METHODS
Transparent, reproducible analysis
Every analysis step is versioned, documented, and open for independent replication. Conventional baselines are never hidden by experimental methods — they remain visible for direct comparison.
CORE PIPELINE
From data registration to evidence record
Discover and register datasets
Identify authoritative archives, APIs, catalogs, event streams, and literature. Record rights, licensing, citation requirements, coverage, and provenance before any data is accessed.
Normalize coordinates, time, units, and metadata
Align all observations to common coordinate frames (ICRS, Galactic), time standards (UTC, TDB), units, calibration versions, and spatial/spectral/temporal resolution. Follow IVOA standards where applicable.
Assess data quality and selection effects
Measure coverage completeness, sensitivity limits, observational biases, and instrumental artifacts. Absence of a detection is not negative evidence without a sensitivity estimate.
Cross-match sources and uncertainty regions
Spatially and temporally cross-match observations across instruments, messengers, and epochs. Propagate localization uncertainty through every match.
Reject or label known explanations
Flag terrestrial interference, instrumental artifacts, duplication, and known natural sources (pulsars, magnetars, AGN, satellite RFI, etc.) before anomaly detection begins.
Run conventional anomaly detection
Apply domain-specific baselines: matched filters, periodicity searches, spectral line detection, transient classification, and catalog cross-matching. These baselines remain visible and are never hidden by experimental methods.
Run Entropica1 RSCS sequenceEntropica1 RSCS
Apply the pattern-before-data sequence in fixed order: Entropy Drop Zones → Symbolic Gradient Fields → Ensemble-SINDy. Results are compared against conventional baselines, not substituted for them.
Compare, ablate, and test on held-out data
Measure agreement between methods, perform ablation tests, estimate uncertainty, and validate on held-out sky regions or epochs. Injection/recovery tests quantify false-positive and false-negative rates.
Create or update an evidence record
Write a versioned evidence record. Never silently overwrite earlier results. Every revision is logged with timestamp, analyst, and change description.
Route significant candidates to human review
Candidates above a configurable significance threshold are flagged for human and independent review before any public communication.
ENTROPICA1 RSCS
Reflexive Symbolic Cognition Sequence
The RSCS sequence is applied in a fixed order. Each stage builds on the previous. Results are experimental and must be compared against conventional baselines — pattern or model confidence does not prove intelligence.
EDZ
Entropy Drop Zones
Localize unexpectedly ordered or low-entropy regions and events in the data. An entropy drop indicates structure that is more regular than the surrounding noise or background — a necessary but not sufficient condition for further investigation.
Limitations
Entropy drops can arise from instrumental artifacts, data compression, calibration errors, and natural periodic sources. EDZ results require interference and artifact testing before interpretation.
SGF
Symbolic Gradient Fields
Map the topology, direction, drift, sources, sinks, and transitions of symbolic patterns across the data. SGF identifies how structure propagates and evolves — not just where it exists.
Limitations
Gradient fields are sensitive to the choice of symbolic representation and window size. Results must be compared against conventional spectral and temporal analysis.
SINDy
Ensemble-SINDy
Infer compact, interpretable candidate rules from the data and test whether they generalize beyond the discovery sample. Ensemble methods reduce overfitting; held-out evaluation tests generalization.
Limitations
SINDy-derived rules describe statistical patterns, not physical mechanisms. A rule that generalizes is interesting — it is not evidence of intelligence.
SCIENTIFIC GUARDRAILS
What our methods do not claim
A highly unusual event is not automatically a technosignature.
Missing coverage is not negative evidence.
Correlation is not causation.
Model confidence is not physical probability.
Pattern recognition is not proof of intent.
"Unexplained" does not mean "extraterrestrial."
A single detection without replication is a candidate, not a finding.
Automated scores never produce "confirmed extraterrestrial intelligence."