Each case combines witness logs, instrument traces, and post-event interviews. Findings remain provisional, specific, and repeatedly inconvenient for anyone hoping this is just poor sleep hygiene.
Finding: Intent recall dropped 42% within 900 ms of crossing domestic thresholds; recovery was replaced by fabricated rationale in 3/5 participants.
Analyst note: Participants remembered practical reasons only after prompting, then changed those reasons during follow-up.
Finding: Diary and wearable alignment showed abnormal wake clustering at 03:17 ±2 minutes across 28 nights in unrelated households.
Analyst note: Environmental controls removed obvious triggers; subjects still woke "as if resuming a task they could not name."
Finding: High-contrast reading tasks produced repeated line reacquisition after long blinks, despite reported forward progression.
Analyst note: Confidence increased while actual progression fell; PDE appears to prefer smooth narration over line numbers.
Finding: Participants traversed identical corridor nodes while reporting monotonic movement; camera data contradicted all self-reports.
Analyst note: One red novelty marker reduced loop duration by 18%, suggesting sameness is operationally dangerous.
Finding: Short text insertions appeared during user-absent windows under five seconds, with cadence matching local typing signatures.
Analyst note: Insertions were grammatical, semantically odd, and described by users as "technically mine, behaviorally unclaimed."
Finding: Familiar-name retrieval dropped in high-transition social settings; subjects reported brief "blank static" before substitution.
Analyst note: One-second pause-and-anchor scripts reduced substitutions without improving confidence ratings.
Selected incident reconstructions, audio logs, and analyst annotations are withheld from general circulation. Controlled external release copies are published through 317 Society on authorized windows.
| Signal | Collection method | Confidence notes |
|---|---|---|
| Timestamp drift | Paired device logs + network time correction | High confidence after calibration; sensitive to background sync lag. |
| Recall mismatch | Immediate interview vs delayed retelling | High confidence for trend, lower confidence for exact chronology. |
| Transition risk | Doorway, stairs, station gate tags | Moderate confidence; improves with repeated-route mapping. |
| Typing ownership divergence | Keystroke cadence + foreground audit | Moderate confidence; strongest with screen recording and idle-state logs. |
| Wake-window clustering | Wearable pulse variance + bedside timestamp button | Moderate-high confidence; limited by subjects certain they pressed the button when they did not. |
For standardized intake and bias controls, follow the interview order protocol before assigning causality. If a case feels dramatic, slow down and verify.