Research

Methods

How a persistent signal is separated from an enormous volume of noise, and held to a standard of evidence.

The Laboratories record far more than they can examine. The work of method is to decide, rigorously, what is worth examining -- and to resist, at every step, the explanation the data will not bear.

The pipeline

Record

continuous, archived

Reduce

candidate lines

Test

false-alarm, coincidence

Review

human, by behavior

Catalog

or reject

Fig. 1 From recording to catalog

Every recording is archived in full and reduced to a set of candidate detections. Each candidate is tested against a false-alarm standard and, where separated stations are available, for coincidence. What survives is reviewed by a person and classified by its behavior. What does not survive is rejected -- but the raw recording is kept, because the analysis may improve.

candidate_score = 0.32*P_persist + 0.28*C_coincident + 0.18*S_structure
                + 0.12*Q_calibration - 0.10*N_known_source

P_persist     = recurrence_count / reviewed_observing_windows
C_coincident  = matched_station_count * timing_confidence
S_structure   = spectral_entropy_drop + phase_stability + residual_shape
Q_calibration = 1 - abs(clock_drift_ns / allowed_drift_ns)

public_review_threshold = 0.74
class_III_review        = candidate_score >= threshold AND N_known_source < 0.12

Fig. 1a Candidate scoring sketch -- Simplified public notation. Internal weights vary by instrument, campaign, and access class.

Source: Methods office public training notation
As of: 2026 methods review
Method: Variables are public descriptions; operational coefficients are not published

The standard of evidence

A signal enters the catalog described by what it does -- its frequency, its recurrence, its coincidence across stations -- and never by any theory of where it comes from. Cause is a conclusion, not a starting point. The Laboratories' false-alarm framework governs what does and does not enter the record; it is deliberately conservative.

Fig. 2 Detection and rejection -- most candidate lines are interference and are rejected; a persistent line that survives every test is retained.

Source: FAL public training reconstruction
As of: 2026 methods review
Method: Representative bins derived from released review categories, not raw restricted signal data

Coincidence

The strongest evidence the Laboratories can offer for a genuinely anomalous detection is coincidence: the same signal, in the same instant, at stations far enough apart that no local fault could produce it. Coincident detections that survive every test for chance and cross-talk are catalogued as Class III, unresolved, described precisely and kept under study.

Input	Question asked
Clock agreement	Did the station clocks agree within the relevant timing window?
Instrument state	Were receivers, clocks, filters, and gain states within baseline?
Known-source rejection	Could a transmitter, weather event, vehicle, or maintenance action explain it?
Spatial residual	Does the signal leave the same shape after station-specific noise is removed?
Repeatability window	Does the behavior recur under comparable observing conditions?

Fig. 3 Coincidence review inputs -- A detection must survive boring explanations before it receives interesting language.

References & notes

1.FAL Technical Report 22-011, A false-alarm-rate framework for long-baseline persistent-signal detection.
2.FAL Technical Report 04-009, Coincidence analysis across separated stations.