technosig_lab

Reproducible SETI technosignature benchmark and vetting harness.

This project is not a claim detector. It is a small research tool for measuring when a narrowband drifting-signal detector works, when it fails, and when a real-data event should be downgraded instead of over-interpreted.

It generates controlled artificial signals, runs them through simplified channel/background models, calibrates drift-search thresholds, and logs Pd/Pfa style results as JSONL. It can also read reduced Breakthrough Listen .h5 / .fil waterfall data through blimpy when that optional dependency is installed.

Useful public framing:

Technosig Lab is a reproducible benchmark and vetting harness for narrowband
drifting-signal detection in synthetic and real Breakthrough Listen waterfall
backgrounds.

Install

pip install -r requirements.txt

Only numpy is required for synthetic experiments. matplotlib is needed for plots. blimpy and h5py are needed for real Breakthrough Listen files.

Repository Scope

This repository should contain code, tests, docs, small examples, and metadata. It should not contain large Breakthrough Listen data files or generated result directories. See docs/DATA.md for the data policy and docs/CANDIDATE_GATES.md for the candidate gate definitions.

Reproducibility wrappers live under scripts/. They write local data to ignored data/ and results/ directories.

Quick Synthetic Run

python -m technosig_lab.cli thresholds-synthetic --out results/thresholds_synthetic.jsonl
python -m technosig_lab.cli synthetic-pd --out results/synthetic_pd.jsonl
python -m technosig_lab.cli plot --input results/thresholds_synthetic.jsonl results/synthetic_pd.jsonl --out figures

Reproducibility Scripts

Run from the repository root:

pwsh scripts/fetch_voyager_positive_control.ps1 -DownloadData
pwsh scripts/run_voyager_positive_control.ps1
pwsh scripts/fetch_bldr1_sband_index.ps1
pwsh scripts/run_hip194_vetting.ps1

run_hip194_vetting.ps1 expects the large HIP194 HDF5 file to already exist locally under data/bl_sband/; it does not download that file automatically.

Real BL Waterfall Run

python -m technosig_lab.cli real-candidates ^
  --input path\to\file.h5 ^
  --max-frames 64 ^
  --max-bins 2048 ^
  --chunk-bins 256 ^
  --max-chunks 16 ^
  --peak-threshold-z 6 ^
  --out results/real_candidates.jsonl

python -m technosig_lab.cli real-thresholds ^
  --input path\to\file.h5 ^
  --max-frames 64 ^
  --max-bins 2048 ^
  --chunk-bins 256 ^
  --max-chunks 16 ^
  --out results/real_thresholds.jsonl

python -m technosig_lab.cli real-injection ^
  --input path\to\file.h5 ^
  --max-frames 64 ^
  --max-bins 2048 ^
  --chunk-bins 256 ^
  --max-chunks 16 ^
  --out results/real_injection.jsonl

Voyager Positive Control

Voyager should be treated as a positive control, but not by blindly looking for 8.4 GHz inside a locally cut waterfall. Reduced data can use local or relative frequency axes. The machine criterion here is morphology-first:

python -m technosig_lab.cli real-candidates ^
  --input path\to\voyager_file.h5 ^
  --positive-control ^
  --positive-control-top-n 5 ^
  --max-frames 64 ^
  --max-bins 2048 ^
  --chunk-bins 256 ^
  --max-chunks 16 ^
  --peak-threshold-z 6 ^
  --out results/voyager_candidates.jsonl

PASS for this command means a narrowband_drifting cluster appears in the top-N candidate clusters. That is only a shape check. A real claim still needs drift-search scoring, header frequency checks, ON/OFF cadence, and RFI vetting.

Recorded Voyager Control Results

Machine-readable benchmark entry:

• Catalog: known_signals_catalog.json
• Dataset id: voyager1_bl_gbt_xband
• Interpretation: known artificial carrier with Voyager-associated secondary narrowband-drifting responses

Positive control:

• Input: data/voyager1_bl/voyager_f1032192_t300_v2.fil
• Carrier window: 8420.2163-8420.2166 MHz
• Result file: results/voyager_candidates_carrier_run.jsonl
• Result: PASS
• Morphology: narrowband_drifting
• Peaks: 6
• z_max: 38.77
• z_mean: 35.03
• Morphology score: 514.85
• Estimated drift: -0.107 Hz/s

Independent drift-search validation:

• Search window: 8420.18-8420.25 MHz
• Result file: results/voyager_thresholds_8420_18_25_run.jsonl
• Top score: 85.42
• P99 threshold: 31.77
• Estimated drift: -0.10 Hz/s

Negative/off-target control:

• Candidate window: 8420.3000-8420.3003 MHz
• Result file: results/voyager_candidates_offtarget_run.jsonl
• Result: FAIL as expected, no narrowband_drifting cluster
• Off-target drift-search window: 8420.30-8420.37 MHz
• Result file: results/voyager_thresholds_offtarget_run.jsonl
• Top score: 4.66
• P99 threshold: 4.59
• Interpretation: marginal near-threshold drift-search score without a corresponding morphology cluster

Conclusion: the stand recovers a known artificial narrowband drifting carrier from real filterbank data and does not label a nearby off-target window as the same positive control. This validates the stand as a real-data positive-control detector, not merely a synthetic-signal detector. It does not yet prove blind survey performance, calibrated false-alarm behavior across observing conditions, or robust RFI rejection.

Wide carrier-neighborhood scan:

• Search window: 8420.00-8420.60 MHz
• Result file: results/voyager_thresholds_wide_8420_00_60_absolute.jsonl
• Chunks: 826
• Top score: 93.86
• P99 threshold: 5.13
• Top estimated frequency: 8420.216449 MHz
• Top estimated drift: -0.10 Hz/s

Conclusion from the wider scan: the dominant score in the broader Voyager neighborhood lands at the tutorial carrier frequency, rather than appearing uniformly across nearby spectrum.

Interpretation of secondary Voyager responses

The wide-window drift-search produced several secondary responses near the main Voyager carrier neighborhood. Targeted morphology review showed that these secondary windows are not empty off-target regions.

8420.23895-8420.23925 MHz:
  morphology: narrowband_drifting
  peaks: 4
  clusters: 1
  z_max: 13.20
  score: 86.95
  drift: -0.107 Hz/s

8420.19395-8420.19425 MHz:
  morphology: narrowband_drifting
  peaks: 5
  clusters: 1
  z_max: 10.17
  score: 109.62
  drift: -0.094 Hz/s

8420.19365-8420.19395 MHz:
  morphology: narrowband_drifting
  peaks: 5
  clusters: 1
  z_max: 12.00
  score: 115.67
  drift: -0.094 Hz/s

These responses are not classified as independent blind SETI candidates. They are recorded as Voyager-associated sideband/neighborhood responses because they appear in the same Voyager observation context and show drift values consistent with the main carrier response.

This distinction matters:

main carrier:
  strongest wide-search response
  matches known Voyager carrier neighborhood

secondary responses:
  morphology PASS
  drift-consistent with the Voyager response family
  require separate instrumental/source interpretation

off-target control:
  morphology FAIL
  no peaks
  no clusters

The detector therefore distinguishes an empty off-target window from structured Voyager-associated secondary narrowband-drifting responses.

Frequency separation analysis

The top wide-search responses form a structured frequency neighborhood around the main carrier. The table below combines wide-search estimated frequencies with targeted morphology-review drift estimates for the secondary rows.

frequency_mhz	delta_from_main_khz	morphology	drift_hz_s
8420.216449386	0.000	main carrier	-0.100
8420.239107666	+22.658	narrowband_drifting	-0.107
8420.194108997	-22.340	narrowband_drifting	-0.094
8420.193791106	-22.658	narrowband_drifting	-0.094

The symmetric pair at +22.658 kHz and -22.658 kHz is recorded as a Voyager-associated secondary response pattern, not as multiple independent signals.

The comparison JSON is generated with:

python -m technosig_lab.cli compare-candidates ^
  --input results\voyager_thresholds_wide_8420_00_60_absolute.jsonl ^
  --reference-f0-mhz 8420.216449386 ^
  --out results\voyager_candidate_frequency_offsets.json

The generated JSON stores the wide-search drift estimates from results/voyager_thresholds_wide_8420_00_60_absolute.jsonl.

BLDR1 S-band HIP194 Initial Evidence

hip194_sband_2375p931_bldr1_event is tracked as an initial BLDR1 S-band event-table candidate, not as a known positive control.

• File: data/bl_sband/spliced_blc2021222324252627_guppi_57992_30093_HIP194_0014.gpuspec.0000.h5
• Header summary: data/bl_sband/HIP194_2375p931_HEADER_SUMMARY.json
• Evidence report: results/hip194_sband_candidate_evidence.json
• Catalog status: initial_analysis_complete
• Classification: low_strength_event_table_hit
• Candidate strength: low

Header checks:

frequency range: 1818.457034-2720.800781 MHz
contains event frequency 2375.931298 MHz: true
contains requested 2380.000000 MHz: true

Gate report:

score_p99: weak_pass
quantile_stability: fail_borderline_tail
drift: fail_static
static_rejection: fail_static_supported
morphology: fail_no_narrowband_drifting
neighbor_window: pass_locally_distinct
positive_control: fail_not_known_positive

Quantile and static-drift evidence:

score: 48.132
p99 margin: 1.598
p999 margin: 0.160
p9999 margin: 0.016

score/p99: 1.034
score/p999: 1.003
score/p9999: 1.0003

nonstatic rerun, abs(drift) >= 0.5 Hz/s:
  nonstatic score: 7.688
  score drop: 40.443
  top nonstatic drift: -0.55 Hz/s

Neighbor-window vetting:

event rank among five neighbor windows: 1
neighbor max score: 4.952
gate: pass_locally_distinct

The event window has a weak local score excess, but the top drift is static/near-static and the morphology gate does not recover a narrowband_drifting cluster. It is therefore recorded as initial detector evidence only.