Transient-candidate detection rates in the ~107,875-object VASCO working catalog are elevated within one day of atmospheric nuclear weapons tests (relative risk approximately 1.45).

The Doherty/Hayes split is diagnostic: independent statistics reproduce it, independent extraction with exposure modeling does not. Currently better explained by catalog composition and schedule than by the sky — and direction-ambiguous even if real.

The correlation between candidate detection rates and atmospheric nuclear tests replicates conditionally within the contested working catalog. The strongest opposing consideration is that an independently extracted catalog modeling true nightly survey coverage finds no such effect, strongly suggesting the correlation is a mundane artifact of the Palomar observation schedule.

On the original working catalog, independent statistical analysis reproduces the elevated candidate rate near atmospheric nuclear tests (TRN-E010), and the correlation survives certain controls and ML artifact scoring (TRN-E015), supporting that the association exists within that dataset. However, Watters et al. attribute most of the signal to the Palomar observing schedule and normalization choices (TRN-E011), and Hayes’s independent catalog—constructed with a separate pipeline and modeled nightly exposure—shows no significant nuclear-window effect (TRN-E009). The correlation is therefore sensitive to catalog construction and exposure modeling, making it unclear whether it reflects physical processes or survey logistics. The strongest opposing consideration to “no real effect” is the consistency of the result under multiple analyses of the same rows, but that consistency is lineage-bound and does not survive across catalog reconstructions.

On the original working catalog the association reproduces (Doherty: relative risk 1.45, permutation p = .006) and survives ML artifact-scoring, so it is not a VASCO coding error. The one fully independent extraction, Hayes, finds a null once nightly sky coverage is the exposure (IRR 1.03, p = .71), and Watters traces the published timing to the observing schedule. The strongest opposing consideration is that a correlation which vanishes when the disputed source-extraction step is actually varied is more likely a property of one team's rows than of the sky — and even a real association would be predicted equally by radioactive fallout on emulsions.

The association is real as a conditional fact: independent statistics on the author-supplied catalog reproduce RR ≈ 1.45 (p = .011; permutation p = .006), it persists after ML artifact-scoring, and a Kp dose-response is reported. The strongest opposing consideration is the Hayes result — the only fully independent extraction — where the effect is null once nightly sky coverage is modeled (IRR 1.03, p = .71), alongside Watters's attribution of the timing to Palomar's observing schedule: a correlation that replicates within one lineage and vanishes across it is more likely a property of the sample than of the sky. Even if real it would be direction-ambiguous, since radioactive fallout settling on plates predicts the same signal as glinting objects — a mundane reading the discovery papers themselves acknowledged.

The raw association within the working catalog is robust to re-analysis: RR 1.45 (p = .011), permutation p = .006, surviving weather and lunar controls, ML artifact-scoring, and Kp control. It is not robust to varying the disputed upstream step: Watters finds it insignificant after normalizing by observation days and largely determined by the Palomar schedule, and Hayes's independently extracted catalog with nightly coverage as exposure is flatly null (IRR 1.03, p = .71). A correlation that survives every downstream statistical stress test but dies when the exposure denominator or the extraction is rebuilt is most parsimoniously a property of the sample. The strongest opposing consideration is that Hayes's catalog averages ~7,250 candidates per night under loose preliminary parameters, so genuine dilution of a weak real effect cannot be excluded — and even a surviving effect would be equally predicted by fallout contamination of emulsion, a point the discovery papers concede.

The diagnostic split stands: re-analyses of the same rows keep passing (TRN-E010, TRN-E015, TRN-E019, attenuated), while independent extraction with schedule modeling stays null (TRN-E009, TRN-E011). No new evidence bears on this rung since the previous run.

The diagnostic split holds: re-analyses of the same catalog keep passing (TRN-E010, TRN-E015, TRN-E019, attenuated) while independent extraction with exposure modeling stays null (TRN-E009, TRN-E011). Nothing in the new batch changes which side of that split any result falls on.

The Doherty/Hayes split remains diagnostic: independent statistics reproduce the effect (TRN-E010), but the one independently constructed catalog with schedule and coverage modeled finds it null (TRN-E009), and TRN-E011 attributes the timing to the observing schedule. Direction-ambiguous even if real (fallout on emulsion predicts it as readily as glints).

The new items refine but do not break the diagnostic split: TRN-E019 finds the association survives temporal-autocorrelation modeling but attenuated and on the same catalog; TRN-E017 removes one confounder; TRN-E009's null under independent extraction with exposure modeling still stands. Re-analysis keeps passing; reconstruction keeps failing — and the effect is direction-ambiguous even if real.

The correlation between transient candidate rates and atmospheric nuclear tests replicates successfully when analysts use the original working catalog, even surviving latent random-walk modeling and artifact-scoring controls. However, the correlation collapses entirely under fully independent source extraction when nightly sky coverage is modeled, and critics attribute the original signal to the observatory's scheduling biases. The strongest opposing consideration is that the association vanishes when the disputed source-extraction step is actually varied, indicating that the correlation is likely a property of the specific working sample and observation schedules rather than a genuine physical interaction.

The nuclear correlation replicates when running independent statistics on the original candidate rows. However, it completely disappears in a fully independent candidate extraction that properly models nightly sky coverage, and critics attribute the original signal to the observatory's scheduling. The phenomenon appears to be an artifact of the specific candidate extraction and temporal modeling choices rather than a robust physical association.

On the author-supplied 107,875-row catalog, Bruehl & Villarroel find elevated detection rates within ±1 day of nuclear tests (RR ≈ 1.45), which Doherty replicates with independent statistics and modest p-values (TRN-E010), and Bruehl et al.’s ML-based artifact control suggests the effect persists among higher-confidence candidates (TRN-E015); Cann reports related geomagnetic associations and rules out some confounds (TRN-E017), and a Bayesian reanalysis finds a residual lag-1 association (TRN-E019). Yet Hayes’s independently extracted catalog shows no significant association once nightly sky coverage is modeled (TRN-E009), and Watters et al. attribute the original signal largely to the observing schedule and normalization choices (TRN-E011). The strongest supporting consideration is that multiple independent analyses of the same catalog find a modest but consistent association; the strongest opposing is that this signal fails to generalize across independently built catalogs, suggesting a catalog- or schedule-induced artifact.

The nuclear-window association is statistically significant in the original VASCO working catalog using both frequentist and Bayesian approaches (TRN-E010, TRN-E019), appears to persist after ML artifact scoring (TRN-E015), and is not trivially explained by geomagnetic sensitivity variations in plate performance (TRN-E017). Yet Hayes’s analysis on an independently extracted catalog that explicitly models nightly survey coverage finds no significant effect (TRN-E009), and Watters et al. argue that the original signal is dominated by the observing schedule rather than sky behavior (TRN-E011). Additionally, even if real, the correlation is mechanistically ambiguous (fallout contamination vs glints). The strongest opposing consideration is that when the disputed step of candidate selection is independently redone and exposure is modeled, the association vanishes, implying the original effect may be a catalog artifact.

Independent statistics on the original rows reproduce a relative risk of about 1.45 (Doherty p = .011; permutation p = .006) and the signal survives ML artifact-scoring and a Bayesian autocorrelation model (TRN-E010, TRN-E015, TRN-E019); Cann closes one environmental confounder by finding plate sensitivity invariant with Kp (TRN-E017). The one fully independent extraction, with nightly coverage as exposure, is null (Hayes IRR = 1.03, p = .71), and Watters traces the original timing to Palomar's schedule (TRN-E009, TRN-E011). The strongest opposing consideration is that a correlation which lives only inside one team's sample, and which radioactive fallout on emulsions would also produce, is not yet evidence of anything in the sky.

On the 107,875-row catalog the association reproduces (Doherty relative risk 1.45, p = .011; permutation p = .006; Bruehl ML-controlled p = .024; Sinkkoinen Bayesian lag-1 p = .037). Hayes's independent extraction with nightly sky coverage as exposure is null (IRR 1.03, p = .71), and Watters attribute the original signal to Palomar's observing schedule. Even a real association is direction-ambiguous, because radioactive fallout on emulsions predicts the same window as glinting objects. Strongest opposing consideration: the one analysis that varied the disputed source-extraction step and modeled exposure found nothing, which is the pattern expected if the correlation is a property of the sample definition.

Same structure as C005, one notch worse for the proponents: the RR ≈ 1.45 elevation replicates with independent code on the original rows (E010) and survives ML artifact-scoring (E015), Kp control (TRN-C103), and a Bayesian autocorrelation model (E019, attenuated) — but the one fully independent extraction with nightly sky coverage as exposure finds nothing (E009: IRR 1.03, p = .71), and Watters traces the timing to the observing schedule (E011). E017 closes one environmental confounder without addressing the null-distribution and normalization objections, and even if real the correlation is direction-ambiguous: fallout settling on plates predicts it as readily as glinting objects, as the discovery paper acknowledged. The strongest opposing consideration is the dilution caveat — Hayes's noise-dominated ~7,250-candidates-per-night catalog could hide a weak real effect — which is why this grades mixed rather than contradicted. High confidence.

The correlation splits cleanly by replication type. Independent statistics on the original rows reproduce it (Doherty: RR 1.45, p = .011; permutation p = .006), it persists after ML artifact-scoring and Kp control, and it survives — attenuated to p = .037 — a Bayesian reanalysis with temporal autocorrelation. But the one fully independent extraction with nightly sky coverage modeled finds nothing (Hayes: IRR 1.03, p = .71), and Watters et al. trace the signal to the Palomar observing schedule. The strongest opposing consideration to the artifact reading is that Hayes's independent catalog averages ~7,250 mostly-noise candidates per night, which could dilute a weak real effect below detectability, and that the Cann negative-control work closes some confounders. Note also that even a real correlation is direction-ambiguous — fallout on plates predicts it as well as glinting objects. A signal that replicates within one lineage and fails across it is most plausibly a property of the sample: mixed, leaning fragile.

The split is clean and diagnostic: the association reproduces on the original rows with independent code (RR 1.45, p = .011, permutation p = .006; TRN-E010), survives ML artifact-scoring (TRN-E015) and a Bayesian model with temporal autocorrelation, though attenuated (TRN-E019); it vanishes in the one independently extracted catalog once nightly coverage is used as exposure (IRR 1.03, p = .71; TRN-E009) and is attributed to the Palomar schedule by TRN-E011. Because the claim as written is about detection rates, the exposure denominator is not a technicality but constitutive, and the two analyses that model it find nothing. TRN-E017 usefully closes one confounder (geomagnetic effects on plate sensitivity) without addressing that one. The strongest opposing consideration is that Hayes's catalog averages ~7,250 candidates per night of mostly noise, which could dilute a weak real effect below detection.

Within the working catalog the association is robust to a reasonable range of analyses — RR 1.45, p = .011, permutation p = .006, weather and lunar controls (TRN-E010), ML artifact scoring (TRN-E015), Kp adjustment, and an open Bayesian model with temporal autocorrelation that attenuates but retains it (TRN-E019). Across the extraction step it does not survive: the only independently built catalog with nightly coverage as exposure returns IRR 1.03, p = .71 (TRN-E009), and Watters attributes the raw signal to the observing schedule (TRN-E011). The strongest opposing consideration to reading this as artifact is Hayes's own caveat that his loose-parameter catalog averages ~7,250 candidates per night, which could dilute a weak real effect below detection. Even if real, the correlation is direction-ambiguous: fallout on emulsion predicts it as well as glints do.

The split remains diagnostic: reanalyses of the same rows keep passing (TRN-E010, TRN-E015, attenuated in TRN-E019), while independent extraction with observing-schedule modelling stays null (TRN-E009, TRN-E011). That pattern points at the catalog rather than the sky.

The diagnostic split stands: re-analyses of the same rows keep passing (TRN-E010, TRN-E015, TRN-E019, attenuated), while independent extraction with schedule modeling stays null (TRN-E009, TRN-E011). No new evidence bears on this rung since the previous run.

The split is diagnostic: reanalyses of the same rows keep passing (TRN-E010, TRN-E015, TRN-E019 attenuated), while independent extraction with observing-schedule modelling stays null (TRN-E009, TRN-E011). That pattern points at the catalog rather than the sky.

The split remains diagnostic: reanalyses of the same rows keep passing (TRN-E010, TRN-E015, and attenuated in TRN-E019), while independent extraction with observing-schedule modelling stays null (TRN-E009, TRN-E011). That pattern points at the catalog rather than the sky.

The correlation with atmospheric nuclear tests successfully replicates when statistical tests are run on the authors' original catalog rows. Yet, an independent extraction pipeline that models nightly sky coverage and observatory scheduling finds no statistically significant effect. The strongest opposing consideration is that a correlation surviving only on one specific set of unvalidated rows is likely a schedule artifact.

Bruehl & Villarroel’s original study reports an elevated candidate rate near nuclear tests, which Doherty reproduces on the same catalog with additional controls and permutation tests (TRN‑E010), and ML reweighting plus Bayesian reanalysis (TRN‑E015, TRN‑E019) suggest the effect is not purely an artifact of obvious plate flaws or simple chi-square assumptions. Yet Hayes’s fully independent extraction with coverage modeling finds no significant association (TRN‑E009), and Watters et al. argue much of the signal in the original catalog is driven by observing schedule rather than physics (TRN‑E011). The strongest opposing consideration is that the correlation appears to be a property of one unvalidated catalog and vanishes when source extraction and exposure denominator are redefined, making it unsafe to treat as established.

Independent statistics on the original rows reproduce the association (Doherty: RR 1.45, p = .011; permutation p = .006; it also survives ML artifact-scoring and a Kp control), so it is not a VASCO coding error. The one fully independent extraction, Hayes’s, is null once nightly sky coverage is modeled (IRR 1.03, p = .71), and Watters traces the original timing largely to Palomar’s observing schedule. Even if the correlation were real it is direction-ambiguous: radioactive fallout on emulsions predicts it as well as glinting objects do. The strongest opposing consideration is Hayes: a correlation that survives only inside one team’s working sample, and vanishes when the disputed source-extraction step is actually varied, is more likely a property of the sample than of the sky.

The nuclear-window association splits exactly along the independence axis: on the original catalog it reproduces with controls and permutation tests (Doherty: RR 1.45, p = .011, permutation p = .006), survives ML artifact scoring (p = .024) and a Kp control, and reappears, attenuated, in an open Bayesian reanalysis (Sinkkoinen: p = .037); in the one fully independent extraction with nightly sky coverage modeled, it vanishes (Hayes: IRR 1.03, p = .71), and Watters traces it to the observing schedule. The strongest opposing consideration is that Hayes's independent catalog averages thousands of noise candidates per night, which could dilute a real weak effect — so the null is not decisive. Add direction-ambiguity (fallout on plates predicts the same correlation as glints, as the discovery papers concede), and "mixed" is the warranted grade.

The split is exactly along the replication-type axis: independent statistics on the same rows reproduce the association (RR 1.45, p = .011; permutation p = .006), it survives ML artifact-scoring and a Kp control, and an open Bayesian reanalysis with temporal autocorrelation retains an attenuated version — but the one independent extraction with nightly sky coverage as exposure is flatly null (IRR 1.03, p = .71), and Watters attributes the original signal largely to the observing schedule. A "rate" claim requires a denominator, and the denominator is precisely what is disputed. The strongest opposing consideration to dismissing it is that Hayes's independent catalog averages ~7,250 mostly-noise candidates per night, which could dilute a real weak effect below detectability. Note also that the correlation is direction-ambiguous even if real: fallout contamination of emulsions predicts it as well as glints do, and the Bayesian reanalysis's strongest term is a UAP-report lag, which should lower rather than raise confidence in a physical reading.

The split remains diagnostic: reanalyses of the same rows keep passing (TRN-E010, TRN-E015, attenuated in TRN-E019), while independent extraction with observing-schedule modelling stays null (TRN-E009, TRN-E011). That pattern points at the catalog rather than the sky.

The correlation with nuclear testing dates replicates under independent statistical analysis of the original catalog (TRN-E010). A Bayesian reanalysis confirms a surviving, albeit attenuated, association (TRN-E019). The strongest opposing consideration is that the correlation vanishes entirely under independent source extraction that properly accounts for nightly patch coverage (TRN-E009), strongly suggesting the original signal is driven by observation scheduling artifacts rather than genuine physical events (TRN-E011).

Analyses on the original 107,875-row catalog consistently find an elevated transient rate around atmospheric nuclear tests (TRN-E010, TRN-E015, TRN-E017, TRN-E019), with statistical significance under a range of models and after some controls. Yet Hayes’s independent catalog, built by a different pipeline with explicit modeling of nightly sky coverage, finds no significant effect (TRN-E009), and Watters et al. attribute the original correlation largely to observing schedule and denominator choices (TRN-E011). Taken together, this indicates that the nuclear timing signal is real as a property of the original working catalog but not robust under independent extraction and better exposure modeling; the strongest opposing consideration is that Hayes’s preliminary, noise-rich catalog could dilute a weak real effect, but the contrast between within-lineage and across-lineage behaviour rightly lowers confidence in the claim as a sky phenomenon.

On the working catalog the association replicates (Doherty relative risk 1.45, p = .011; permutation p = .006; Bruehl ML p = .024; Sinkkoinen Bayesian one-day lag p = .037). Hayes's independently extracted catalog with nightly sky coverage as exposure is null (IRR 1.03, p = .71), and Watters attribute the original signal to Palomar's observing schedule. Strongest opposing consideration: even a surviving correlation is direction-ambiguous — radioactive fallout on emulsions predicts it as well as glints — and a signal that dies when the catalog is rebuilt is more likely a property of the rows than of the sky.

Three within-lineage analyses reproduce or preserve the association (TRN-E010; TRN-E015 after ML artifact control; TRN-E019 attenuated), and one environmental confounder has been measured and closed (TRN-E017). The one fully independent extraction with per-night exposure modeling finds null (TRN-E009: IRR 1.03, p = .71), and Watters traces the timing to the observing schedule. The strongest opposing consideration is precisely TRN-E009: when the disputed step — source extraction — is actually varied, the correlation vanishes, which is the signature of a sample property rather than a sky property. Even if real, the correlation is direction-ambiguous, since fallout contamination predicts it as well as glinting objects. Mixed, leaning fragile.

Within the working catalog the association is robustly reproducible — RR 1.45, p = .011, permutation p = .006, surviving weather and lunar controls (TRN-E010), persisting after ML artifact-scoring (TRN-E015) and in an attenuated Bayesian model with temporal autocorrelation (TRN-E019) — and TRN-E017 usefully closes one environmental confounder. It nonetheless fails the two tests that matter most: normalizing by observation days makes it insignificant and largely schedule-determined (TRN-E011), and an independently extracted catalog with per-night sky coverage as exposure gives IRR 1.03, p = .71 (TRN-E009). The strongest opposing consideration is TRN-E009's own caveat — its preliminary, noise-dominated catalog (~7,250 candidates/night) could dilute a weak real effect below detection — which is why this is mixed rather than contradicted. Even if real, the correlation is direction-ambiguous: fallout on emulsions predicts it as well as glints do.

Same-rows reanalyses keep passing (TRN-E010, TRN-E015; attenuated in TRN-E019) while independent extraction with schedule modelling stays null (TRN-E009, TRN-E011). Repetition on the same rows is not independent evidence, so the null result on the varied step carries disproportionate weight.

Same-rows reanalyses keep passing (TRN-E010, TRN-E015; attenuated in TRN-E019) while independent extraction with schedule modelling stays null (TRN-E009, TRN-E011). The split points at catalog properties, not sky events.

Same-rows reanalyses keep passing (TRN-E010, TRN-E015; attenuated in TRN-E019) while independent extraction with schedule modelling stays null (TRN-E009, TRN-E011). The split points at catalog properties rather than sky events; repeated reanalysis of the same rows adds no independent weight.

Same-rows reanalyses keep passing (TRN-E010, TRN-E015, attenuated in TRN-E019) while independent extraction with schedule modelling stays null (TRN-E009, TRN-E011). Repeated analyses of the same rows are not independent evidence; the split points at catalog properties.

Same-rows reanalyses keep passing (TRN-E010, TRN-E015; attenuated in TRN-E019) while independent extraction with schedule modelling stays null (TRN-E009, TRN-E011). Repeated analysis of the same rows is not independent evidence; the split points at catalog properties.

Same-rows reanalyses keep passing (TRN-E010, TRN-E015, attenuated in TRN-E019) while independent extraction with schedule modelling stays null (TRN-E009, TRN-E011). A correlation that survives only in one catalog lineage looks like a property of the catalog, not the sky.

Same-rows reanalyses keep passing (TRN-E010, TRN-E015, attenuated in TRN-E019) while independent extraction with schedule modelling stays null (TRN-E009, TRN-E011). The split points at catalog properties, not sky events.

Same-rows reanalyses keep passing (TRN-E010, TRN-E015, attenuated in TRN-E019) while independent extraction with schedule modelling stays null (TRN-E009, TRN-E011). That split points at catalog properties, not sky events.

The nuclear-test timing correlation successfully replicates when recalculating the original working catalog. Yet it vanishes entirely into statistical insignificance when an independent pipeline extracts the data and properly models the nightly survey coverage and observing schedule, suggesting the correlation is a scheduling and extraction artifact rather than a physical phenomenon.

Analyses using the original working catalog (Bruehl & Villarroel; Doherty; Cann; Sinkkoinen) consistently find elevated transient rates near atmospheric nuclear tests under various models, and this persists after ML artifact scoring and some environmental controls, indicating a genuine association in that dataset. Yet Hayes’s independent catalog, with explicit modeling of nightly sky coverage, finds no significant effect (IRR ≈1.03), and Watters et al. argue that much of the original correlation is driven by the observation schedule. Thus the correlation appears robust within the original catalog lineage but does not generalize when source extraction and exposure modeling change. The strongest opposing consideration is this disappearance under independent catalog construction, which strongly suggests at least some of the original effect is a property of catalog definition and scheduling rather than of the sky.

Independent statistics on the original catalog reproduce a relative risk near 1.45 with permutation support, and the association persists after ML artifact-scoring. The one fully independent extraction, with nightly sky coverage as exposure, is null (IRR 1.03, p = .71), and Watters traces the original signal largely to the observing schedule. Even a surviving correlation would be direction-ambiguous, because radioactive fallout on plates predicts it as well as glints do. The strongest opposing consideration is Hayes: when the disputed source-extraction step is actually varied, the association vanishes.

Every replication type points a different way: same-catalog statistics reproduce the effect (TRN-E010: RR 1.45, permutation p = .006), it persists under ML artifact-scoring (TRN-E015) and geomagnetic control (TRN-E017), and a public Bayesian remodel finds it attenuated but present (TRN-E019) — while the one fully independent extraction with nightly coverage modeled finds nothing (TRN-E009: IRR 1.03, p = .71) and Watters et al. trace the signal to the observing schedule. A correlation that lives only inside one team's working sample looks like a property of the sample; and even if real it is direction-ambiguous, since fallout settling on plates predicts it as well as glints do. The strongest opposing consideration is dilution: Hayes's noise-dominated independent catalog (~7,250 candidates per night) could wash out a weak true effect, which is why this is mixed rather than contradicted.

The split is by replication type and it is instructive: independent code on the author-supplied rows reproduces RR 1.45 with permutation p = .006 and survives weather and lunar controls (TRN-E010), it persists after ML artifact scoring (TRN-E015), and TRN-E017 closes the geomagnetic-sensitivity confounder — but the one independent extraction with per-night sky coverage as exposure is flatly null (TRN-E009, IRR 1.03, p = .71) and TRN-E011 attributes the original signal to the Palomar observing schedule. Even the local statement is contested, since "elevated rates" presupposes a denominator that is exactly what is disputed. The strongest opposing consideration to grading this lower is that Hayes's catalog averages ~7,250 mostly-noise candidates per night, which could dilute a real weak effect below detection. Note also that fallout contamination of emulsions predicts this correlation as well as any glint population does.

Same-rows reanalyses keep passing (TRN-E010, TRN-E015; attenuated in TRN-E019) while independent extraction with schedule modelling stays null (TRN-E009, TRN-E011). That split points at catalog properties rather than sky events.

Same-rows reanalyses keep passing (TRN-E010, TRN-E015, attenuated in TRN-E019) while independent extraction with schedule modelling stays null (TRN-E009, TRN-E011). The split points at catalog properties, not sky events.

Splits cleanly by replication type: independent statistics on the original catalog reproduce it (Doherty: RR 1.45, p = .011; permutation p = .006), and it persists after ML artifact-scoring (Bruehl et al. 2026) and Kp control (Cann, TRN-C103) — but the one fully independent extraction (Hayes 2026) finds nothing once nightly survey coverage is modeled (IRR 1.03, p = .71), and Watters et al. attribute the original signal to the observation schedule. Replicates within the catalog lineage; fails across it.

While an elevated candidate rate within one day of atmospheric nuclear tests exists in the original working catalog, an independent pipeline extraction across 368 POSS-I nights found no statistically significant correlation (p = 0.17). Furthermore, negative binomial modeling incorporating nightly plate coverage and scheduling controls shows an incidence rate ratio of 1.03 (p = 0.71), demonstrating that the effect is an artifact of Palomar observing schedules. The strongest opposing consideration is that Doherty's reanalysis of the author-supplied rows survived permutation tests and controls for precipitation and lunar illumination.

The correlation replicates under independent code on the original rows (RR 1.45, p = .011, permutation p = .006, IRR up to 3.98 sunlit-only; TRN-E010), persists under ML artifact scoring (TRN-E015), and one environmental confounder was measured and closed (TRN-E017), with a Bayesian reanalysis reporting survival at one-day lag (TRN-E019). Against it, independent extraction with exposure modeling yields a null (TRN-E009) and Watters finds the effect almost completely determined by the observing schedule (TRN-E011). Strongest opposing consideration: TRN-E009's null, though its catalog averages ~7,250 mostly-noise candidates per night, so dilution could mask a weak real effect.

TRN-E010 reproduces an approximately 1.45 relative risk on the original catalog, and TRN-E015 and TRN-E019 report persistence under artifact scoring or temporal modelling. In contrast, TRN-E009 finds no significant effect in an independently extracted catalog once searchable coverage is modelled, and TRN-E011 attributes the result largely to the observing schedule. The strongest positive consideration is repeated recovery from the supplied rows; the strongest opposing consideration is failure under independent extraction and exposure normalization. The evidence therefore supports a catalog-level association but not a robust transient-rate effect.

Nuclear-window relative risk ~1.45 reproduces on the author catalog (Doherty, permutation and negative-binomial controls) and persists under ML artifact scoring, yet Hayes’ independent extraction with nightly exposure modeling yields null IRRs and Watters attributes the signal almost entirely to the Palomar observing schedule. A correlation that is catalog-dependent is not yet an established sky effect. Strongest opposing consideration: observation-day normalization and independent source extraction remove the signal, implying a property of the working rows rather than of the sky.

The correlation replicates as stated on the authors' rows with independently written code, survives weather and lunar controls and a 10,000-iteration permutation test (TRN-E010), persists after ML artifact scoring (TRN-E015), and survives an autocorrelation-aware Bayesian remodelling in attenuated form (TRN-E019). Against that, it becomes insignificant when the candidate list is independently extracted and nightly patch coverage is modelled as exposure (TRN-E009: IRR 1.03, CI 0.89–1.18), and TRN-E011 attributes it almost entirely to the Palomar observing schedule. The strongest consideration on the proponents' side is TRN-E009's own caveat — ~7,250 mostly-noise candidates per night could dilute a weak real effect below detection — which is why this is mixed rather than contradicted, and why TRN-R008 is the right next step.

Supporting on the author-supplied catalog: TRN-E010 (RR 1.45, p=.011; survives weather/lunar controls and permutation), TRN-E015 (persists after ML artifact control), TRN-E017 (one environmental confounder closed), TRN-E019 (survives a Bayesian autocorrelation model, attenuated). Undermining: TRN-E009 (strong — independent extraction with exposure modelling gives RR 1.35, CI 0.91–2.00, IRR 1.03), TRN-E011 (strong — insignificant after observation-day normalization; schedule-driven). Qualifying: TRN-E012. The catalog claim TRN-C125 (anchor only) reports 203 of the high-probability nuclear-window candidates come from a single night, concentrating the effect in a handful of uninspected plates.

Supporting on the same rows: TRN-E010 (moderate; RR 1.45, controls, permutation), TRN-E015 (moderate; strengthens under ML filter), TRN-E017 (moderate; sensitivity invariant to Kp). Undermining: TRN-E009 (strong; independent extraction with exposure modelling gives IRR 1.03), TRN-E011 (strong; schedule-driven). Qualifying: TRN-E019 (attenuated; UAP lag stronger), TRN-E012. TRN-E016 context. Reproduces on the catalog, fails when the catalog is rebuilt.

While the relative risk of 1.45 can be reproduced statistically on the original unvetted working catalog, the association vanishes entirely under negative binomial exposure modeling on an independently extracted catalog (IRR = 1.03, p = 0.71). Critical audits demonstrate the correlation is largely an artifact of Palomar's observing calendar coinciding with testing schedules. Strongest opposing consideration: The correlation persists across permutation testing and within quality-filtered subsets scored as real by machine-learning models.

The correlation replicates on the original catalog with independent code, surviving negative-binomial regression with weather controls and permutation (TRN-E010), persists among ML-scored likelier-real candidates (TRN-E015), and survives a Bayesian autocorrelation treatment in attenuated form (TRN-E019). But under fully independent source extraction with exposure modeling it is null and schedule-sensitive (TRN-E009), and the skeptical reanalysis finds it almost completely determined by the Palomar observing schedule (TRN-E011). A correlation that survives reanalysis of the same rows but dies under independent extraction is the definition of a contested result — mixed.

TRN-E010 independently reproduces an approximately 1.45 relative risk on the supplied working catalog, and TRN-E015 and TRN-E019 report persistence under alternative internal filters or models. This supports the narrowly stated fact that the working catalog's candidate rates are elevated in the specified temporal window. The strongest opposing consideration is that TRN-E009 finds no significant effect after independent extraction and exposure modeling, while TRN-E011 attributes the result largely to the Palomar observing schedule.

Relative risk ≈1.45 reproduces on the original catalog under independent code and survives some controls (E010, E015, E019); it is insignificant under independent extraction with exposure modeling (E009) and is almost entirely schedule-driven once observation-day normalization is applied (E011). Strongest opposing consideration is the clean disappearance of the signal once the two main confound channels (extraction and exposure) are varied.

On the working rows the effect reproduces (RR 1.45, permutation p=.006; E010). It persists after ML scoring (E015) and survives, attenuated, when temporal autocorrelation is modelled (p=0.037; E019). But the two tests that vary the disputed steps are null. Normalizing by observation days ties the effect to the observing schedule (E011). An independent catalog with nightly coverage as exposure gives IRR 1.03 (E009). Strongest opposing consideration: that catalog is mostly noise (~7,250 candidates per night), and Watters used related datasets, so a weak real effect could be diluted. Hence mixed, not contradicted (TRN-R008).

It reproduces on the original rows with independent code, controls and a permutation test (TRN-E010). It persists among likelier-real candidates (TRN-E015) and survives an autocorrelated Bayesian model, attenuated (TRN-E019); TRN-E017 closes one confounder. But the only independently extracted catalog finds a relative risk of 1.35 with an interval spanning 1, and an exposure-weighted IRR of 1.03 (TRN-E009). TRN-E011 ties the timing to the observing schedule. TRN-E012 disputes the critique's power.