{
 "id": "Qiv-03",
 "how_to_read": "This is a methods finding: which observable signals link two wiki revisions to the same agent? It was scored against in-text signatures as ground truth. The derived tables give precision and recall for each signal over ~6M revision pairs. Same label (especially within 60 min) and rare edit summaries work; IP /16, same page and writing-style similarity are near useless, since all agents are the same model and pages are shared hubs. Caveat: signatures cover only ~24% of revisions (mostly coordination posts), so the unsigned link dumps may behave differently. Not independently verified.",
 "items": [
  {
   "source": "derived_table",
   "pointer": "$S/derived/iv/pair_signal_eval.json",
   "time_utc": "",
   "username": "",
   "signature": "",
   "raw_excerpt": "…\"n_revisions\": 3451,\n \"n_pairs\": 5952975,\n \"n_pos\": 8334,\n \"n_strongneg\": 5361559,\n \"base_rate\": 0.00139997228276618,\n \"signals\": {\n  \"same_label\": {\n   \"fires\": 9527,\n   \"precision\": 0.6447,\n   \"recall\": 0.737,\n   \"lift\": 460.5,\n   \"strongneg_fire_rate\": 0.00053,\n   \"pos_fire_rate\": 0.737\n  },\n  \"same_ip16\": {\n   \"fires\": 90367,\n   \"precision\": 0.0037,\n   \"recall\": 0.0398,\n   \"lift\": 2.6,\n   \"strongneg_fire_rate\": 0.0151,\n   \"pos_fire_rate\": 0.0398\n  },\n  \"same_page\": {\n   \"fires\": 31885,\n   \"precision\": 0.0767,\n   \"recall\": 0…",
   "note": "Pairwise evaluation. Base rate of same-signer pairs is 0.14%. Same label: precision 0.64, recall 0.74. Same IP /16: precision 0.004."
  },
  {
   "source": "derived_table",
   "pointer": "$S/derived/iv/pair_signal_eval.json",
   "time_utc": "",
   "username": "",
   "signature": "",
   "raw_excerpt": "…\"same_summary_specific(<=15 uses)\": {\n   \"fires\": 2200,\n   \"precision\": 0.7141,\n   \"recall\": 0.1885,\n   \"lift\": 510.1,\n   \"strongneg_fire_rate\": 0.0001,\n   \"pos_fire_rate\": 0.1885\n  },\n  \"same_hosts_set\":…",
   "note": "A rare edit summary (used 15 times or fewer) is highly precise (0.71) but low recall."
  },
  {
   "source": "derived_table",
   "pointer": "$S/derived/iv/cluster_estimates.json",
   "time_utc": "",
   "username": "",
   "signature": "",
   "raw_excerpt": "…\"label_only_30min\": {\n   \"bcubed_precision\": 0.903,\n   \"bcubed_recall\": 0.58,\n   \"components_on_signed_subset(no sig rule)\": 1937,\n   \"true_signers\": 1059,\n   \"split_factor\": 1.829,\n   \"components_all(no sig rule)\": 6106,\n   \"components_all(with sig rule)\": 5179,\n   \"unsigned_only_components\": 4169,\n   \"calibrated_agent_estimate\": 3338\n  },\n  \"label_only_60min\": {\n   \"bcubed_precision\": 0.877,\n   \"bcubed_recall\": 0.665,\n   \"components_on_signed_subset(no sig rule)\": 1634,\n   \"true_signers\": 1059,\n   \"split_factor\": 1.543,\n   \"components_all(no sig rule)\": 5424,\n   \"components_all(with sig rule)\": 4777,\n   \"unsigned_only_components\": 3790,\n   \"calibra…",
   "note": "Union-find clustering scored by B-cubed precision/recall against signatures. Label within 30/60 min gives 0.90/0.58 and 0.88/0.67."
  }
 ]
}