–
hits verified (blocks mined)
The verification work, explained
What every mined block actually is
Galaxy SIGNAL doesn't hash arbitrary numbers. Each block starts from a real radio-astronomy detection (a "hit") from Breakthrough Listen's public L-band survey, and the network's Proof-of-Useful-Verification (PoUV) is four deterministic RFI-vs-candidate checks, re-verified by every node -- the same kind of decision the original survey's own pipeline had to make -- plus a declared confidence score, published as metadata on every block but never independently verified, with no effect on what that block pays.
The network's work, so far
Every confirmed block is one hit that cleared the four rejection rules above. This updates live from the chain itself.
–
distinct target stars
–
classifier versions seen on-chain
Out of ~28.86 million raw hits in the source dataset -- only those that clear four deterministic rejection rules (see below) and a declared score threshold ever become a block. See the explorer for the full, browsable list, including sky position (RA/DEC) for each.
Where every mined hit actually falls, live
This chart is not a fixed example -- it's every block the network has ever mined, replotted from the live chain on every page load. It grows on its own as more blocks come in. The shaded bands mark frequency ranges known to be crowded with terrestrial transmitters -- the eleven bands the classifier's known-band feature checks. That feature barely moves the model's score either way (see below), because roughly 99% of this corpus falls inside one of these bands -- so falling "outside" them is the residual, expected case, not a signal of anything. The real filtering is the four deterministic rules below, which don't use this mask at all.
Loading chain data…
mined hits (binned by frequency)
known crowded band
What actually decides whether a hit becomes a block
Four deterministic rules, recomputed independently by every node from the same public dataset -- the real triage, not the classifier below. A hit either clears all four or it doesn't; nothing about passing them comes in degrees.
No-drift exclusion. A hit with zero measured Doppler drift is rejected -- a population specific to this corpus, at least partly an instrument-processing artifact (confirmed for ≥15.8% of it; the rest is open across three named hypotheses -- see /audit §6 for the full breakdown), that modern data generation no longer produces.
Reference-position exclusion. Every observing cadence alternates the target star with reference pointings. A hit has to come from an actual target-pointing position, not merely a file whose name suggests that.
Cadence coincidence. Rejected if any detection from the same cadence's reference positions falls within a tolerance derived from the pair's own drift rate and the real time elapsed between them.
Cross-target coincidence. Rejected if a detection from a different target star sits within 50Hz of it anywhere in the dataset -- a threshold anchored to the measured frequency spread of a real, known terrestrial emitter.
Unpredictable candidate window. Which hits are even eligible to try for the next block is derived from the previous block's own hash -- the window doesn't exist, even to the miner who ends up using it, until the block that seeds it does. A public dataset with a fixed reading order would otherwise let anyone pre-score candidates for several blocks ahead of time, off-network.
Hit authenticity. Every node holds its own local copy of the canonical dataset and looks up any hit a block claims by its exact position, requiring every field to match. A set of numbers engineered to clear the classifier but never actually observed by any telescope gets rejected here, not accepted as a genuine detection.
These four rules were retroactively applied to this network's own early mined history and found a real error in how the second rule had been implemented -- see /audit for the full record.
The declared score -- published, not verified
A model trained on real labels -- not a hand-tuned heuristic. Labels come from cross-referencing "ON" (pointed at the star) against "OFF" (pointed away) observations of the same cadence: a signal that shows up in both is terrestrial interference (RFI), not a signal from the star.
Four features per hit -- frequency, drift rate, signal-to-noise ratio, and whether the frequency falls in a band known to be crowded with terrestrial transmitters -- feed a Random Forest classifier. The output is a single number from 0 to 1: the model's confidence that a hit is NOT interference. A block's declared score has to clear a fixed threshold to be valid at all.
But this score isn't what actually decides whether a hit can become a block -- the four rules above do that. The score itself is declared by the miner, and is never recalculated by other nodes when they check a block -- the same way classifier_version only has to be present, not match a specific model. Every block that clears the gate pays the same fixed reward regardless of the exact score attached to it -- the number is published, not paid for.
Every score the network has actually produced
No offline comparisons, no hand-picked examples -- these are the declared verification scores behind every block the network has mined, straight from the live chain, refreshed as the network mines more. Every block shown here already cleared every deterministic check and the declared-score threshold; what varies is the score itself, and whether its hit sits inside a frequency band already known to be crowded with terrestrial transmitters.
Loading chain data…
outside known crowded bands
inside a known crowded band
A permanent copy, for citing this outside the chain
Everything above is live, which is exactly the point -- but a live page changes under the same URL every time a block is mined, which makes it a poor thing to formally cite. This is what the deposit below is for.
By design, every 100 confirmed blocks the network publishes a snapshot to Zenodo -- a research-data archive run by CERN and OpenAIRE, the standard place independent and citizen-science work gets a permanent, citable DOI without needing a journal's gatekeeping first. Each snapshot contains the full hit data behind every block mined so far, the verification score, and the exact classifier model file that produced it -- so a third party can independently re-score any row and reproduce the number, without trusting this project's own servers. Automated publishing is currently paused after a real incident tied it to the wrong chain following a consensus reset -- see /audit §10.3 for the full record.
Each snapshot is published as a new version of the same record, under one permanent DOI that always resolves to whichever is most recent, and every earlier version stays exactly as published, forever -- that part of the mechanism is unaffected by the pause above. See /dataset for the current DOI, files, and download links, always pulled live from Zenodo's own record.
What this is not
Worth saying plainly, not burying in a footnote.
This system does not claim to have found a technosignature. The underlying data is public, and the original survey already concluded -- after a much more thorough, multi-signal analysis -- that none of these target stars host the kind of transmitter it searched for.
What's novel here isn't the astronomy -- it's the system: a way to turn continuous, incentivized, independently-auditable triage of real scientific data into the actual consensus work of a blockchain, instead of throwing away the electricity on arbitrary hashing.