twentyOne2x
9a77615e06
Reassert Leo read-only tools after runner init
2026-07-16 08:49:51 +02:00
twentyOne2x
226b4f7323
Harden Leo benchmark with structured KB reads
2026-07-16 08:41:12 +02:00
twentyOne2x
b59bee8844
Fix Leo restart truth and read command reliability
2026-07-16 07:14:31 +02:00
twentyOne2x
021aaf6f09
Harden blind scoring for natural Leo replies
2026-07-16 06:13:53 +02:00
twentyOne2x
813d4cd91d
Harden blinded Leo reasoning repair gate
2026-07-16 05:19:50 +02:00
twentyOne2x
27fbfa49a9
Accept receipt-bound natural answers in the Leo OOS scorer ( #172 )
...
* Make OOS scoring accept receipt-bound natural answers
* Test short-ID causal attribution in OOS ablation
* Close OOS provenance and schema qualifier gaps
2026-07-15 23:44:46 +02:00
twentyOne2x
8a4325f42a
Harden OOS benchmark provenance and leakage checks
2026-07-15 12:06:35 +02:00
twentyOne2x
93eddbcd4f
Model conditional no-DB receipt gaps
2026-07-15 06:23:15 +02:00
twentyOne2x
8506dd9359
Bind Leo responses without masking semantic failures
2026-07-15 05:54:45 +02:00
twentyOne2x
da65e05316
Ground Leo reasoning in bounded live retrieval
2026-07-15 05:30:10 +02:00
twentyOne2x
ae117097af
Add blinded live Leo reasoning benchmark
2026-07-15 04:02:41 +02:00
