Repository navigation
scoring: define and enforce session-measurement reuse in maps #98
Description
The importer validates map entries independently, so distinct survey IDs can reference one session measurement. This can count the same observation more than once in reported coverage. Define whether any reuse is legitimate, then reject unsupported reuse with a specific error and regression tests. Keep genuinely shared evidence represented by one survey measurement where appropriate.
Follow-up from PR #4 discussion_r4114069346, deferred under the approved non-safety convergence rule. Maps remain operator-authored research inputs, not production clearance evidence.
Activity
SamGu-NRX commentedon Sep 27, 2026
The new Greptile thread in #4 supplies a concrete false-result reproduction under the current 08:30Z standard. Using the committed synthetic fixture, copying c1-gas onto c2-gas in the map changes c2/gas from UNSURE to PASS without a second accepted measurement. This is now escalated as a blocker in discussion_r4115046414 rather than treated as coverage-only duplication. No repair made.
Require distinct camera frames before casting 2D detections (#98)