ControlPoint Manual — versioned with the code; every worked example below is a REAL corpus job and its numbers are pinned by the test suite.
5 — Read the report
What it does
The report is the deliverable: inputs with SHA-256 hashes, weighting provenance, per-setup health, blunder detection (dual standardised residuals), reciprocal checks, control comparison, coordinates with confidences, and a full residual appendix where every line carries its audit key (fieldbook:line) so any number can be traced to the booking in seconds. Worst first, everywhere.
Worked example — 2HR19's single-target fault
The report isolated FREE1→N11 as a single-target fault of 9.6 mm (orientation-chase ratio −1.474 against the theoretical −1.463 for one bad ray in that setup). Excluding that one observation moved N11 by 3.5 mm and nothing else — exactly what the diagnosis predicted.
Theory in one paragraph
A standardised residual w = v/σv compares each correction to what the geometry allows it to be; |w| > 3.29 (Baarda, α = 0.001) flags an outlier worth investigating. Redundancy numbers say how testable each observation was — a fault on a low-redundancy line hides in the coordinates instead of the residuals, which is why the report prints both.