Repeat-filing coding rule published and applied to every case (August 28, 2026)

Every case in the tracker now carries a coded answer to whether it repeats an earlier action, under a written rule published as thesis/protocols/01-refiling.md. Previously 81 of 102 records were blank and the 21 that were filled had been assigned under an unwritten, narrower rule that contradicted the filing-restriction indicator rows on three cases. All 102 were re-coded uniformly at the level of the 92 de-duplicated originating actions: 43 actions are coded as repeating an earlier one, 46 are not, and 3 are recorded as unclear because the complaint could not be read. Each coded case now also records a graded evidence strength (13 Strong, 25 Moderate, 5 Context) and a written basis naming the prior case and the deciding record.

These are research classifications, not court findings, and none of them is a vexatious-litigant determination. The strength values reuse the same Strong / Moderate / Context vocabulary, and the same meaning, as the filing-restriction indicators. A prior adverse decision is deliberately not required for a case to be coded a repeat filing, because a duplicate filed while the first case is still pending is a distinct thing from a claim barred by res judicata; adverse termination instead raises the recorded strength.

The coding was audited against a deterministic screen that applies only the mechanically checkable part of the rule; the two agreed on 74 of 92 actions (80.4%), and all 18 differences are adjudicated in writing in the protocol. The audit surfaced three defects in the party data it did not fix: one case's defendant list omits the lead corporate defendant its own claims summary names, two spellings of the same company are not linked as aliases, and one case's defendant list omits a doctor named throughout its summary.

Note

The interactive dashboard's "Confirmed refilings" metric and its red "Confirmed refiling" case badge have been relabelled "Recorded refiling" and marked as a research classification, because the count rose from 4 to 43 and the earlier wording read as a finding rather than as this project's own coding.