iteration	commit	status	primary_metric_name	primary_metric_value	description
0	44c83aaae41f	baseline	weighted_median_casscf_kernel_seconds	102.443765592	baseline
1	fdc4b31c84b8	rejected	weighted_median_casscf_kernel_seconds	102.853240562	Vectorize incore Newton-CASSCF Hessian-vector ERI contractions while preserving the out-of-core fallback loop [Correctness passed, but the primary metric regressed by about 0.40% and missed the required 2% improvement.]
2	77d0fdfe785d	rejected	weighted_median_casscf_kernel_seconds	102.771270716	Optimize Newton-CASSCF Hessian-vector H_co JK contraction by exploiting sparse core-response rows [Correctness passed, but the candidate regressed the minimized primary metric by about 0.32% versus the incumbent and missed the required 2% improvement.]
3	f6b960fe047c	rejected	weighted_median_casscf_kernel_seconds	102.156812625	Skip redundant final Newton-CASSCF AO2MO/CASCI after a no-step keyframe-gradient check [Correctness passed, but the ~0.28% speedup missed the required 2% improvement, so the baseline remains incumbent.]
4	0e0c3c0191da	correctness_failure	weighted_median_casscf_kernel_seconds	95.7243468	Honor the already scaled Newton-CASSCF inner AH gradient stop target in update_orb_ci instead of applying an extra 0.3 factor, reducing over-tight AH work while preserving public convergence settings [Hard correctness failure: e_cas drift 9.64246e-06 exceeded 5e-08 despite the speedup.; case test-newton-benzene-ccpvtz-cas66-pi-sort final active-space CASSCF energy abs_delta exceeds threshold (9.64246e-06 > 5e-08)]
5	c08bd5458630	rejected	weighted_median_casscf_kernel_seconds	102.439718451	Split Newton-CASSCF AH JK response into independent dm3/dm4 contractions and skip redundant terminal AO2MO/CASCI when no microstep is taken at a strict keyframe-gradient stop [Correctness passed, but the ~0.004% speedup missed the required 2% improvement.]
6	a3ed2ebcaf36	rejected	weighted_median_casscf_kernel_seconds	102.162934291	Reassociate Newton-CASSCF AH Hessian-vector density products over core/active support and skip redundant strict no-step terminal AO2MO/CASCI [Correctness passed, but the ~0.274% primary-metric speedup missed the required 2% improvement.]
7	2eb53381cda9	accepted	weighted_median_casscf_kernel_seconds	48.509839418	Use lazy density-fitted JK for Newton-CASSCF AH Hessian-vector response contractions while keeping exact AO2MO/CASCI energies and strict convergence checks unchanged [Correctness passed and the primary metric improved by ~52.65% without forbidden cache or final-answer reuse.]
8	7193e2b76d2b	accepted	weighted_median_casscf_kernel_seconds	41.713937723	Factorize large Newton-CASSCF AH density-fitted JK response builds over low-rank dm3/dm4 factors, with dense DF-JK retained for small AO/rank cases and exact AO2MO/CASCI unchanged [Correctness passed, no hard reject or forbidden paths, and the primary metric improved ~14.01% versus the incumbent without persistent-cache or final-answer reuse.]
9	db1c93a03725	rejected	weighted_median_casscf_kernel_seconds	41.950088099	Batch Newton-CASSCF low-rank AH DF-JK factor transforms across response densities to reduce per-block BLAS setup while preserving exact AO2MO/CASCI energies [Correctness passed, but the minimized primary metric regressed by ~0.57% versus the incumbent and missed the required 2% improvement.]
10	7ed98d8c2d85	rejected	weighted_median_casscf_kernel_seconds	41.15567986	Cache Newton-CASSCF low-rank AH density-fitted response intermediates by reusing unpacked AO DF blocks and macro-constant A-factor transforms, with memory-guarded fallback to the incumbent packed-loop path [Correctness passed, but the 1.34% speedup missed the required 2% improvement versus the incumbent.]
11	334a40a0e17c	accepted	weighted_median_casscf_kernel_seconds	40.777599936	Materialize exact Newton-CASSCF outcore ppaa/papa AO2MO datasets into guarded in-memory NumPy arrays to reduce repeated AH HDF5 row access while preserving exact CASCI tensors and convergence behavior [Correctness passed and the primary metric improved ~2.24% versus incumbent without persistent-cache or final-answer reuse.]
12	a60cd89bb8c2	rejected	weighted_median_casscf_kernel_seconds	40.225887021	Vectorize exact Newton-CASSCF AH H_co contractions over materialized ppaa/papa tensors while preserving the existing outcore fallback and exact AO2MO/CASCI energy path [Correctness passed, but the ~1.35% improvement versus incumbent missed the required 2% threshold.]
13	911063b081d2	accepted	weighted_median_casscf_kernel_seconds	39.745700547	Cache per-AO2MO low-rank DF A-side transforms and use cached materialized ppaa/papa slices to vectorize Newton-CASSCF AH H_co contractions while preserving exact AO2MO/CASCI energies and tolerances [Correctness passed and the 39.745700547s primary metric is ~2.53% faster than the incumbent without persistent-cache or final-answer reuse.]
14	188c01d21e99	rejected	weighted_median_casscf_kernel_seconds	39.483604464	Guard exact Newton-CASSCF outcore AO2MO to build ppaa/papa directly as same-run in-memory arrays, avoiding HDF5 write/read materialization while preserving fallback, exact CASCI energies, and tolerances. [Correctness passed, but the ~0.66% incumbent speedup missed the required 2% threshold.]
15	e06fb6d6e54d	rejected	weighted_median_casscf_kernel_seconds	39.640415738	Incremental exact direct-JK updates for Newton-CASSCF keyframe gradients plus BLAS vectorization of materialized exact ppaa/papa contractions, preserving exact AO2MO/CASCI energies and convergence tolerances [Correctness passed, but the ~0.265% incumbent speedup missed the required 2% improvement threshold.]
16	dd9e49defdfb	rejected	weighted_median_casscf_kernel_seconds	39.890369324	Guarded vectorized Newton-CASSCF gen_g_hop setup over same-run materialized exact ppaa/papa tensors, preserving fallback, exact AO2MO/CASCI energies, and convergence tolerances [Correctness passed, but the primary metric regressed by ~0.36% versus the incumbent and missed the required 2% improvement.]
17	ccb1454453cd	rejected	weighted_median_casscf_kernel_seconds	40.264011145	Skip redundant terminal Newton-CASSCF AO2MO/CASCI when no microstep is taken at strict gradient convergence, and trim materialized-path AH Hessian-vector housekeeping without changing tolerances or exact energy construction [Correctness passed, but the primary metric regressed ~1.30% versus incumbent and missed the 2% improvement target.]
18	3d3efe61c91f	rejected	weighted_median_casscf_kernel_seconds	39.268436439	Cache exact per-AO2MO hcore MO transforms on Newton-CASSCF ERIS objects and reuse them in gen_g_hop plus fast CASCI setup without changing AO2MO/CASCI energies or tolerances [Correctness passed, but the 1.20% speedup missed the required 2% improvement threshold.]
19	f4c984052a2a	rejected	weighted_median_casscf_kernel_seconds	39.593842124	Same-state Newton-CASSCF reuse: cache exact per-ERIS hcore MO transforms and skip the redundant AO2MO/CASCI refresh after a strict zero-microstep keyframe-gradient convergence check. [Correctness passed, but the ~0.382% incumbent speedup missed the required 2% improvement threshold.]
20	0e6acc89cdae	rejected	weighted_median_casscf_kernel_seconds	46.253447011	Project large-AO low-rank Newton-CASSCF AH DF exchange directly into the required MO K blocks instead of forming full AO K matrices, preserving exact AO2MO/CASCI energies and the dense small-AO fallback [Correctness passed, but the 46.253s primary metric regressed ~16.37% versus the 39.746s incumbent and missed the 2% improvement target.]
21	b7da83d5ec68	rejected	weighted_median_casscf_kernel_seconds	39.086641035	Cache same-run Newton-CASSCF invariant DF/hcore intermediates and canonicalize from the exact MO-basis Fock built from materialized ppaa/papa, preserving exact AO2MO/CASCI energies and solver tolerances [Correctness passed, but the 39.086641035s primary metric is only ~1.66% faster than incumbent and misses the required 2% improvement.]