List the cases you can investigate, with their status on the discovery ladder.

The Strange Stars explanation classes. Hypotheses must use one of these names.

Start here. What the case is, what data exist (per series: span, points, brightness range), the stellar context from Gaia, and the claims and hypotheses recorded so far.

Points of one series, e.g. series='NEOWISE:W1' or 'AAVSO:V'. Optional time window and binning (median per bin_days). Long series are binned automatically to max_points.

Summary statistics, long-term slope, and the single most significant change in level (a changepoint: when, how big, which direction) for one series.

Median brightness of a series in window A vs window B, with the significance of the difference.

Does every independent observer see the same thing? For each AAVSO observer (or camera) with data in both windows: their own baseline median, their faintest and median point in the event window, and the depth relative to their own baseline. Observer zero points cancel.

Period of short, deep, rare events (eclipses) that Lomb-Scargle misses: find the period at which every point more than depth mag below its series median falls at one phase, and no quiet point does. Combine instruments (e.g. ["ZTF:ZTF_r", "GAIA:G"]) for a long-baseline ephemeris.

Fold a series on a period: phase-binned medians, primary depth (phase 0) and secondary depth (phase 0.5) within +/- window_phase, the faintest bin away from the primary (a secondary at any phase, for eccentric orbits), and primary depth on even vs odd cycles (catches period halving).

Fit a gravitational microlensing event: a point source magnified by a point lens, without and (by default) with annual parallax, to one or more series at once (e.g. ["ATLAS:ATLAS_o", "ATLAS:ATLAS_c", "NEOWISE:W1"]). Each series is binned per night and turned into flux / its own baseline (median outside [start, end], or over the baseline window). Errors include that series' baseline night-to-night scatter and 1% of flux. Returns t0, tE, u0, A_max (and piE, fs), 95% ranges for t0 and tE from chi-square profiles (not covariance), how well the peak and wings are sampled, and how much the parallax model improves on the static one. A static fit much worse than the parallax fit means the light curve is asymmetric; a peak with no points means t0 and A_max rest on the wings. Microlensing is achromatic: fit the bands together and compare their residuals.

Lomb-Scargle periodogram with the survey's spectral window, flagging peaks that match Earth/spacecraft rhythms (day, lunar month, TESS orbit, year) or window aliases.

Find dips (fainter) and brightenings against a running baseline: start, end, depth, ingress/egress durations, asymmetry, and whether the start or end fell in a data gap. Restrict to a window (start/end) to get a local noise level, e.g. for quiet years.

How series A changes when series B changes (near-simultaneous pairs): correlation and slope dA/dB, with the interstellar-dust expectation for comparison.

NEOWISE W1-W2 colour per visit and whether it tracks W1 brightness (warm dust signature).

Infrared excess in NEOWISE over a baseline (photosphere) epoch: excess flux in each band as a fraction of the star, and the blackbody temperature of the excess from its W1/W2 ratio.

Gaia DR3 sources near the target: separation, brightness difference, and fraction of the target's flux. In blind mode, identities are removed.

Run the Skeptic: deterministic checks for blends, saturation, quality cuts, single-observer data, survey-rhythm aliases, unobserved onsets, TESS spacecraft systematics and more.

Record a claim. tag must be observed | computed | inferred | speculated. Cite the call_id(s) of the tool results it rests on (computed and observed claims without calls are weak).

Retract a claim you no longer stand behind (a mistake, a probe, a superseded number). It stays in the record, struck through, with your reason. Nothing in Oddsky is silently deleted.

Propose an explanation. klass from list_classes(). List claim ids that support it and that are in tension with it, and say what observation would falsify it.

Change a hypothesis's status (live | favored | disfavored | rejected) and confidence, with a reason.

Register a falsifiable, machine-checkable prediction. Example: 'the median of AAVSO:V between start and end will be more than 0.3 mag fainter (op '>', value 0.3) than its median over the baseline window'. Without a baseline, value is an absolute magnitude. probability = your stated chance the check comes out true. The registration time is sealed by the ledger and git.

Grade a registered prediction against the data now in the case. Predictions whose window has not closed yet are reported as pending and not graded. dry_run=True never writes. (Not available in blind trials, where the future is held back.)

Open a new case on any star by name (SIMBAD/AAVSO resolvable) and fetch its data from Gaia, NEOWISE, AAVSO, ZTF, ASAS-SN and TESS. Takes a minute or two.

Recent papers mentioning the target (SIMBAD bibliography). Open mode only. Read them after forming your own view, not before, if you want your investigation to count as independent.

Run the Librarian: is it new? For a sky position (ra, dec in degrees; or give a case to use its target), look up the objects in SIMBAD and NED, merge the papers SIMBAD and ADS attach to them, ask the big catalogues (VSX, VarWISE, Milliquas, WISE AGN) and check the shelves of published samples. verdict: unlisted (no object anywhere) | catalogued (an object, but no paper title matches the topics; not evidence that nothing is published) | mentioned (titles match, but only in papers that list more than 500 objects) | studied (a title matches in a paper about the object, a shelf holds it, SIMBAD types it as a transient, or a catalogue of variables lists it: read those before calling anything new). An AGN-catalogue entry is reported and does not change the verdict. complete = false means a source could not be queried, or ADS held more papers than were read, and the verdict is not final. topics: transient, dimming, infrared_excess, binary (default all). epoch: the date of the position as a decimal year (2010.5 for AllWISE); give it for anything that may be a nearby star, so that SIMBAD's stars are moved to that date by their proper motion (default: no motion). Open mode only. Logged to the ledger when a case is given.