iomgaa
4b6d1d8a50
feat(harness): correctness 三口径 + gate 块按 unit 跑
进化引擎与 gate e-process 从 question_id 口径迁至 unit_id 口径,AR pair
双向 AND 折叠为单元、不被 P/Q 单题计分污染;逐题 predictions 仅作溯源。
- question_units: 新增 unit_correctness_view(units, per_q)->dict[unit_id,bool]
作为逐题→单元折叠的唯一入口(复用 unit_correctness)。
- core/evolution/validate: pair_block/compute_accuracy 参数改 unit_ids、
分母按单元数(键即 unit_id)。
- app/harness/validate(gate 实际执行路径):阶梯题序聚合为单元并保持信息
阶梯序(_ladder_units),gate 块按单元切分(AR pair 整锁不跨块拆);
baseline_cache 键含 unit_id、存单元级对错;候选逐题读回后折叠成单元视图;
n_used/W/L/四象限/准确率均按单元计;证据行按 unit 口径,candidate_correctness
独立保留逐题对错供 runner 二轨合并。
- runner: probation 结算按 unit 折叠计 W/L(_probation_unit_flips);quadrant
四象限 id 承载 unit_id。
核心算法保真 #5(信息阶梯 e-process):本次仅迁移 correctness 口径,不改冷启动
2:1 / gamma-EMA / 反泄漏算法本身(gate_ladder 迁移见 Task 8)。
2026-07-15 07:31:03 -04:00
..
2026-07-06 11:37:59 -04:00
2026-07-14 16:00:14 -04:00
2026-07-14 15:47:21 -04:00
2026-07-14 15:55:36 -04:00
2026-07-14 16:33:09 -04:00
2026-07-14 17:07:55 -04:00
2026-07-14 16:16:38 -04:00
2026-07-14 15:35:11 -04:00
2026-07-11 08:48:04 -04:00
2026-07-06 22:36:31 -04:00
2026-07-06 22:32:59 -04:00
2026-07-14 14:31:30 -04:00
2026-07-15 06:46:28 -04:00
2026-07-06 22:39:57 -04:00
2026-07-11 11:56:36 -04:00
2026-07-12 22:34:47 -04:00
2026-07-12 22:33:49 -04:00
2026-07-15 07:31:03 -04:00
2026-07-09 12:20:32 -04:00
2026-07-07 10:04:48 -04:00
2026-07-14 14:47:21 -04:00
2026-07-07 01:45:09 -04:00
2026-07-07 09:34:04 -04:00
2026-07-07 10:26:58 -04:00
2026-07-09 05:36:07 -04:00
2026-07-11 23:14:22 -04:00
2026-07-15 07:31:03 -04:00
2026-07-14 14:33:22 -04:00
2026-07-07 09:38:21 -04:00
2026-07-14 05:50:47 -04:00
2026-07-11 08:19:45 -04:00
2026-07-14 13:51:51 -04:00
2026-07-14 05:55:08 -04:00
2026-07-09 00:12:38 -04:00
2026-07-15 06:46:28 -04:00
2026-07-07 13:13:37 -04:00
2026-07-12 22:46:28 -04:00
2026-07-07 12:55:12 -04:00
2026-07-15 06:57:48 -04:00
2026-07-12 22:45:18 -04:00
2026-07-07 13:01:04 -04:00
2026-07-07 12:55:12 -04:00
2026-07-14 00:58:54 -04:00
2026-07-15 06:46:28 -04:00
2026-07-07 12:36:36 -04:00
2026-07-07 13:20:43 -04:00
2026-07-07 12:41:59 -04:00
2026-07-15 07:09:54 -04:00
2026-07-15 06:30:14 -04:00
2026-07-07 06:13:25 -04:00
2026-07-07 09:44:05 -04:00
2026-07-14 21:08:00 -04:00
2026-07-14 15:39:57 -04:00
2026-07-14 13:40:46 -04:00
2026-07-15 06:09:14 -04:00
2026-07-09 12:20:16 -04:00
2026-07-11 23:23:37 -04:00
2026-07-12 00:04:42 -04:00
2026-07-07 04:45:19 -04:00
2026-07-15 05:45:34 -04:00
2026-07-15 05:52:30 -04:00
2026-07-06 22:39:42 -04:00
2026-07-09 00:54:27 -04:00
2026-07-09 00:23:04 -04:00
2026-07-09 07:53:59 -04:00
2026-07-07 02:37:50 -04:00
2026-07-14 13:59:29 -04:00
2026-07-12 00:14:12 -04:00
2026-07-14 05:43:28 -04:00
2026-07-07 06:11:13 -04:00
2026-07-07 05:46:25 -04:00
2026-07-07 05:53:33 -04:00
2026-07-11 09:02:03 -04:00
2026-07-07 05:52:54 -04:00
2026-07-06 11:37:59 -04:00
2026-07-14 06:49:03 -04:00
2026-07-14 13:53:17 -04:00
2026-07-14 06:51:50 -04:00
2026-07-06 22:43:01 -04:00
2026-07-14 15:30:18 -04:00
2026-07-09 11:57:41 -04:00
2026-07-09 07:40:27 -04:00
2026-07-09 00:11:54 -04:00
2026-07-11 09:02:03 -04:00
2026-07-09 11:57:41 -04:00
2026-07-11 23:05:38 -04:00
2026-07-07 09:47:45 -04:00
2026-07-09 11:57:41 -04:00
2026-07-11 11:14:45 -04:00
2026-07-07 01:49:55 -04:00