Method, evaluation,
and the results that
went against us.
Written while the work is happening, including the parts that did not go our way. Every note has to clear the same four conditions before it goes up.
Before a note goes up.
A number exists only if our own evaluation code wrote it.
A result that went against us is reported at the same prominence as one that did not.
Where the evidence is simulated, the note says so in the first paragraph.
A claim we later found to be wrong is corrected in place, dated, with the old claim left visible.
The notes, newest first.
A metric twin of the room, from cameras alone
Multi-view RGB in, a metric 3D reconstruction out, with every point carrying an entity label and pairwise distances validated against time-of-flight ground truth.
The headline sterility number is a rule, not a model
The 0.768 that the operating-room literature quotes for sterility-breach detection is a zero-shot proximity heuristic, reported without an alarm rate. We read it at source.
Phase detection that has to commit
A causal, forward-only decode at one inference per operating-room second. A model that cannot revise its own history is harder to build and is the only kind that is useful in the room.