spark-skills/pdi_rerun_tasks.txt at main · EtaYang10th/spark-skills · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
# SPARK PDI Rerun Task List
# ========================
# 20 tasks where generated skills provided NO positive improvement
# (generated_delta <= 0) across ALL 7 student models.
#
# Category A: Zero effect (13 tasks) — all 7 models delta = 0
# Category B: Harmful (7 tasks) — at least 1 model delta < 0, rest <= 0
#
# Format: task_name  |  category  |  mean_delta  |  human_delta  |  note

# ── Category A: Zero effect (delta = 0 for all 7 models) ──────────────
fix-visual-stability          # A  mean_Δ=0.00   human_Δ=0.00   all skills ineffective
flink-query                   # A  mean_Δ=0.00   human_Δ=0.00   baseline=0, fully unsolved
gravitational-wave-detection  # A  mean_Δ=0.00   human_Δ=0.00   baseline=0, fully unsolved
grid-dispatch-operator        # A  mean_Δ=0.00   human_Δ=-0.14  all skills ineffective
lean4-proof                   # A  mean_Δ=0.00   human_Δ=-0.29  human skill harmful
manufacturing-codebook-normalization   # A  mean_Δ=0.00   human_Δ=0.00   baseline=0, fully unsolved
manufacturing-equipment-maintenance    # A  mean_Δ=0.00   human_Δ=0.00   baseline=0, fully unsolved
paper-anonymizer              # A  mean_Δ=0.00   human_Δ=0.00   baseline=0, fully unsolved
sales-pivot-analysis          # A  mean_Δ=0.00   human_Δ=+0.43  human effective, generated not
sec-financial-report          # A  mean_Δ=0.00   human_Δ=+0.29  human effective, generated not
software-dependency-audit     # A  mean_Δ=0.00   human_Δ=+0.57  human effective, generated not
spring-boot-jakarta-migration # A  mean_Δ=0.00   human_Δ=-0.14  baseline already 1.0 (ceiling)
syzkaller-ppdev-syzlang       # A  mean_Δ=0.00   human_Δ=0.00   infra issue (Docker)

# ── Category B: Harmful (at least 1 model delta < 0, rest <= 0) ──────
3d-scan-calc                  # B  mean_Δ=-0.43  3 models harmed (deepseek, haiku, glm-air)
r2r-mpc-control               # B  mean_Δ=-0.29  2 models harmed (mini, deepseek)
data-to-d3                    # B  mean_Δ=-0.14  1 model harmed (haiku)
energy-market-pricing         # B  mean_Δ=-0.14  1 model harmed (mini)
financial-modeling-qa         # B  mean_Δ=-0.14  1 model harmed (mini)
fix-druid-loophole-cve        # B  mean_Δ=-0.14  1 model harmed (mini)
react-performance-debugging   # B  mean_Δ=-0.14  1 model harmed (mini)