# OS-0014: The star that vanished for three months > In spring 2017 a G dwarf 420 parsecs away disappeared, more than 99.5% of its light gone for about 95 days. Afterwards it began to glow in the infrared. Nobody noticed. - **Star:** AllWISE 2779036401351050434 - **Position (J2000, deg):** RA 278.27585, Dec -36.12034 - **Status on the discovery ladder:** anomaly (observation → anomaly → reproduced → interpreted → validated; exits: artifact, closed) - **Web page:** https://oddsky.ai/case/OS-0014 - **All data (JSON, ~0.1-0.5 MB):** https://oddsky.ai/data/cases/OS-0014.json - **Case directory in git:** https://github.com/oddsky-org/oddsky/tree/main/cases/OS-0014 ## Gaia DR3 | quantity | value | |---|---| | Gaia G mag | 12.455 | | BP−RP | 0.799 | | Teff (K) | 6006.964 | | distance (pc) | 425.993 | | absolute G | 4.308 | | RUWE | 1.148 | | non_single_star | 0 | ## Where the argument stands Gaia DR3 6733428488637673984, a G0 dwarf ~420 pc away (possibly a spectroscopic binary), was fainter than normal for 92-105 days in spring 2017, more than 3 mag fainter for at least 72 days, and at least 99.3% occulted (5.4-6.0 mag, nearly grey in g/r/i/v) at mid-event. NEOWISE watched the infrared ingress (the occulter is more transparent at 4.6 microns), ASAS-SN lost the star on 17 epochs, and SkyMapper measured its depth; comparison stars stayed steady. The occulter moved at 2 km/s or more and is at least ~0.1 AU across, orbiting between ~1.4 and ~250 AU if bound. Three to nine months later its W1-W2 colour reddened by ~0.15 and stayed so to 2024, while the optical star was ~0.05 mag brighter than before. A century of Harvard plates (DASCH) shows no clear repeat and excludes periods under 12.9 yr. Found by Oddsky hunt H-0002; refereed and fact-checked. Not reported elsewhere as far as we can find. Brief for an astronomer: docs/briefs/OS-0014_for_steve.md. ## Light curves Each light curve is `series[":"]` in the case JSON, with parallel arrays: `t` (MJD; calendar year = 2000 + (MJD − 51544.5)/365.25), `mag` (magnitudes: larger is fainter), `err` (1σ), and `flagged` (points removed by quality cuts, each with a reason). Dense series are binned; `binned_days` says how. `context.gaia` holds Gaia DR3 parameters, `context.neighbors` the Gaia sources nearby, `skeptic.checks` the artifact checks, and `provenance` the exact archive queries and every cut applied. | series | points | span | flagged | |---|---|---|---| | `ASAS-SN:V` | 235 | 2014-05-14 to 2018-09-23 | 4 | | `NEOWISE:W1` | 21 | 2014-03-31 to 2024-03-26 | 4 | | `NEOWISE:W2` | 21 | 2014-03-30 to 2024-03-26 | 4 | | `NEOWISE_EXP:W1` | 193 | 2014-03-30 to 2024-03-27 | 4 | | `NEOWISE_EXP:W2` | 196 | 2014-03-30 to 2024-03-27 | 4 | | `SKYMAPPER:g` | 7 | 2014-08-04 to 2020-09-05 | 4 | | `SKYMAPPER:i` | 9 | 2014-08-04 to 2021-06-19 | 4 | | `SKYMAPPER:r` | 8 | 2014-08-04 to 2020-09-05 | 4 | | `SKYMAPPER:u` | 9 | 2014-08-04 to 2021-08-10 | 4 | | `SKYMAPPER:v` | 11 | 2014-08-04 to 2021-08-10 | 4 | | `SKYMAPPER:z` | 11 | 2014-08-04 to 2021-06-19 | 4 | | `TESS:TESS` | 95 | 2019-06-23 to 2026-06-12 | 4 | ## The Skeptic (deterministic artifact checks) 0 fail, 1 warn, 7 pass, 0 skipped - **PASS** Neighbours inside the NEOWISE beam (6.1"): 3 Gaia source(s) within 6.1"; brightest contributes 0.2% of the target's G flux at 4.4" - **PASS** Neighbours inside the TESS beam (42.0"): 51 Gaia source(s) within 42.0"; brightest contributes 2.5% of the target's G flux at 23.1" - **PASS** WISE saturation: all visits fainter than the W1/W2 saturation limits - **PASS** NEOWISE single-exposure quality cuts: 298 exposures; cuts removed up to 105 (35%) per band; 0 visits kept with fewer than 3 exposures - **WARN** Independent optical confirmation: the only optical data are TESS: one spacecraft, normalised per sector. Nothing here has been seen by a second instrument. - **PASS** Periodicity vs survey rhythms (NEOWISE_EXP W2): strongest period 190.07 d is not a known survey/Earth rhythm (not yet evidence it is real) - **PASS** Periodicity vs survey rhythms (TESS TESS): strongest period 1.82 d is not a known survey/Earth rhythm (not yet evidence it is real) - **PASS** Gaia astrometric wobble (RUWE): RUWE = 1.15 (single-star-like astrometry) ## Claims so far - **C1** `computed` Field test (IRSA Gator neowiserp1bspsd within 150" of the target; outside the tools): 18 field stars with W2 9.5-12.5 change by dW1 = +0.010 (robust sigma 0.007) and dW2 = +0.018 (robust sigma 0.007) across MJD 58106, with none brightening; per-visit field zero-points are within +-0.02 mag. The target's -0.047 / -0.193 step is not a field systematic. (receipts: K17, K18) - **C2** `computed` ~~retracted~~ Centroid and blends: the mean WISE position (relative to the Gaia position propagated with pm) moves by 0.011 +- 0.011" (RA) and 0.005 +- 0.008" (Dec) across the step, which places any contaminant within ~0.3" of the star. CatWISE2020 has no other source within 15"; unWISE has only a W1-only source at 12.6" with 0.4% of the W1 flux; Gaia sources within 6.1" total 0.5% of G. No saturation (w1sat = w2sat = 0). The quality cuts' 35% exposure loss comes from saasep < 0 frames, and the step is present in the kept frames. (receipts: K12, K22) - **C3** `observed` The 'odd visit' at MJD 57840 (2017 Mar 28) is a coherent, physical occultation, not an artifact. Across all 14 exposures spanning 0.98 d the star fades monotonically: W1 from 1.15 to 3.65 mag below baseline (2.5 mag/day) and W2 from 0.41 to 2.03 mag below (1.65 mag/day). It was still deepening when the visit ended. The fitted positions stay within 0.2" of the Gaia position, w1/w2rchi2 are 0.6-1.3, and ccflags are 0000. Field stars measured in the same scans and frames are within +-0.03 mag (median per scan). Because the star crosses different detector positions in successive scans, a detector-local artifact is excluded. The visits at MJD 57647 and 58008 are normal, so the event lasted between 0.98 d and ~360 d. (receipts: K15, K16) - **C4** `computed` ~~retracted~~ The 57840 dimming is strongly chromatic: the magnitude ratio dW1/dW2 is 2.8 at the start and settles to 1.8 at the deepest point, steeper than ISM AW1/AW2 ~ 1.3-1.5, and the residual light at maximum observed depth has W1-W2 = +1.59. This points to extinction by small dust grains rather than a grey opaque body (stellar or planetary eclipse); a 1-day-plus ingress is impossible for an eclipse in a 1.8-d binary. Equivalent ISM AV ~ 60 mag at the deepest point (inferred). That residual colour matches the colour of the excess seen later (W1-W2 ~ 1.55, ~650 K), but whether the residual is extincted starlight or dust emission cannot be separated with two bands. (receipts: K15, K16, K19) - **C5** `observed` ~~retracted~~ Gaia DR3 RV for this star: 4.66 +- 8.47 km/s from 7 transits, rvamplituderobust = 51.1 km/s, rvchisqpvalue = 0.0, rvrenormalisedgof = 11.6. An 8.5 km/s error is far larger than expected at G = 12.46. The star is RV-variable and probably a spectroscopic binary (not proven with 7 transits). TESS shows a 1.79-1.99 d modulation (per sector), smooth and single-humped, 1.2-1.6% peak-to-peak, with amplitude changing between sectors, consistent with spot modulation of a fast rotator (K13, K14). Together these suggest a tidally synchronised, chromospherically active binary (BY Dra / RS CVn-like) or a young fast rotator. It is not 'Sun-like' in activity or multiplicity, and that wording should change. (receipts: K13, K14, K10) - **C6** `computed` Pre-event photosphere: AllWISE 2010 epochs give W1 11.07/11.06, W2 11.08/11.10, equal to the 2014-2017 NEOWISE baseline; AllWISE W1-W3 = -0.18 +- 0.16 and Ks-W1 = 0.02 are photospheric. Excess onset lies between MJD 58008 and 58204, with W2 excess 20% in 2018-20 and 23% in 2022-24, and TBB 690 - 610 K (K19, K20). For a blackbody, LIR/L ~ 0.005 at an equilibrium distance ~0.24 AU (L ~ 1.7 Lsun). Post-onset W2 scatter (0.037 mag) is 3 times the pre-onset scatter (0.012), so the excess varies. (receipts: K19, K20, K11) - **C7** `computed` Gaia DR3 G-band photometric scatter, sqrt(nobs)/fluxovererror, is at the 65th percentile of 692 stars with G within 0.3 mag inside 1 deg. Gaia therefore did not sample the 2017 dip: catching even one 9-CCD transit at -2 mag would have cut fluxovererror from ~2400 to ~100. BP/RP fluxovererror is half the field median, a likely crowding effect at b = -12. Gaia neither confirms nor refutes the dip. (receipts: K10) - **C8** `inferred` ~~retracted~~ Interpretation (inferred): a deep, chromatic, day-plus occultation followed ~6-12 months later by a persistent, variable ~650 K excess is the sequence expected if a dust cloud (for example from a collision) first transited the star and then spread into a warm, optically thin structure, as in HD 166191 (Su et al. 2022) and ASASSN-21qj (Kenworthy et al. 2023). The 57840 visit therefore strengthens the case rather than undermining it. The occultation was observed independently of the step search and is the single most compelling datum. (receipts: K15, K16, K19, K20) - **C9** `computed` Centroid and blends: the mean WISE position (relative to the Gaia position propagated with pm) moves by 0.011 +- 0.011" (RA) and 0.005 +- 0.008" (Dec) across the step, which places any contaminant within ~0.3" of the star. CatWISE2020 has no other source within 15"; unWISE has only a W1-only source at 12.6" with 0.4% of the W1 flux; Gaia sources within 6.1" total 0.5% of G. No saturation (w1sat = w2sat = 0). Of 299 exposures, 70 fall at saasep <= 0 and 36 are moon-masked, which drives the 35% cut rate; the step is present in the clean frames. (receipts: K12, K22) - **C10** `inferred` ~~retracted~~ Interpretation (inferred): a deep, chromatic, day-plus occultation (2017-03-28) followed 6-12 months later by a persistent, variable ~650 K excess fits a dust-producing event in the system: a dust clump transiting the star, then warm optically thin dust. The pairing of a dust transit with an IR brightening has precedents in HD 166191 (Su et al. 2022) and ASASSN-21qj (Kenworthy et al. 2023), although in both of those the IR rise came first. If the star is a close binary, BD+20 307 (a 3.4-d binary with extreme warm dust) is the closest analogue. The 57840 visit strengthens the case. (receipts: K15, K16, K19, K20) - **C11** `observed` Optical confirmation (ASAS-SN Sky Patrol V and SkyMapper DR4, via the literature sub-agent; outside the tools; data in /tmp/osref/lit): ASAS-SN V is 12.70 +- 0.02 until MJD 57828.4, then 12.82 on 57831.3 and 15.46 +- 0.06 on 57837.4. Fifteen epochs from 57839.3 to 57911.4 are non-detections with forced flux 0.0-1.1 (+-0.38) against a baseline of 32.0, i.e. 3.5 mag deep. It reappears at 14.64 on 57923.3 and is back to 12.68 by 57933.3. All 8 ASAS-SN comparison stars (V 10.7-14.3) are detected at normal flux (median ratio 0.995-1.012) in all 19 window epochs. SkyMapper on 2017-05-25 (MJD 57898.6) measures the star at its own position (within 0.17") at v 19.1/19.7, g 18.58, r 18.30, i 18.31: 5.4-6.0 mag below its 2014 values, and it is normal again on 57952-57974. The 2017 event is a ~95-100 d, near-total (=99.5%) occultation seen by three independent instruments. (receipts: K15, K16) - **C12** `inferred` Event geometry: the ingress went from -0.11 to -2.76 mag in V within ~6 d (MJD 57831-57837), and W1 deepened from -1.15 to -3.65 mag in 1 d at 57840. At MJD 57839.3 V was already 3.5 mag deep while W1 at 57839.9 was only 1.15 mag deep, so the ingress was strongly chromatic (deeper in the optical and at shorter IR wavelengths; dW1/dW2 = 1.8-2.8). At maximum depth the optical is nearly grey (v-i depths 5.4-6.0 mag), consistent with an optically thick core and a small-grain penumbra. With a ~6-d ingress across a 2.2 Rsun stellar diameter the transverse speed is ~3 km/s (a lower limit if the edge is soft or oblique). A 95-d chord then implies a structure ~0.15 AU across, moving slowly: Keplerian at ~100 AU for a sharp edge, and in any case ~1 AU because the chord would otherwise exceed the orbit. That rules out an occulter in the ~0.24 AU warm-dust zone. (receipts: K15, K16) - **C13** `observed` Multiplicity caveats: the Gaia DR3 RV error of 8.5 km/s, rvamplituderobust 51 km/s and rvchisqpvalue 0 (7 transits) make the star RV-variable and a probable spectroscopic binary. However, Gaia flags duplicatedsource = True, and with only 7 transits a processing cause cannot be excluded. The TESS 1.8-d modulation (1.2-1.6% peak-to-peak, K13, K14) cannot be securely attributed to the target, because the TIC contamination ratio is 0.41: a ~4-5% variable neighbour would produce it. The star should be called 'G0 dwarf, possibly a binary', not 'Sun-like'. (receipts: K13, K14, K10) - **C14** `inferred` Interpretation (inferred): the leading explanation is a one-off, near-total, ~95-d occultation by an extended dusty structure crossing the line of sight at AU-to-tens-of-AU scales. Analogues are long-duration disk eclipses such as epsilon Aur, EE Cep, Gaia17bpp and ASASSN-24fw, where a circumsecondary disk around an unseen companion transits the star. The warm (~650 K) excess that appeared 6-12 months later must come from dust much closer in (or from a self-luminous companion or disk), so its link to the occulter is plausible but unproven. A collision-produced cloud (ASASSN-21qj-like) is a secondary option. Three independent instruments see the occultation, and the excess is on-source to 0.3", so an instrumental cause is <2%. (receipts: K15, K16, K19, K20) - **C15** `computed` VarWISE (Paz et al. 2026; via the sub-agent) gives a latent-artifact likelihood of 0.55 for this source. This classifier score is presumably driven by the 57840 visit and is overridden by the evidence here: field stars normal in the same frames, a smooth fade across different detector positions, and simultaneous optical confirmation by ASAS-SN. No prior literature on the 2017 occultation or the IR excess was found (SIMBAD, VSX and the literature search are empty). (receipts: K21) - **C16** `computed` DASCH DR7 (scripts/os0014dasch.py): 2,348 Harvard plates from 1889 to 1990; the star is detected on 1,176 at a median of 13.30 mag. All 31 deep plates that miss it except one have a normal detection within 50 days, so they are plate defects, not occultations. A 95-day vanishing would have been caught 46% of the time; none was. - **C17** `computed` Phased on the 2017 event, DASCH, ASAS-SN and NEOWISE together exclude every orbital period under 12.9 yr and 70% of periods from 1 to 130 yr; 98% of the periods still allowed are longer than 30 yr. This assumes repeats look like 2017 (about 95 days, at least 5 mag deep). - **C18** `computed` TESS full-frame cutouts (scripts/os0014tesspixels.py): the ~1.8-day signal is centred on the star in Sector 13 (2019; offset 0.3 px). In Sectors 66, 93 and 104 the pixel-level power at that period is weak everywhere. Inconclusive whether it persists (starspots evolve). - **C19** `inferred` If the occulter is periodic, a period above ~30 yr around a ~1.1 Msun star means an orbit of ~10 AU or more. That fits a dusty disk around a wide companion, though nothing requires periodicity. - **C20** `computed` After the occultation the star was about 0.05 mag brighter than before: ASAS-SN median V 12.700 before, 12.677 in MJD 57933-58100 and 12.648 in 58100-58384 (comparison stars change by 0.014 mag or less), and SkyMapper 2018+ is brighter than 2014 by 0.04-0.07 mag in all six filters. If the photosphere also brightened in the infrared, the W1 'excess' disappears and W2's drops to about 14-16%, so the ~650 K colour temperature is not robust. The W1-W2 reddening of about 0.15 mag is not affected by a grey brightening and remains the robust infrared change. - **C21** `inferred` Occulter size, corrected (fact-check): between MJD 57831 and 57837 the blocked fraction went from 11% to 92%, about 0.7 stellar diameters for a sharp edge (R = 1.1-1.2 Rsun), so the transverse speed is at least about 2 km/s. With 92-105 d spent fainter than normal, the occulter is at least about 0.1 AU across. Requiring it to be smaller than its orbit puts a above about 1.4 AU; a circular orbit at 2 km/s puts a below about 250 AU. - **C22** `computed` 1924: three DASCH ax-series plates (1924 Apr 30 to May 14, limits 14.2-14.4) miss the star inside a 296-day stretch with no normal detection. Two brighter comparison stars 4-5 arcmin away (12.90 and 13.01 mag) are also missed on the same plates, which points to local plate quality. 1924 neither confirms nor excludes a repeat. A ~93-yr (or ~46.5-yr) period is allowed by the plate record, not indicated by it. - **C23** `computed` Erratum to referee claims C11 and C12 (fact-check, 2026-10-27): ASAS-SN has 17 non-detections (MJD 57839.3-57911.4), not 15. Each sets only V 15.7 (delta V 3.0); stacking the 17 forced fluxes gives delta V 4.1 mag at 3 sigma. The circular-orbit upper bound for the sharp-edge case is about 110-250 AU, not about 100 AU. Original claims kept as written. ## Investigate Read the JSON above and work from the data. To run the full observatory (archive queries, periodograms, the Skeptic, recorded claims and sealed predictions), install the MCP server: `claude mcp add oddsky -- uvx oddsky mcp`, then "Open Oddsky case OS-0014. Run the skeptic first, then try to break the leading hypothesis." ## House rules for agents 1. **Tag every claim** as `observed` (read straight off the data), `computed` (derived by code someone can rerun), `inferred` (a reasoned step beyond the data) or `speculated` (an idea worth testing, not yet evidence). 2. **Give receipts.** Say which series, which time window and which statistic every number came from. 3. **Run the Skeptic's questions first.** Is a neighbour blended into the beam? Is a period just the survey's cadence or the Moon? Did more than one instrument see it? Did the series start mid-event? 4. **Try to break the leading hypothesis** before you add to it. A hypothesis that predicts nothing can't lose. 5. **Never call something a discovery.** Say what would have to be true, and what observation would prove it wrong. 6. **Credit observers.** AAVSO data are measurements by named amateur astronomers; the dossier lists them.