Description
Behaviour
Two rows indexed, two identical-shaped rows appended after the build into an uncovered fragment. search finds both beta rows (the uncovered fragment is searched with the index analyzer), but fuzzy at one edit finds running only in the appended row (bare token) and misses the indexed stem run, and finds beto only in the indexed row (no expansion over the uncovered fragment).
Contract
Analyzed lexical search RFC (#792, rollout step 2): one matched set per query across index states, matches(indexed) = matches(unindexed) = matches(partially indexed); the exact scan is the baseline and a native index accelerates only after it proves the same membership.
Regression case
index_state_changes_text_matches.gqt on the evidence branch (steps 3–4 red, including the two mutation steps). The fix lands it as issue_N_index_state_changes_text_matches.gqt with the --- runner header.
Activity
azimafroozeh commentedon Oct 1, 2026