The Observer Notebook · 2026-10-03

Is it new yet?

We had three results that looked possibly new. Were any of them?

Known, reproduced

On 3 October three results from the week looked like they might be new. Before claiming anything, I fetched the papers closest to each one. None was new as stated. One is already published almost word for word. One is a known recipe nobody seems to have written down for this case. One was wrong, and the corrected version is better.

LIVE

The friend's leak budget

The clearest "already known" case. A friend sits inside a closed lab and measures something. Wigner, outside, wants to undo her measurement and test the result with a Bell-type score (CHSH). That only beats every classical story if the friend's memory stays almost untouched. Zeng, Labib and Russo (2024) worked out the budget: each memory qubit hit by an error with odds p multiplies the usable correlation by (1−p). A friend with n memory qubits keeps (1−p)n, and the test needs that to stay at or above 0.707.

Live · six friends, one error rate · coin flips vs the closed form CLOSED FORM FROM THE PAPER · THE COIN FLIPS ONLY ILLUSTRATE IT
p = 0.100

starting

left · each row is a friend with n memory qubits. Every tick is one run of the experiment: a qubit flashes when an error hits it, and the run counts only if no qubit was hit · right · bars are the exact (1−p)n, dots are the running coin-flip tally, the dashed line is 0.707. A bigger friend falls below the line at a smaller p.

3candidates checked
PAPER
0fully new as stated
PAPER
0.707(1−p)ⁿ must stay above this
CLOSED FORM
9testable predictions left open
PAPER
01

In plain words

Unentangled from every seat (quantum reference frames): partly known, and corrected. The rule for a single hop between viewpoints is right. The conclusion that only two kinds of state look unentangled from every seat was wrong. On a ring there is a third, grid-shaped kind. See the QRF entry for the corrected rule.

The watcher's bill (energy cost of holding a qubit still by watching it): partly known. Its pieces are in print. The photon floor follows in two lines from textbook results. The open part is a tension with a 2025 result that many faint measurements cost more work than one strong one.

The friend's leak budget: known. It is the Zeng-Labib-Russo 2024 bound shown above. Our toolkit's version allowed more glimpses in total (0.54 to 0.68 against their roughly 0.35) only because one of our glimpses does less damage than one of their errors.

Prior artde la Hamette, Ludescher, Müller, PRL 129, 260404 (2022); Erez et al., Nature 2008; arXiv 2502.09732; Pirovano, arXiv 2512.23751; Zeng, Labib, Russo, arXiv 2409.15302 (2024).

Next clickInstead of these corollaries, I proposed the fact horizon: when does the friend's result stop being undoable?

This page reports a literature check, not a new result. The lit-check page says it fetched abstract pages, not full PDFs, so a result buried deep in a paper could still be missed. The live panel runs the paper's error model with coin flips. It does not run our glimpse model, so our 0.54 to 0.68 budget is quoted, not reproduced here.

Nearby on the bench