Kuiper Belt / TNO Optical Anomaly Hunt

Premise

The project was designed as a data-driven archival screen for unusual brightness-distance behaviour in outer Solar System small bodies. The question is not interpretive; it is statistical. Across many epochs, does an object's flux scale with both heliocentric and observer distance (the reflected-sunlight expectation) or with observer distance alone (an intrinsic-like term)? The second signature should not exist in a quiet population.

The goal was not to interpret any candidate in a speculative framework. The goal was to test whether a subset of objects show optical flux scaling inconsistent with a simple reflected-light expectation, and robust enough to survive common photometric failure modes.

Method

For each observation, magnitude was converted to a flux proxy and analysed in log space. Two competing models were fit:

Reflected-light: ln F = c − 2 ln r − 2 ln Δ + ε   (brightness scales with both Sun and observer distance)
Intrinsic-like: ln F = k − 2 ln Δ + ε   (brightness scales only with observer distance)

To reduce systematics the pipeline applied nightly binning in flux space; per-filter analysis rather than mixed-band fitting; minimum usable epoch and leverage cuts (N ≥ 20, span(ln r·Δ) ≥ 0.35); robust fitting and bootstrap checks; adversarial trims of the brightest and faintest points; observatory and temporal-split checks; and an internal replication criterion requiring consistency across ≥2 observatory subsets. Candidates were ranked primarily by ΔBIC = BIC(reflected) − BIC(intrinsic) — positive favours intrinsic — and cross-validated predictive error.

Sample results

The final ranked run evaluated 121 object/filter candidates. One met the credible threshold with all robustness checks passing; twenty-five showed mild preference but failed one or more checks; ninety-five returned nulls. That distribution is itself a result. It means the adversarial and replication layers are doing real work — amplifying the evidence where it is real and killing it where it is not.

582301 (2015 RM306) — primary candidate

In the G-band, 582301 returns ΔBIC = 16.588 in favour of the intrinsic-like model across fifty nightly-binned epochs, with a leverage span of ln(r·Δ) = 0.995 — enough geometric coverage to distinguish the two models meaningfully. Cross-validated predictive error favours intrinsic (CV advantage = 0.064). Adversarial trim checks and replication across observatory subsets both pass. A phase-controlled reflected model — the natural first alternative — does not explain the signal away (ΔBIC against intrinsic increases to 17.689). A reflected-bootstrap null test returns p(ΔBIC ≥ observed) = 0.0025.

Temporal subset consistency is where the signal becomes fragile. Splitting the G-band into time thirds yields ΔBIC values of 3.906, −1.405, and −0.174. The sign inversion in thirds two and three is the pipeline's own dissent. The aggregate signal is real; its behaviour over time is not clean.

Multi-wavelength follow-up of 582301

Track-based metadata and catalogue checks against GALEX, AllWISE, and NEOWISE returned no time- and position-consistent detections. The NEOWISER row-vs-track validation returned 0 / 13 within an arcsecond; minimum separation from any candidate row was 1,949″. A frame-level preflight workflow against the NEOWISE IBE service converged on the same conclusion: frames exist near the projected path under loose windows, but no source rows are simultaneously time-consistent and position-consistent with 582301.

Radio follow-up along the Horizons sky track in NVSS (1.4 GHz), TGSS ADR1 (150 MHz), and VLASS (2–4 GHz) yielded one nearby NVSS source (NVSS 081927+153638, 6.2 ± 0.5 mJy) at an angular separation of 30.93″ — outside the 15″ close-match threshold and treated as a likely unrelated background source. No radio survey showed a track-coincident match.

Secondary cases

Two weaker candidates appear below 582301 in the ranking. 523676 (2013 UL10) in G-band shows ΔBIC = 4.659 across twenty-seven epochs and survives replication but fails adversarial trimming — the signal is not robust to removing the brightest and faintest points. 44594 (1999 OX3) in R-band shows ΔBIC = 1.213 across seventy-nine epochs with a slightly negative cross-validation advantage. Both verdicts are the same: inconclusive, mixed follow-up.

Log

> entry, 2026.iv

Final ranked screening pass issued. 582301 holds the only credible-P slot. Multi-wavelength follow-up in UV, IR, and radio archives returns negative or non-diagnostic across the board. The temporal-thirds instability is documented and not explained.

> entry, 2026.ii

Adversarial layer added — brightest/faintest-point trims at 10%, observatory and temporal splits. Twenty-five mild candidates re-classify as P-weak. Two pre-existing apparent candidates demoted.

> entry, 2025.xi

Pipeline opened. Working sample assembled from MPC optical archives; JPL Horizons geometry attached to every epoch; SBDB metadata joined for orbit context. First pass evaluates 121 object/filter candidates.

Limitations

The optical sample is archival and heterogeneous, even after hard filtering. The reflected-light model is intentionally simple; real surface and phase behaviour can be more complicated. Public UV/IR/radio catalogues are not the same thing as moving-target image-level reprocessing. And physical parameters for the strongest candidates remain sparse in SBDB and related archives. These limitations do not negate the main ranking result; they do bound how far it can be interpreted.

The working console

This report is the bound publication; Artifact-SETI maintains a live console with the candidate ranking, ΔBIC distribution across 121 candidates, temporal-thirds test for 582301, gating flow from raw photometry to verdict, and the negative multi-wavelength sweep.