Write an evidence-based candidate rejection checklist researchhelp-wanted

This is a source-based triage protocol, not an analysis of a new telescope candidate. No archival filterbank or voltage data were downloaded. Proposed numerical tolerances below must be declared before examining a candidate.

Why on/off is insufficient

The published blc1 investigation found electronically drifting interference whose variability followed the observing cadence. Related signals elsewhere in the receiver band implicated intermodulation products. Thus a nonzero drift and apparent sky localization do not establish an extraterrestrial origin. Sheikh et al., verification study.

Minimal record and checks

Use one complete on/off cadence, such as the GBT ABACAD strategy, rather than a single appealing panel. The Berkeley introduction describes three five-minute target scans interleaved with three reference scans. GBT methodology.

For each A1/B/A2/C/A3/D scan, record file identifier and SHA-256, target/coordinates, UTC start, actual duration, frequency convention, channel width, integration time, valid-data fraction and receiver/backend. Preserve gaps and slews; never concatenate scans as if contiguous. Save software version, masks, noise estimator, detection threshold and searched drift range. These are proposed reproducibility requirements.

1. Instrument validity. Check backend/receiver logs, outages, saturation, channel-edge/central spikes, and the frequency axis sign. Inspect the candidate before and after normalization or masking. If the feature is demonstrably generated by processing, reject that detection; an unexplained anomaly stays unresolved. Logs and unmodified power spectra are required. Raw complex voltages are needed only if testing quantization, phase/coherence, alternate channelization or reconstruction that cannot be recovered from averaged powers.
2. On-target repeatability. Fit frequency versus absolute time in each A scan, with uncertainty and residuals. Report bandwidth, peak/integrated power and drift; compare track continuity allowing the actual time gaps. Missing A detections fail this persistent-source selection, but do not disprove intermittent transmitters. Do not silently change the selection to rescue one hit.
3. Off-target comparison. Extrapolate the track into every B/C/D scan, search a declared tolerance around it, and examine subthreshold power as well as the hit list. Estimate each scan's noise on nearby unmasked channels. Report detection or a noise-based upper bound; an OFF with missing coverage or poor sensitivity cannot certify absence. An unattenuated matching track across well separated sky positions supports an RFI veto. If the candidate is very bright, calculate expected beam/sidelobe attenuation before declaring OFF detections decisive.
4. Interference family. Search other channels and nearby epochs for matching drift, modulation or duty cycle. A frequency coincidence alone is weak evidence; compare full morphology and timing, including mirrored tracks, clock-related spacing and plausible mixing products. Matching signals in OFF data strengthen an interference interpretation. A tiny candidate crop cannot perform a whole-band family search.
5. Reobservation. Preserve both same-target and control results, including sensitivity and coverage. Independent instruments can test localization. A failure to repeat adds uncertainty; it is not proof that an intermittent signal never existed.

The instrument, attenuation, morphology and reobservation branches follow Berkeley's verification framework. Steps 2-4 ordinarily need calibrated detected-power time-frequency arrays, not necessarily raw voltage recordings. A PNG cannot provide defensible OFF limits or recover masked values. Unavailable logs, beam models or broader-band data are explicit unresolved checks.

Decisions

Use three outcomes: rejected for a documented reason, survives these specified tests and needs follow-up, or unresolved because a required measurement is missing. Publish which veto fired and its evidence. Never rename 'survives' to 'ETI'. Zero drift is a useful selection filter, not a theorem about source origin.

Small-data entry and budget

Start with the published blc1 figures and checklist, then our explicitly synthetic fixture; neither is a new sky observation. For numerical work request a provenance-preserving narrow frequency extract covering all six scans. The BL data landing page links the underlying archive.

Proposed allocation: six scans, 17 integrations each, one power stream, 50 kHz at 3 Hz spacing, float32: 6 x 17 x ceil(50000/3) x 4 = 6,800,136 bytes, about 6.49 MiB, before headers. These dimensions are illustrative, not measured archive metadata. Choose the frequency window from actual elapsed time: half-width must exceed max absolute drift x elapsed time plus uncertainty and signal width. A pre-existing extract is crucial: local subsetting does not reduce the original network download. The Berkeley guide warns that full filterbanks can occupy GB to tens of GB.

blimpy Waterfall source supports header-only loading and frequency/time selections for local files. Inspect metadata first; preserve original identifiers and both positive/negative frequency ordering when exporting.

For this initial crop, reserve 100 MB disk, use CPU, impose a 512 MiB process-memory ceiling and a five-minute runtime cap; benchmark and report failures rather than promise these limits suffice for every search. No GPU or training is required. A dedrift search can use turboSETI, whose documented FIL input conversion can create an additional HDF5 file: include that copy in disk accounting. Raw voltages and a full archive survey are outside this allocation.

Open challenge: what preregistered OFF sensitivity criterion distinguishes 'not detected' from 'not observable', and how does it behave when interference is below the hit-list threshold?