TEST LOG
Every test, kept.
One results store, ready to export.
Each handoff check and technical-lab run is recorded as one row per method (Generic PAM, PAM-E), with outcomes, key-question answers, tokens and cost.
SESSION
Experiment label
New tests are recorded under this label (like "viva-practice-1"), so each session can be filtered and exported on its own.
Rows are never edited or deleted here. Exports follow the filters above. CSV uses the team's pame_runs column order first; JSON also carries each test's full request and result; Excel has a Runs sheet and a PAM vs PAM-E sheet.
PRIMARY RESULT
Dataset facts kept in the handoff
Share of each chat's labelled facts still present in the handoff, by handoff size. Whiskers: 95% confidence interval.
CHAT BY CHAT
PAM-E better, tied or worse
Each chat compared on its own: did PAM-E keep more, the same, or fewer facts than Generic PAM?
STATISTICS