What the caller count implies about changing the signature:
Drill 3 — naming grammar
Queries tried:
cc main-thread owner: impl-thread counterpart:
*Client found: inverted dependency direction & why:
*Delegate found: inverted dependency direction & why:
Drill 4 — generated code
.idl declaration:
Blink method:
Name-mangling rule:
CSS property chosen: inherited? animatable? custom parsing?
Feature flag not enabled by default: where "enabled by default" is recorded:
Drill 5 — history
CL/spec justification for <input type=hidden>:
Oldest surviving line found:
Most recently changed line found:
Where the risk lives in this subsystem:
Drill 6 — tests
Unit test: Web test: WPT: Where expected-failures are recorded: Candidate failing WPT for Phase 6:
Drill 7 — invariants
| DCHECK | Invariant in one sentence |
|---|---|
CHECK found: why unrecoverable rather than a bug:
Drill 8 — layering
Unexpected include rule: Documented exception + the compromise: Team boundary inferred from OWNERS:
Cross-check — the ladder
Source-only: time __ confidence __ answer __ Tracing-only: time __ confidence __ answer __
Why is tracing rung 3 if it was faster here?