Everything that dims.
1,011 entries, none thrown away.
Every consensus dip of the panel, every brightening and every star's variability from the sweep, each classified by a stated rule, with its confidence, its cues and its provenance: the marks, the weights and the Sentinels' own words. Sentinels run on Claude (Anthropic) by default; other providers are supported and labelled. Probes are deterministic matched-filter detectors.
The sweep so far
Every star searched, every dip kept and classified, nothing thrown away. So far: Sector 1: 200 of the 15,898 SPOC 2-minute targets searched (Probes only; by SetuHQ). A first slice, not the sky.
- Planet candidate3null 2 · excess +1 (95%: −3 to +5)not distinguishable from noise in this slice
- Eclipsing binary116null 77 · excess +33 (95%: −11 to +89)not distinguishable from noise in this slice
- Flare11null 5 · excess +6 (95%: −6 to +19)not distinguishable from noise in this slice
- Pulsator or rotational variable18null 2 · excess +6 (95%: 0 to +12)not distinguishable from noise in this slice
- Systematic742null 728 · excess −8 (95%: −103 to +88)not distinguishable from noise in this slice
- Unclassified121
Read from the sweep's committed catalogue (built 2026-10-20 13:14 UTC); nothing is extrapolated. Each class is set against its own null on the same 200 star-sectors (what its rule finds where its signal cannot be; the rules table says which). In this slice no class exceeds its null: every one is not distinguishable from noise in this slice. No entry is a new planet.
A first slice, stated plainly. Sector 1: 200 of the 15,898 SPOC 2-minute targets searched (Probes only; by SetuHQ). 200 stars searched, 200 star-sectors, 936 consensus dips reviewed. This catalogue is separate from the discovery feed. Nothing here is a new or confirmed planet. A class is what the light curve's cues suggest; there is no pixel or centroid check, so a blended binary can look like anything. The planet-candidate rule found 3 on the real light curves and 2 on flux-inverted copies of the same stars (dips turned into bumps, so real dips cannot pass): an excess of +1 (95%: −3 to +5). They are not distinguishable from noise in this slice, so they are not presented as candidates. 0 match a known TOI.
Each class against its own null, measured on the same 200 star-sectors: what its rule finds where its signal cannot be. Dips: the flux turned upside down (a real dip becomes a bump; a real flare or pulsation maximum becomes a dip, so this null is not empty of real signals). Flares: downward events with the real dips masked. Variability: a red-noise copy of the star (AR(1) fitted to its bins) through the same rule, which needs the same best period (within 10%) in both halves of the light curve. Of its periodogram peaks not at a binary's period, 8 are coherent; unclassified are 20 longer than a half can test (coherence not tested), 5 at the edge of the period grid and 3 that wander between halves. Limits: it finds a 300 ppm sinusoid in strong red noise only 3/20 times (scripts/variability_rule_check.py), and cannot test periods longer than about 3.8 d in one sector. Kind is compared with kind, star-sector is the unit: the 95% interval is a paired bootstrap over star-sectors, beside McNemar's test. A class that does not pass both is not distinguishable from noise in this slice: Planet candidate, Eclipsing binary, Flare, Pulsator or rotational variable, Systematic. Of the 11 flare-class brightenings, 8 rise faster than they decay and 2 the other way (1 equal); real flares rise fast, noise has no preferred direction.
TIC
No entries match these filters. Clear filters
The rules, and each class against its null
First match wins, in this order: systematic, eclipsing binary, variable, then planet candidate; a brightening is a flare unless it is systematic. The rules are in src/setuhq/variables.py and are re-applied to every contributed entry on ingest. Excess: real minus null over the same star-sectors, with a 95% interval from a paired bootstrap over star-sectors (10,000 resamples). The second table has every comparison, kind by kind.
| Class | Rule | Confidence | Real | Null | Null measured on | Excess |
|---|---|---|---|---|---|---|
| Planet candidate | Passed vetting, labelled a potential planet by half the panel's weight, not off its star's own periodic signal, and taken by no rule above. A candidate, unconfirmed. | medium: on a periodic signal with two or more transits, neither V-shaped nor density-mismatched; low: otherwise. Never high. | 3 | 2 | the flux-inverted copy: its dips of this class | +1 (95%: −3 to +5) not distinguishable from noise in this slice |
| Eclipsing binary | Odd/even depth mismatch or secondary eclipse; or a depth of 30,000 ppm or more; or V-shaped and 5,000 ppm or more; or a stellar-density mismatch at a period under 1 d; or half the panel's weight saying eclipsing binary. Variability at an eclipsing binary's period, or half of it. | high: odd/even mismatch or secondary eclipse; medium: depth, V shape, a short-period density mismatch, or the eclipse period; low: the panel's reason alone. | 110 | 77 | the flux-inverted copy: its dips of this class | +33 (95%: −11 to +89) not distinguishable from noise in this slice |
| Flare | A brightening: 3 or more consecutive points at least 3σ above a 0.5-day running median, peaking at 5σ or more, against both the star's scatter and the scatter within 0.25 d; not on an instrumental cue. | high: 5 or more points, a peak of 10σ or more and a faster rise than decay; medium: two of those; low: otherwise. | 11 | 5 | downward events with the real dips masked | +6 (95%: −6 to +19) not distinguishable from noise in this slice |
| Pulsator or rotational variable | A periodogram peak with Baluev FAP below 0.001 and a semi-amplitude at least 2× the scatter of 10-minute bins, coherent: the same period (within 10%) is the best peak of each half of the light curve, and not within 10% of the longest period searched; a period shared by 3 or more stars of a sector is systematic. A dip lasting 0.25 of that period or more is a trough of it; so is a dip half the panel's weight called stellar variability. | high: FAP below 1e-10 and a semi-amplitude 5× the binned scatter or more; medium: semi-amplitude 3× or more; low: otherwise (or a dip by the panel's reason alone). | 8 | 2 | a red-noise copy of the same star (AR(1) fitted to its bins), through the same rule and coherence test | +6 (95%: 0 to +12) not distinguishable from noise in this slice |
| Systematic | On an instrumental cue: a subject edge; a momentum dump, a scattered-light flag or a pointing or manual-exclude flag within 0.1 d; a data gap over 0.5 d within 0.1 d; or half the panel's weight calling it instrumental. | high: two or more instrumental cues; medium: one; low: the panel's reason alone. | 720 | 728 | the flux-inverted copy: its dips of this class | −8 (95%: −103 to +88) not distinguishable from noise in this slice |
| Unclassified | Everything else, with the reason no class applied. | n/a. | 121 | — | — | — |
| Comparison | Real | Null | Excess (95%) | Star-sectors with any: real / null | McNemar p | In this slice |
|---|---|---|---|---|---|---|
| Planet-candidate rule (dips) | 3 | 2 | +1 (95%: −3 to +5) | 3 / 2 | 1 | not distinguishable from noise in this slice |
| Eclipsing binary (dips) | 110 | 77 | +33 (95%: −11 to +89) | 26 / 20 | 0.238 | not distinguishable from noise in this slice |
| Pulsator or rotational variable (coherent periodogram peak) | 8 | 2 | +6 (95%: 0 to +12) | 8 / 2 | 0.109 | not distinguishable from noise in this slice |
| Variable-star troughs (dips) | 10 | 17 | −7 (95%: −18 to +1) | 3 / 4 | 1 | not distinguishable from noise in this slice |
| Flare (brightenings) | 11 | 5 | +6 (95%: −6 to +19) | 5 / 3 | 0.727 | not distinguishable from noise in this slice |
| Systematic (dips) | 720 | 728 | −8 (95%: −103 to +88) | 97 / 99 | 0.871 | not distinguishable from noise in this slice |
| Systematic (brightenings) | 22 | 25 | −3 (95%: −42 to +27) | 11 / 6 | 0.227 | not distinguishable from noise in this slice |
Not compared, for want of a null of the same kind: 6 eclipsing binary entries of another kind.
Who searched
- SetuHQ 200 star-sectors · sector_0001.json.gz
Anyone can add a sector with their own reviewers and be credited by a name they choose: see CONTRIBUTING.md (keys stay in your environment, never in a bundle). A bundle is sealed with content hashes, not signed; ingest searches a sample again and credits only searches that are new.