Uncited Press › Star Trek Open the interactive journal →
Trek · Physics & Instrumentation

Counting Causal Loops: Detection Bias in 342 Classified Incidents of the Department of Temporal Investigations, 2310–2374

Dr. Wendell Pargeter1
1 Federation Department of Temporal Investigations
Received 10 Sep 2026 · Revised 15 Oct 2026 · Accepted 24 Oct 2026

Abstract

Incident records of the Department of Temporal Investigations place each event in one of three categories, a closed loop, a branch or an incursion later reversed, and the counts in those categories have been used to argue that closed loops are the commonest kind of temporal event. Reversed incursions undo most of their own consequences, so that they leave fewer records than the other two categories, and the raw counts may therefore mislead. Here we employ a two-list estimator on 342 classified incidents recorded between 2310 and 2374, using the contemporaneous field report and the Department's later audit of retained subspace and sensor logs as the two lists. The raw series holds 163 closed loops, 121 branches and 58 reversed incursions, a ratio of loops to reversed incursions of 2.81 (95% CI 2.08 to 3.79). After correction the estimated totals are 177.9, 140.9 and 138.5, and the ratio falls to 1.28 (95% CI 0.78 to 2.13). The series is estimated to hold 91.6% of closed loops but only 41.9% of reversed incursions. We regard it as likely that the two kinds occurred at comparable rates, although the estimate rests on two lists whose independence cannot be tested, the interval for reversed incursions is wide, and plausible misclassification or dependence confined to closed loops would move the ratio to between 1.4 and 1.6.

1. Introduction

The two-list approach to counting a partly observed population was devised for groups whose members can escape every register, and its logic transfers to the incident records of the Federation Department of Temporal Investigations (hereafter the Department) with one adjustment, which concerns the assumption that the two lists are independent. Here we employ it on the 342 classified temporal incidents of the Department's closed series for 2310 to 2374, taking the contemporaneous field report and the Department's later audit of retained subspace and sensor logs as the two lists (Federation Department of Temporal Investigations, 2375). We write in 2376, and no incident later than 2374 is counted, while the latest record used is the Department's series of 2375, compiled after the audit closed.

Temporal incidents have drawn sustained work in the Department, in part because the cost of misjudging one is severe, and in part because the Temporal Prime Directive depends on a clear account of what has and has not been altered. Each incident in the series is placed in one of three categories, namely a closed loop (a predestination in which the event is its own cause), a branch (a divergence that persists as a separate line of events) and an incursion later reversed (a change that was undone, so that the history containing it no longer exists). The scheme has been in use since 2372 (Pargeter, 2372), and in the Department's own audits of the 2360s its categories have rarely been overturned, among which cases the 2366 incursion in which the Federation and the Klingon Empire were at war, undone when the Enterprise-C returned through a temporal rift, is the example most often cited for reversed incursions.

A scheme that sorts incidents cannot by itself say how often each kind occurs, because the categories differ in the evidence they leave behind.

Closed loops leave their evidence in place, since nothing is undone and the participants, the logs and the artefact at the centre of the loop all remain. A reversed incursion removes most of its own consequences, so that only the instruments and persons shielded from the reversal retain any trace of it (Oskarsen, 2371). If the Department's records are taken as counts of what happened, they would therefore overstate the share of closed loops, and the excess of loops over reversed incursions that has been read in liability proceedings as a sign that history resists alteration would follow from the way incidents enter the record. Whether that is so, and by how much, is the question the paper puts to the series.

2. Methods

The Department's closed series of classified incidents (Federation Department of Temporal Investigations, 2375) holds 371 entries for the years 2310 to 2374. We removed 14 duplicate entries, each describing an incident already in the series under another filing agency, and excluded 15 incidents that the Department had been unable to place in any of the three categories, which left 342 incidents. The series was closed by the Department in 2375. Final categories are those of the Department's review panels, staffed by agents of the kind who conducted the 2373 review of the Deep Space 9 case (Dulmer and Lucsly among them), and they may differ from the category in either source's first record.

Source A is the contemporaneous report, defined as a written account filed within 90 days of the event by the starship, station or Department officer who was present. Source B is the Department's retrospective audit, in which analysts searched the raw subspace and sensor logs that Starfleet retains (Starfleet Command, Records Division, 2374) for chroniton residue and for discontinuities in log sequence, without reference to any field report, which the analysts did not see during the search. The audit was carried out from 2373 to 2375, and each source recorded a provisional category from its own evidence alone. An incident enters the counts as recorded by A only, by B only or by both, and every one of the 342 incidents was recorded by at least one source.

For each category separately we estimated the number of incidents recorded by neither source with a bias-corrected two-list estimator (Quillfeather, 2369), which treats the count found by both sources as the information about how closely the lists overlap. The variance comes from the same source, and intervals for the unrecorded count are formed on the log scale so that they cannot fall below zero. The estimator assumes that, within a category, each list captures incidents independently of the other and that every incident has the same chance of capture. Positive dependence between lists, which arises when one durable trace feeds both, makes the estimator undercount (Thessaly, 2372), and we therefore add a sensitivity analysis in which the unrecorded count is multiplied by an odds ratio of dependence (ψ) of 1.5 and of 2.

We compared categories by the ratio of the corrected loop count to the corrected reversed count, with a 95% interval and a Wald test on the log scale, and we compared the raw counts with an exact conditional binomial test, the standard Poisson comparison of two counts observed over the same 65 years. All tests are two-sided at the .05 level. With a count of only 7 incidents in both lists for the reversed category, the Wald interval for that category is crude, and we say so where it matters.

3. Results

The series records 342 incidents, of which 163 are closed loops (47.7%), 121 are branches (35.4%) and 58 are reversed incursions (17.0%). That is 2.81 loops for every reversed incursion (95% CI 2.08 to 3.79), and an exact conditional binomial test rejects equal counts decisively (p < .001). The raw rates are 2.51 loops a year against 0.89 reversed incursions.

Source A recorded 254 incidents and source B 238 (Table 1). In all, 150 incidents appear in both of the lists. Among the incidents found by both sources, the provisional categories agreed in 138 (92.0%), and overlap differed greatly between categories, in particular with 89 of 163 closed loops (54.6%) in both lists against 7 of 58 reversed incursions (12.1%). Agreement on the provisional category was lowest for the reversed incursions, 4 of 7 (57.1%), against 84 of 89 closed loops (94.4%) and 50 of 54 branches (92.6%).

Table 1 sets out the counts for each category together with the corrected totals.

The estimated number of incidents that neither source recorded was 14.9 closed loops, 19.9 branches and 80.5 reversed incursions, 115.3 in all (95% CI 63.0 to 211.1). The corrected totals are therefore 177.9 closed loops (95% CI 170.5 to 192.8), 140.9 branches (131.1 to 160.2) and 138.5 reversed incursions (93.4 to 241.2), 457.3 in all (405.0 to 553.1). The series is therefore estimated to hold 91.6% of closed loops, 85.9% of branches and only 41.9% of reversed incursions.

Against the benchmark of the raw counts, the correction reduces the ratio of loops to reversed incursions from 2.81 to 1.28 (95% CI 0.78 to 2.13; Wald p = .33), so that the corrected counts no longer differ detectably. The corrected annual rates are 2.74 closed loops and 2.13 reversed incursions a year. The share of reversed incursions among all incidents rises from 17.0% to 30.3%, and the share of closed loops falls from 47.7% to 38.9%.

Positive dependence, the only kind modelled here, can only raise each corrected total. With ψ of 1.5 the corrected ratio is 1.04 and the total is 515 incidents, and with ψ of 2 the ratio is 0.88 and the total is 573, so that reversed incursions then outnumber closed loops (Table 2). If dependence is confined to closed loops (ψ of 2 for loops and 1 for reversed incursions), the ratio is 1.39.

If the five audit-only reversed incursions in excess of the field-only count are reassigned to closed loops, the corrected totals become 185.3 closed loops and 119.1 reversed incursions, and the ratio is 1.56 (95% CI 0.95 to 2.55).

4. Discussion

In these records, most of the surplus of closed loops over reversed incursions in the raw series is a product of detection. The raw ratio of 2.81 falls to 1.28 once the unrecorded incidents are estimated, which is a clear improvement over the raw count as a basis for comparison, and the corrected interval includes equal frequency. We regard it as likely that closed loops and reversed incursions occurred at comparable rates between 2310 and 2374, although the interval for reversed incursions is wide (93.4 to 241.2) and extends well above the corrected loop count.

One alternative is that the shift reflects classification and not detection. If audit analysts assigned ambiguous residue to the reversed category more freely than field officers did, the reversed count would be inflated in the audit list and the overlap with field reports would fall. The provisional categories agreed in 92.0% of the 150 incidents found by both sources, but the reversed category is the only one in which the audit alone found more incidents than field reports alone (28 against 23), which is the direction a generous audit would produce, and agreement among the 7 reversed incursions found by both lists was the lowest of the three categories. Reassigning that excess to closed loops raises the corrected ratio to 1.56, an interval that still includes equal frequency, so misclassification could account for part of the shift but not for most of it.

Positive dependence raises every corrected total, and its effect on the ratio depends on which category it affects most: a common ψ lowers the ratio, whereas dependence confined to closed loops, the category with the most durable shared trace, raises it to 1.39. The sensitivity analysis shows how far a common ψ could go, with a ratio of 1.04 at ψ of 1.5 and 0.88 at ψ of 2. These values depend entirely on the assumed dependence. We offer neither of these values as estimates, because two lists cannot identify the dependence, which would need a third source that is independent of both and which the Department has not yet compiled for this period.

For Department practice, the consequence is that the count of classified reversed incursions is a floor and not a rate. The point matters for every hearing that relies on the series. A decade with few classified reversed incursions may simply be a decade whose logs were not kept, and a hearing that treats the surviving ratio of loops to reversed incursions as a property of causality relies on a statistic that the record itself produces.

5. Limitations

The estimate rests on two lists, which is the smallest number that allows any correction, and with two lists the independence assumption cannot be tested from the data. The reversed category is estimated from 7 incidents found by both sources, so that, owing to that small overlap, its interval is wide and its Wald variance is crude; the lower bound of 93.4 lies not far above the 58 recorded. Detection probabilities are taken as constant within a category, whereas fewer raw logs survive from the earlier decades under the retention schedule (Starfleet Command, Records Division, 2374), and heterogeneity of that kind lowers the estimate for the oldest incidents. The 15 incidents excluded as unclassifiable are missing from every count, and the corrected totals describe only events that some agency treated as incidents at all, so that a Department that adopted a stricter or looser definition of an incident in any decade of the 65 years would have produced a different series and a different correction, a change that the present data cannot separate from a change in the underlying rate.

Our result stands or falls on the reach of the method, and the method reaches only incidents that at least one list could have recorded, so a reversed incursion of which no log retains a trace and which no witness recalls lies beyond it. A third source independent of both, such as the testimony of personnel who remember a history that no longer exists, would allow dependence between the lists to be estimated directly, and that possibility opens an avenue for a corrected count with a narrower interval for reversed incursions.

temporal incidentsclosed causal loopsdetection biascapture-recapturereversed incursionsincident classification

References

  1. Federation Department of Temporal Investigations (2375). Classified incident series, temporal incidents 2310–2374, with source and category fields. Starfleet Incident Records, series SIR-75-TI.
  2. Pargeter, W. (2372). A three-category scheme for the classification of temporal incidents in liability proceedings. Federation Council Legal Research Office Reports, LRO-72-09.
  3. Oskarsen, L. (2371). Persistence of chroniton residue in shipboard sensor logs after a reversed timeline. Journal of Applied Field Dynamics, 11(1), 40–63.
  4. Quillfeather, M. (2369). Estimating the size of a partly observed population from two overlapping lists. Proceedings of Applied Speculative Statistics, 17(3), 142–165.
  5. Thessaly, R. (2372). Dependence between lists and the bias of two-list estimators. Proceedings of Applied Speculative Statistics, 20(2), 77–99.
  6. Starfleet Command, Records Division (2374). Retention schedule for raw subspace and sensor logs, 2300–2374. Starfleet Fleet Telemetry Archive, schedule FT-RS-74.
Read this article inside the full journal experience — browse by faculty, search across universes, and filter by discipline.
Open in Uncited Press →