SIGNALTESTNET

Citable, permanent evidence

The dataset, archived off-network

The live chain is the real evidence -- this page exists so it can also be cited. Every confirmed block and the hit that cleared four deterministic rejection rules to produce it, deposited periodically on Zenodo with a permanent DOI -- automated publishing is currently paused, see below. Once published, a given version can never be silently altered -- that's a feature for citation, not a limitation.

Deposits published before 2026-08-26 predate a protocol correction and may include invalid blocks. An audit found 120 of the network's first 263 blocks were paid on a miscalibrated filter (reference-position detections, not primary targets) -- those blocks are present in any snapshot version covering that range. See /audit for the full record and the exact 120 affected. Those deposits live under a separate, now-frozen record (10.5281/zenodo.22084643, final version explains the cut) -- the record below is a NEW one, started 2026-08-27, created specifically so its version history would be post-correction only.

Open, since 2026-08-28: the record below does not yet reflect the current chain. A second, unrelated consensus reset (a mining-loop race, not a filter miscalibration -- see /audit §10.3) means every version currently published under this record predates it. The automated periodic snapshot is paused rather than silently resuming under the same DOI on the new chain. For the network's real current state, use the live pages (/dashboard, /explorer) until a corrected version is published here.

Latest snapshot

Pulled live from Zenodo's own record for this deposit -- never hand-typed here, so it can't go stale.

Loading the latest deposit record from Zenodo…

Source data

Every hit in this dataset comes from the same corpus described in /audit: a public turboSETI hits table matching Enriquez et al. 2017 (692 stars, L band) -- confirmed independently by an exact match on both the corpus-wide drift-rate search cutoff and the count of distinct primary-target sources. The raw filterbank/HDF5 observation files aren't part of this corpus at all; see /audit's appendix for that boundary. Reproducing any check below requires this same source table, not just the files in a Zenodo deposit.

How to independently verify a row

No step here requires trusting Galaxy SIGNAL -- each one is checkable against a source this project doesn't control.

  1. Check the four rejection rules. Using the same public source table (above) and the row's own status/obs_run/frequency/drift-rate fields, anyone can recompute all four deterministic checks (whitepaper section 4.3) independently -- the actual triage, as opposed to the declared score below.
  2. Re-score it, matched to its own classifier_version. Load the rfi_classifier.joblib whose fingerprint matches the row's own classifier_version field -- not necessarily the single model file bundled with a deposit, if that deposit spans more than one model version (this project has retrained the classifier more than once; a mismatched model will not reproduce every row). Feed it the row's own frequency/drift-rate/SNR/known-band flag and confirm it reproduces that row's declared verification_score -- this checks the number is reproducible, not that the network verified it itself; see below.
  3. Check the block. block_hash is independently verifiable against the live network's own /chain endpoint, or any full node's own copy -- see the whitepaper for the block-hash construction.
  4. Check the hit is real. hit_file_pos is the hit's exact row position in the source table (above) -- anyone with a copy of it can confirm the row genuinely exists, not fabricated.

What "verified" does and doesn't mean here. Every check above is verifiable against the published source table -- but that table itself is the ground truth these checks assume, not something they can confirm. The raw .fil/.h5 observation files it was exported from aren't part of this network's data at all; no node checks it against them. This is mitigated, not eliminated, by the table's own public provenance -- anyone can independently re-derive it from Breakthrough Listen's own published archive.

What this is not

Not a discovery claim. This archives the network's actual, checkable output -- which hits cleared the four deterministic rejection rules, when, and under which exact classifier model was declared alongside them -- as an immutable, timestamped record. It is not a claim that any of these detections is a technosignature; see the whitepaper's own "what this is not" section for why.