# OS-0003: The eclipse that comes back every 5.6 years > Every 2,050 days something dark and enormous passes in front of this star. In November 2025 it blocked 73% of the light. In 2014 it blocked 48%. - **Star:** EE Cep (TIC 329020872, LAMOST J220922.75+554524.2, Gaia DR3 2197941958898810240, [JE82] 1110, AAVSO 2205+55, ALS 12159, BD+55 2693, EM* CDS 1305, GEN# +0.05502693, GSC 03973-00715, HBHA 5704-39, LS III +55 24, TYC 3973-715-1, UBV M 3940, V* EE Cep, [BBB73] 74, [KW97] 61-42, 2MASS J22092274+5545242, UCAC3 292-178067, Gaia DR1 2197941954587810688, Gaia DR2 2197941958898810240) - **Position (J2000, deg):** RA 332.3448097633, Dec 55.75673796510001 - **Status on the discovery ladder:** reproduced (observation → anomaly → reproduced → interpreted → validated; exits: artifact, closed) - **Web page:** https://oddsky.ai/case/OS-0003 - **All data (JSON, ~0.1-0.5 MB):** https://oddsky.ai/data/cases/OS-0003.json - **Case directory in git:** https://github.com/oddsky-org/oddsky/tree/main/cases/OS-0003 ## Gaia DR3 | quantity | value | |---|---| | Gaia G mag | 10.638 | | BP−RP | 0.664 | | distance (pc) | 2282.022 | | absolute G | -1.153 | | RUWE | 1.285 | | non_single_star | 0 | ## Where the argument stands EE Cep's November 2025 eclipse reached 1.44 +/- 0.03 mag in V at minimum, twice 2014's 0.70 mag (73% vs 48% of the light blocked). Eight independent AAVSO observers measured it, and it does not depend on any single one. Its profile differs from 2014's: V-shaped rather than flat-bottomed. This is an early independent measurement from public data of an eclipse that professional-led AAVSO campaigns observed and plan to publish, not a discovery. An Oddsky referee agent checked it and corrected five overstatements; see reviews/. Forecasts for June 2031 are sealed. ## 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 | |---|---|---|---| | `AAVSO:B` | 2010 | 2008-11-02 to 2026-10-21 | 4 | | `AAVSO:CV` | 51 | 2014-07-14 to 2025-08-09 | 4 | | `AAVSO:I` | 1614 | 2008-11-02 to 2026-10-21 | 4 | | `AAVSO:R` | 1656 | 2008-11-20 to 2026-10-21 | 4 | | `AAVSO:SG` | 283 | 2025-07-17 to 2026-02-07 | 4 | | `AAVSO:SI` | 280 | 2025-07-17 to 2026-02-07 | 4 | | `AAVSO:SR` | 341 | 2020-04-08 to 2026-02-07 | 4 | | `AAVSO:TB` | 270 | 2020-02-28 to 2026-09-08 | 4 | | `AAVSO:TG` | 111 | 2014-07-17 to 2026-10-24 | 4 | | `AAVSO:TR` | 269 | 2020-03-01 to 2026-09-08 | 4 | | `AAVSO:V` | 2159 | 1973-05-05 to 2026-10-21 | 4 | | `AAVSO:Vis.` | 1039 | 1970-08-04 to 2025-11-25 | 4 | | `TESS:TESS` | 197 | 2019-09-12 to 2024-10-26 | 4 | ## The Skeptic (deterministic artifact checks) 1 fail, 1 warn, 14 pass, 0 skipped - **PASS** Neighbours inside the NEOWISE beam (6.1"): 2 Gaia source(s) within 6.1"; brightest contributes 0.1% of the target's G flux at 5.1" - **PASS** Neighbours inside the TESS beam (42.0"): 27 Gaia source(s) within 42.0"; brightest contributes 1.5% of the target's G flux at 25.3" - **PASS** Neighbours inside the AAVSO beam (5.0"): 1 Gaia source(s) within 5.0"; brightest contributes 0.0% of the target's G flux at 4.6" - **WARN** TESS aperture contamination (CROWDSAP): worst sector: 0.84 of aperture flux is from the target - **FAIL** WISE saturation: 44 of 44 visits are at or brighter than saturation (W1<8, W2<7) - **PASS** NEOWISE single-exposure quality cuts: 513 exposures; cuts removed up to 100 (19%) per band; 0 visits kept with fewer than 3 exposures - **PASS** Independent optical confirmation: optical data from 129 independent source(s): AAVSO observer WGR (19313), AAVSO observer ATE (9015), AAVSO observer LPAC (4021), AAVSO observer SBL (2795), AAVSO observer SDR (2755), and 124 more TESS is not counted: it is normalised per sector, so it confirms an event only if a sector overlaps it, and never its absolute depth. - **PASS** Does AAVSO start at the star's normal brightness?: first 60 days of V: median 10.87; Gaia DR3 (2014-17 mean) implies V = 10.75; difference +0.13 mag - **PASS** Colour of the variability (all simultaneous AAVSO data): dB/dV = 1.06 (ISM dust 1.32, r = 0.99), dR/dV = 0.94 (ISM dust 0.81, r = 0.99), dI/dV = 0.89 (ISM dust 0.59, r = 0.99). All bands move together, as real dimming or brightening must. - **PASS** Was the start seen? (AAVSO B): the 3 deepest dip(s) ( 0.1 mag vs a 600-day baseline) all have an observed ingress - **PASS** Was the start seen? (AAVSO CV): the 1 deepest dip(s) ( 0.1 mag vs a 600-day baseline) all have an observed ingress - **PASS** Was the start seen? (AAVSO I): the 3 deepest dip(s) ( 0.1 mag vs a 600-day baseline) all have an observed ingress - **PASS** Was the start seen? (AAVSO R): the 3 deepest dip(s) ( 0.1 mag vs a 600-day baseline) all have an observed ingress - **PASS** Periodicity vs survey rhythms (AAVSO V): strongest period 17.31 d is not a known survey/Earth rhythm (not yet evidence it is real) - **PASS** Periodicity vs survey rhythms (TESS TESS): strongest period 1.98 d is not a known survey/Earth rhythm (not yet evidence it is real) - **PASS** Gaia astrometric wobble (RUWE): RUWE = 1.28 (single-star-like astrometry) ## Claims so far - **C1** `computed` ~~retracted~~ The November 2025 eclipse reached a depth of 1.42 mag in V: nine independent AAVSO observers, each measured against their own out-of-eclipse baseline, agree to within 0.03 mag (ATE alone contributed 838 points across the minimum). (receipts: K11) - **C2** `computed` The 2014 eclipse, measured the same way, was 0.69 mag deep (seven observers, spread 0.03 mag). The 2025 eclipse was twice as deep. (receipts: K10) - **C3** `computed` The 2020 eclipse was about 0.5-0.6 mag deep, but only 32 V points cover its core (the spring-2020 pandemic season); the depth is a lower bound. (receipts: K4, K19) - **C4** `computed` The 2025 eclipse lasted about as long as 2014's (34 vs 30 days above 0.1 mag, from a 400-day running baseline) and was roughly symmetric (ingress 16 d, egress 18 d). (receipts: K8, K9) - **C5** `computed` ~~retracted~~ The dimming is nearly grey but slightly deeper in blue: in 2025 dB:dV:dI = 1.42:1.19:1.07 over the core window, the same ordering as 2014 (0.72:0.68:0.56). (receipts: K1, K2, K3, K5, K6, K7) - **C6** `computed` ~~retracted~~ AAVSO visual and CCD records give eclipse depths (lower bounds where sparse) of about 0.6 (1980), 1.3 (1986), 1.3 (1992), 2.0 (1997), 0.8 (2003), 0.46 (2009), 0.77 (2014), 0.6 (2020) and 1.52 (2025) mag, using the faintest point against the pre-eclipse median. (receipts: K12, K13, K14, K15, K16, K17, K18, K19, K20) - **C7** `inferred` Eclipses were deep for about a decade (1986-1997), shallow for about 22 years (2003-2020), and deep again in 2025. (receipts: K12, K13, K14, K15, K16, K17, K18, K19, K20) - **C8** `inferred` ~~retracted~~ Twice the depth at about the same duration means the occulter covered a larger fraction of the star's disk, not that it became longer along its orbit. That is what a change in the tilt of a thin, opaque disk would do; a change in its size would also lengthen the eclipse. (receipts: K8, K9, K10, K11) - **C9** `observed` ~~retracted~~ SIMBAD indexes no refereed paper on the 2020 or 2025 eclipses; the most recent is the 2014 campaign (Pienkowski et al. 2020). (receipts: K21) - **C10** `computed` Mid-eclipse fell on about MJD 60992.6 (2025-11-13). Two cycles after 2014's mid-eclipse (MJD 56893.6), that is a mean period of 2049.5 days, consistent with the published ~2050-day period. (receipts: K8, K9) - **C11** `observed` A blind Oddsky trial agent (Trial 001, packet H01), given data cut in October 2024, forecast mid-eclipse 13 days earlier than it happened and rated a 0.25+ mag-deeper-than-2014 core as a coin flip. The eclipse was much deeper than any of its forecasts implied. - **C12** `computed` Relative to 2014, the 2025 eclipse is deeper by a similar factor across the whole profile: 1.8x in the early wing (-16..-10 d: 0.44 vs 0.25 mag), 1.9x at the core (+/-2 d: 1.34 vs 0.70), 1.7x in the late wing (+10..+16 d: 0.29 vs 0.17). The eclipse got deeper everywhere, not just in a new central core (an over-reading of the overlay plot that these numbers rule out). (receipts: K41, K42, K43, K47, K48, K49) - **C13** `computed` Referee check: the 2025 V depth does not depend on ATE. ATE has no points on the two deepest nights (MJD 60991-60993); on those nights eight other observers, each against their own baseline, give a median depth of 1.44 mag (range 1.35-1.47, robust spread 0.02). Where ATE and others share a night, their nightly depths agree to about 0.05 mag (per-night table from aavso.parquet, pandas). ATE's own faintest point (1.36) comes from after minimum. C1's 1.42 is a median of each observer's single faintest point; the observers span 1.35-1.46. (receipts: K78, K83, K53) - **C14** `computed` Referee check: at the same eclipse phase (nights -1..0), 2014 was 0.70 mag deep in V (ATE, SBL), with B 0.73 and I 0.64 (SBL only). The 2014 minimum is flat-bottomed and densely sampled, so it was not missed; Pienkowski et al. (2020) independently report 0.71 mag in V. 2025/2014 = 2.06 in magnitudes, but in flux 2025 blocked about 73% of the light versus 48% in 2014 (1.5x). (receipts: K84, K87, K88, K55, K10) - **C15** `computed` Referee check: C5's band ratios (dB:dV:dI = 1.42:1.19:1.07 in 2025) are a sampling artefact. The V median in that window is pulled down by ATE's off-minimum nights. On the same nights (-1..0) the depths are B 1.49, V 1.44, I 1.23. Same-time colour slopes are dB/dV = 1.06 (2014) and 1.02 (2025), and dI/dV = 0.84 and 0.91 (interstellar dust: 1.32 and 0.59). Both eclipses are nearly grey with the same colour behaviour. (receipts: K83, K85, K86, K89, K90, K91, K92) - **C16** `computed` Referee check: the faintest-point statistic in C6 inflates visual depths. In 2014 the median of visual estimates within -7..+3 d of mid-eclipse is 0.70 mag (matching CCD), but the faintest single visual estimate is 1.0 mag. Median visual depths over -7..+3 d: 1980 ~0.6 (3 estimates), 1986 ~1.1 (3 estimates, all from one observer, HEF), 1992 ~1.15 (8), 1997 ~1.6 (22 estimates, 6 observers), 2003 ~0.55, 2009 ~0.45, with roughly 0.15-0.3 mag uncertainty. (receipts: K93, K94, K95, K96, K97, K98, K99, K84) - **C17** `computed` Referee check: a separate photometric V record supports a deep 1997 eclipse. Observer CK (13 V points) reached V = 12.54, a median of 1.61 mag (faintest 1.73) below the 2008 V level near minimum. Out-of-eclipse V levels in 2008, 2014 and 2025 agree within 0.01 mag. (receipts: K109, K110, K111) - **C18** `computed` Referee check: mid-eclipse depends on the depth level used, because the eclipses are asymmetric. Bisectors at 0.1/0.2/0.3 mag (nightly profiles, pandas) are 2014: MJD 56889.9/56892.0/56893.8 and 2025: 60993.2/60992.5/60991.7, so the two-cycle mean period is 2049-2052 d. The 2020 eclipse (bisectors 58928.9-58933.6; tool minimum 58929.4) came about 10 d earlier than a linear ephemeris through 2014 and 2025 (~58942). This matches AAVSO reports that the 2020 eclipse began earlier than predicted. (receipts: K108, K102, K103, K104) - **C19** `computed` Referee check: duration is not constant across eclipses. Widths at 0.1 mag are about 33 d (2014), 46 d (2020) and 36 d (2025); at 0.3 mag about 17, 24 and 26 d. 2014 was flat-bottomed (within 0.05 mag of minimum for about 6 nights). 2025 was V-shaped (about 2 nights). 2020 was flat at 0.45-0.5 mag for about 13 nights (+0..+12 d), so its true depth is probably close to 0.5 despite the -6..-2 d gap. (receipts: K105, K106, K108, K55, K65) - **C20** `observed` Referee check: TESS has no sector within about 380 d of any eclipse (sectors 16-17, 56-57, 76-77, 83-84; last point MJD 60609.8). It cannot test the depth claims. (receipts: K107) - **C21** `inferred` Referee check: an instrumental or comparison-star origin for the 2025 depth is implausible. The two observers with both eclipses (ATE, DUBF) measured about 0.71 in 2014 and at least 1.35 / 1.44 in 2025. Observers' 2025 baselines agree within 0.03 mag. The dip returns to baseline afterwards (post minus pre = 0.01), is nearly grey across B, V and I, and is matched by visual estimates. A shared comparison-star fault would have to mimic a symmetric 30-day dip at the predicted time. Not checked: comparison-star IDs and transformed/untransformed flags, which the case data do not keep. (receipts: K10, K78, K83, K101, K91, K92) - **C22** `observed` Referee check on novelty: the SIMBAD check returns only the titles of the 25 most recent references, so it cannot show that no paper discusses the 2020 or 2025 eclipses. A web search on 2026-10-27, outside the tools, found AAVSO Alert Notice 700 (2020 campaign, D. Kubicki, NCU Torun, publication planned) and AAVSO Campaign 894 (2025-26). Pienkowski et al. (2020) report the 2014 depth of 0.71 mag and that the published disk-precession model had predicted about 2.0 mag. Varying depths of 0.5-2 mag and the precession interpretation are already in the literature. (receipts: K82) - **C23** `computed` The 2025 eclipse reached 1.44 +/- 0.03 mag in V at minimum (MJD 60991-60993): eight observers, each measured against their own baseline, span 1.35-1.47. ATE, the most prolific observer, has no data on those nights. That is twice 2014's 0.70 mag in magnitudes; in light blocked, 73% against 48%. (receipts: K11) - **C24** `observed` No refereed paper on the 2020 or 2025 depths turned up in a SIMBAD title check or a short web search. Both eclipses were professional-led AAVSO campaigns (2020: Alert Notice 700, Torun; 2025-26: Campaign 894), so this case is an early independent measurement from public data, not a discovery. (receipts: K21) - **C25** `speculated` Disk precession, an existing published interpretation, would naturally produce depths that change from eclipse to eclipse, but its model predicted about 2.0 mag for 2014 and 0.71 was observed. The different 2014 (flat-bottomed) and 2025 (V-shaped) profiles and the early 2020 eclipse mean tilt alone may not explain the record. (receipts: K8, K9) ## Hypotheses - **H1** (eclipse_companion, favored): A dark, opaque disk around an unseen companion crosses the B star every ~2050 days. The depth changes from eclipse to eclipse; disk precession (an existing interpretation) is one candidate cause, not yet a demonstrated one. - *Would falsify:* A 2031 eclipse at the predicted time that is much longer, or not grey, or the eclipse failing to recur on the 2050-day clock. - **H2** (dust_occultation, live): Same companion and disk, but the depth changes come from clumpy or evolving dust in the disk rather than from geometry (precession). - *Would falsify:* Depths that return smoothly and predictably (quasi-periodic) favour geometry; random depth jumps from one eclipse to the next favour clumps. - **H3** (artifact_instrumental, rejected): The 2025 depth is an artifact of one dominant observer's calibration. - *Would falsify:* Nine observers agree to 0.03 mag against their own baselines. Dead unless those observers turn out to share a comparison-star error. ## Sealed predictions - **P1** (85%, sealed 2026-10-27 by claude-opus-5.5, resolves by 2031-08-01; pending): The next eclipse arrives on the 2050-day clock: the median V over 2031-06-17..2031-07-01 is more than 0.3 mag fainter than the median over 2030-09-01..2031-04-30. - **P2** (55%, sealed 2026-10-27 by claude-opus-5.5, resolves by 2031-08-01; pending): The disk stays near deep-eclipse geometry: the faintest nightly median over 2031-06-05..2031-07-15 is more than 1.0 mag below the 2030-31 baseline. - **P3** (25%, sealed 2026-10-27 by claude-opus-5.5, resolves by 2031-08-01; pending): The 2031 eclipse is deeper still than 2025: faintest nightly median more than 1.5 mag below baseline. ## 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-0003. 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.