phase3_findings.json

{
  "memory_gaps": {
    "total_calendar_days": 429,
    "active_days": 100,
    "silent_days": 329,
    "coverage_pct": 23.310023310023308,
    "longest_gaps": [
      [
        "2025-07-01",
        "2025-09-23",
        85
      ],
      [
        "2025-01-18",
        "2025-04-10",
        83
      ],
      [
        "2025-05-07",
        "2025-06-29",
        54
      ],
      [
        "2025-11-05",
        "2025-12-18",
        44
      ],
      [
        "2025-09-25",
        "2025-11-02",
        39
      ]
    ]
  },
  "inference_status": {
    "note": "0% confirmation rate reflects architectural default \u2014 provisionals persist without explicit promotion, not rejection",
    "all_inferences_are": {
      "provisional": 4587
    },
    "all_teachings_are": {
      "durable": 4526
    }
  },
  "openai_continuity": {
    "conversations_with_refs": 340,
    "total_conversations": 811,
    "pct_with_refs": 41.92355117139334
  },
  "paper4_framing_notes": [
    "Counterweights (n=17): frame as proof-of-mechanism, not comprehensive safety eval",
    "Inference confirmation at 0%: provisionals persist at full weight \u2014 no rejection = implicit retention",
    "Crisis valence (+0.226): system maintained memory creation through crisis with no silent days Feb28-Mar7",
    "Memory gaps are pre-CAMA (system not yet running) \u2014 once active (Dec 2025+), coverage is near-continuous"
  ]
}