# Re-run of the pn_model_k1 system of residual-ext.yaml (the first run named it with a capital L, which Postgres folds in the index name), both C1 cells.
# Same harness, same dataset, 1 and 8 clients. Cells: C1 (1% key + 0.3% residual, 33 rows), C2 (10% key + 0.3% residual, 326 rows).
#   pn_pool_k1 / pn_pool_k10: relaxed_order, cap 20000, rescore_multiplier {50,100,200} x scan_fraction {0.5 (ramp), 1.0}
#   pn_model_k1: mode_l_threshold left at the shipping 0.03, so the 1% key routes to mode L (exact over the key rows)
name: residual_ext2
seed: 42
k: 10
warmup_queries: 20
hardness_recall_threshold: 0.97
min_throughput_seconds: 5.0

postgres:
  maintenance_work_mem: 8GB
  min_wal_size: 1GB
  max_wal_size: 4GB

dataset:
  kind: real
  artifact: wikidpr_768d_1m.npz
  n: 1000000
  dim: 768
  n_centers: 1000
  kmeans_sample: 100000
  kmeans_iters: 15
  n_queries: 500

metric: l2

filters:
  query_modes: [out_region]
  run:
    - {selectivity: 0.01, mode: correlated, residual: {selectivity: 0.003, mode: uncorrelated}}
    - {selectivity: 0.01, mode: correlated, residual: {selectivity: 0.01, mode: uncorrelated}}
    
systems:
  - name: pn_model_k1
    type: pg_pinecone_native
    skip_unfiltered: true
    record_scan_stats: true
    explain_probe: true
    build_params: {seed: 42, filter_column: filt_correlated_0010_eq}
    sweep_param: mode_l_threshold
    sweep_values: [0.03]
    iterative_scan_values: ["off"]

concurrency: [1, 8]

churn:
  enabled: false

sift1m:
  enabled: false
