Selectivity Reduces a Generated File From 20000 Lines to Two

The line-reduction demonstration starts with a deliberately oversized text file whose answer is known in advance. It has 20000 lines: alpha appears on line 702, omega on line 17891, and every other line contains an irrelevant row label. The predicate succeeds only when both trigger lines are present. This makes a useful check of the reducer, because a result that loses either trigger is wrong and a result that keeps unnecessary lines is visibly unfinished.

In one recorded run of Selectivity 0.1.1 in a Linux amd64 container, Selectivity reduces the 408866-byte file to 12 bytes containing alpha and omega on their own lines. The engine performs 80 predicate executions with repeats set to 1 for this demonstration. Baseline and final verification are included in that count. The saved summary reports completion with no unresolved checks and local deletion minimality at line granularity.

A generated 408866-byte file becomes a 12-byte input containing its two required trigger lines

The command uses an explicit shell predicate that requires both exact line matches:

./selectivity --input large.txt --out lines --repeats 1 \
  -- sh -c 'grep -qx alpha "$1" && grep -qx omega "$1"' check '{input}'

Selectivity does not know which two lines matter. It discovers that through candidate deletions and the predicate’s answers. The final content is compared byte for byte with the expected two-line file. The original remains available as a snapshot, and the JSONL audit records the candidate checks that led to the result. The fixture generator is included in scripts/demo.sh, so the example can be recreated rather than trusted as a screenshot.

This is a synthetic reduction example, and a fairly cooperative one. It establishes that this input can be reduced to its known answer under the supplied predicate. It does not measure a parser crash, prove that all reductions need 80 checks or demonstrate a globally smallest result across arbitrary edits. A real CSV or configuration file often needs validity checks in its predicate. The reference explains that boundary; the recorded evidence preserves the run details.