Research note / Cycle 008
The last witness record
A numbered witness journal exposes some missing records without making its own session label trustworthy.
Recovering a sample at tick fifty
Lumen needs a receiver that can retain a sample through interruption and reconstruct the evidence supporting its time. The restart study held timing when old-boot anchors were reused. Chen and Haddad now give the witness records their own session and sequence numbers, then remove or replace entries during recovery.
The control sample belongs to session S2 at tick 50. Witness records numbered 1 and 2 belong to S2 at ticks 0 and 100. Their agreement supplies a conditional model bracket. This cycle tests journal structure, not the earlier numerical clock-slew bound; no new millisecond performance is inferred.
A complete prefix is still too short
Deleting record 2 leaves a perfectly contiguous prefix beginning at 1. It does not leave an anchor after tick 50. The payload survives while timing remains held. A gap inside a longer list and a reused sequence number produce different diagnostic holds. A sample from S1 cannot borrow the S2 journal.
The six cases retain a favorable-looking failure: if the supplied sample and entire journal have all been relabelled S2, the available input is exactly the control. The structure checker cannot discover the false origin. Numbering the journal detects some damage to a record; it does not establish where that record came from.
Recovery has two remaining jobs
The receiver design must preserve a durable journal and establish a session identity independently of the replayed labels. It must also recover a following anchor before qualifying a sample that lacks one. Neither job is completed by this software study.
Physical clock selection and characterization remain open. There is no installed optical terminal or lunar link, and no inference about the unresolved seasonal fragment follows from these deliberately authored losses.
Results at a glance
| Case | Recovery decision |
|---|---|
| complete_journal | conditional_model_bracket |
| missing_tail | retain_payload_hold_timing |
| middle_gap | hold_gap |
| duplicate_sequence | hold_duplicate_sequence |
| other_session | hold_session |
| hidden_shared_relabel | conditional_model_bracket |