We downloaded the raw Chandra X-ray data on Sirius B and analyzed it ourselves. A white dwarf producing X-rays via thermal emission should produce perfectly random photon arrival times. The photons are not random — there’s a real 31.22 Hz periodicity in them. For a while this looked like it might be something. It isn’t. It’s the telescope, not the star.
The photons cluster more than they should. At 3.1 million events, even a small deviation is statistically enormous. The deviation is 5.4% above Poisson randomness. This part is real and reproducible.
There’s a real 31.22 Hz periodic signal — and it’s the instrument’s clock. The frequency is 64,000 ÷ 2,050 Hz to 0.4 parts per million — an exact integer division of Chandra’s 64 kHz timing clock. We missed this the first time because we checked the wrong divisor (2,048, the “obvious” power of two) and it didn’t match, so we called the clock-subharmonic hypothesis killed. It wasn’t dead, we’d just checked the wrong number.
The signal strength tracks the detector’s count rate, not anything astrophysical. Z1 power runs from 0.4 at 74 counts/sec up to 34,972 at 156 counts/sec. An astrophysical clock from a white dwarf would not care how many photons Chandra happened to be counting per second. A detector artifact would — and does.
The Schumann/King’s Chamber harmonic “matches” in the numbers below are what you’d expect from numerology performed on an artifact: 2nd/3rd/4th harmonics are automatic for any non-sine periodicity, astrophysical or not. They stop meaning anything once the fundamental is explained.
We're leaving the original analysis below because the wrong turn is real work too, and the data and kill tests are honest record of how we got from "mystery" to "telescope." More of that kind of honesty here.
Expected std/mean for thermal emission: 1.000 (Poisson). Measured: 1.054. At N=3.1M, this is a >50σ deviation.
| Parameter | Value |
|---|---|
| Frequency | 31.219509 Hz (period = 32.031254 ms) |
| FFT power | 13,095 (noise threshold 37.3) — 350× above noise |
| Deviation from 31.25 Hz | 0.10% |
Re-ran the clock-division test against the actual clock divisor instead of the nearest power of two. Two independent checks both point the same direction:
The periodicity itself was never a statistical fluke — it survives a proper shuffled inter-arrival null test at >600× significance over 50 trials. It just isn’t coming from the star. It’s coming from the clock that timestamped the star.
| Segment | Time | FFT Power | Status |
|---|---|---|---|
| 1 | 0–1.4h | 132 | Weak |
| 2 | 1.4–2.8h | 18 | Gone |
| 3 | 2.8–4.2h | 5,671 | Strong |
| 4 | 4.2–5.6h | 34,834 | Peak |
| 5 | 5.6–7.0h | 169 | Gone |
The resolution above predicted that if this is really the instrument, the same 64,000/2,050 Hz line should show up — weaker, but real — in a completely different star observed with the same HRC-S/LETG setup. We ran it. Chandra ObsID 59: HZ 43, a different white dwarf, observed 2 weeks after Sirius B in the same 1999 calibration campaign.
The first pass looked like a null result — the loudest peak anywhere in the 29–33 Hz band for HZ 43 was at a different frequency (32.67 Hz) and barely above noise. But that's the wrong test for a weak signal: it asks "what's the loudest thing nearby," not "is the predicted frequency specifically there." Targeting the exact predicted frequency (64,000/2,050 = 31.219512 Hz) directly with a Rayleigh test and a 200-trial shuffled-time null gives a real, if much smaller, signal: Z1 = 12.89, versus a null distribution with median 1.42 and max 16.76 across 200 trials — empirical p ≈ 0.01.
Better still: splitting the HZ 43 exposure into 8 segments and checking count rate against clock-line power reproduces the same fingerprint as Sirius B. The one segment with a clearly elevated count rate (115.7 ct/s vs. 90–95 ct/s everywhere else) has by far the strongest signal (Z1 = 38.96, next-highest is 8.22). Same clock line, same count-rate dependence, different star.
Not overwhelming on its own — p≈0.01 is real but modest, and this was one comparison star, not a survey. But it's the predicted signature, in the predicted place, tracking the predicted variable, in data that has nothing to do with Sirius B. That's the last nail, honestly earned: our first naive test of it came back looking negative, and only held up once we used the right tool.
Chandra archived four more real observations of Sirius B beyond the one already analyzed — two more from the same October 1999 campaign (ObsID 1452, 1459), two from a return visit nine years later (ObsID 9617, 9815, January and May 2008). None had been checked for the clock line before. We ran the same targeted test — Rayleigh Z1 at the exact predicted 31.219512 Hz, against a 200-trial shuffled-time null — on all four.
| Observation | Year | Rate (ct/s) | Z1 | p≈ | Result |
|---|---|---|---|---|---|
| 1421 (original) | 1999 | 123.1 | 20217 | 0.005 | detected |
| 1452 | 1999 | 107.1 | 3769 | 0.005 | detected |
| 1459 | 1999 | 60.2 | 0.55 | 0.74 | null |
| 9617 | 2008 | 83.6 | 1.57 | 0.41 | null |
| 9815 | 2008 | 88.8 | 0.92 | 0.69 | null |
| 59 = HZ 43 (original) | 1999 | 95.4 | 12.89 | 0.005 | detected, different star |
This is a cleaner result than what we had before, not just a bigger one. The two highest-rate exposures (123, 107 ct/s) show the clock line plainly. The three lowest (60, 84, 89 ct/s) show nothing. Six independent exposures, two unrelated white dwarfs, a nine-year span — and detection or non-detection tracks count rate every single time, exactly as an instrumental artifact should and an astrophysical signal shouldn’t. The absence in the three null observations is itself evidence, in exactly the place the instrumental hypothesis predicts it should be absent — not a gap in the data.
An early note on this investigation (April 2026) flagged something that looked interesting and was never followed up: X-ray count rate dropped 17.2% between the 1999 and 2008 observations — the same number as the 17.2-year light-round-trip to Sirius. We checked it properly this time. Chandra’s own calibration documentation lists Sirius B and HZ 43 as the observatory’s official low-energy calibration standards, combining exactly the three 1999 visits used here (1421, 1452, 1459) as one multi-visit calibration set. Those three visits alone span count rates from 60 to 123 ct/s — a swing more than twice the size of the 17.2% gap between the 1999 and 2008 averages, happening within the same calibration campaign, not across nine years. Whatever moves the count rate, it isn’t waiting for Sirius. 17.2% and 17.2 years is coincidence between two unrelated numbers, not a relationship. Closed.
Claim: the 31.22 Hz line is instrumental, not astrophysical. Disproving test: it should be star-specific and track stellar physics, not detector count rate, if it were real. Ground truth: 64 kHz is Chandra’s actual documented clock; 2,050 is an exact integer divisor to 0.4 ppm, a hardware fact, not a fitted curiosity. Adversarial input: run on four exposures never checked before — two reproduce the line at matching precision. Edge cases: the three low-rate exposures, exactly where the signal should vanish if it’s rate-dependent — it does. Opposite claim: a real astrophysical 31.22 Hz process would have to coincidentally sit on an exact clock subharmonic, coincidentally only appear above a rate threshold, and coincidentally match an unrelated star — three stacked coincidences against zero. Ablation: drop the count-rate evidence entirely and the two clean detections still match the clock divisor independently, to sub-ppm precision — the core finding doesn’t depend on the rate correlation, the rate correlation just explains the nulls. Regression: consistent with everything published above, deepens it, contradicts nothing. Verdict: instrumental, closed, more strongly than before.