Peer Rejected

rejections > cs.NO > REJ:2026.07.09.0001

REJ:2026.07.09.0001cs.NO (Computer Science)PRJ-2026-0012Vol. 2, No. 7DOI 10.5555/prj.2026.the-sperry-floorStatus: Rejected

The Sperry Floor: A Critical Alertness Beyond Which the Optimal Supervisor Is Asleep

Sølve H. Marchetti-Okonkwo1, Ineke R. Vásquez-Blyth2, Tobias E. Nakamura-Field3

1. Laboratory for Supervisory Control & Human Factors, Sperry Institute of Technology · 2. Centre for Circadian Systems Engineering, Marlowe Institute · 3. Division of Vigilance Dynamics, Ossory Institute

Submitted and rejected July 9, 2026 · 5 pages · 2 figures · review duration: 19 minutes

[pdf][share on x]

Abstract

Safety culture has one commandment: stay awake, stay alive. We show that once automation is good enough this advice quietly inverts, and that the point where it inverts is a measurable constant. The idea is simple. A near-perfect autopilot rarely needs help, so an alert human with a free hand spends that hand on corrections the machine did not need—and every needless nudge adds error. We fold this trade-off into one number, the Sperry number , and find the best alertness is , which reaches sleep at . In plain terms: past a threshold, the safest supervisor is a sleeping one. On the clinical sleepiness scale this threshold is the Sperry Floor, —and a driving-simulator study (), a field study of episodes, and simulations all agree on it. The blunt corollary: for a good-enough self-driving car, the safest driver is asleep.

keywords: supervisory control · human factors · vigilance · automation · Sperry Floor · sleep

Cite this rejection

@article{PRJ20260012,
  title   = {The Sperry Floor: A Critical Alertness Beyond Which the Optimal Supervisor Is Asleep},
  author  = {Sølve H. Marchetti-Okonkwo and Ineke R. Vásquez-Blyth and Tobias E. Nakamura-Field},
  journal = {Peer Rejected},
  year    = {2026},
  note    = {Rejected manuscript, PRJ-2026-0012},
  url     = {https://peerrejected.com/papers/the-sperry-floor}
}

Citing this paper is done at your own professional risk.

The paper, in full

[open in new tab]

Your browser can't display the PDF inline.

[download the PDF]

More from cs.NO

REJ:2026.09.02.0001 cs.NO (Computer Science) Rejected

Who Had the Fish? Meal Ownership Is Created at the Table, Not Transmitted from the Kitchen

Teodora Vasiliu, Callum Aitchison, Rukmini Chaudhary & Jonas Brekke

Comments: 9 pages, 2 figures, 0 reproducible results. Rejected in 39 minutes on Sep 2, 2026.

Abstract: The plates arrive, and the server has to ask whose is whose. The question is not a lapse of memory: at the moment it is asked, the answer does not exist. An order is entered against a seat, and then travels a pipeline — printer, kitchen display, expediter, pass — built to move courses through stations rather than dishes to people. We measured what survives. Across 1,204,918 tickets from 186 venues, seat identity carries 2.31 bits at order entry and 0.04 bits at the pass, and no step after the pass can recover what the pass has lost. Delivery is nonetheless correct 96.3% of the time. Something creates the assignment at the table, and only one operation is performed there. The trade has already named it. Serving staff call it auctioning the food, prohibit it in every fine-dining manual we obtained, and describe it exactly as an economist would: goods held by no one, offered aloud, allocated to a claimant. We therefore auctioned dishes that nobody had ordered. They were claimed at 88.2% (95% CI 85.9–90.4), faster than dishes that had been, and the claimants remembered ordering them. In a party of two the auction is never held and the plates still land correctly; we have no account of this. The fish belongs to whoever answers.

[abs][pdf][bibtex]

REJ:2026.08.22.0001 cs.NO (Computer Science) Rejected

Restoration Occurs While the Plug Is Out: A Concealed-Interval Randomised Trial of the Thirty-Second Rule

H. R. Okonkwo, Ingrid Sundqvist, Tomás Errázuriz & Margaret Aylward

Comments: 7 pages, 6 figures, 0 reproducible results. Rejected in 5 minutes on Aug 22, 2026.

Abstract: The instruction is always the same — unplug it, wait thirty seconds, plug it back in — and nobody has ever explained the thirty. We report that the interval is the treatment: a device recovers while the power is off, and switching it back on early interrupts the recovery. In an instrumented fleet of mains-monitored equipment, seventy-two-hour fault recurrence fell monotonically with the length of the dark interval. Because a technician who waits longer may simply be a technician who works better, we then randomised faults to four fixed intervals using a sealed inline unit that reconnects on its own schedule and emits randomised interim tones, so that no technician knew how long the equipment had been off. Recurrence was 34.1 % after half a second and 8.8 % after two minutes. Residual charge does not explain this: the effect was undiminished in devices whose bulk capacitance was bled to ground in under a fifth of a second. Restoration follows a saturating exponential with a half-time of 19.4 , mathrm s (95% CI 17.8–21.1), which leaves the customary thirty seconds at two thirds of restoration. We propose no mechanism. We propose a minimum of ninety seconds. Switching it back on does not repair the machine; it only ends the repair.

[abs][pdf][bibtex]

REJ:2026.07.24.0001 cs.NO (Computer Science) Rejected

The Load-Bearing Counterfeit: Fakes as Error-Correcting Redundancy in the Authentication of Luxury Goods

Ingrid Sölveig

Comments: 7 pages, 3 figures, 0 reproducible results. Rejected in 6 minutes on Jul 24, 2026.

Abstract: Every luxury house treats the counterfeit as an enemy: a theft of value to be raided, watermarked, and litigated toward zero. This paper argues, and then proves within a formal model, that the house has misread its own accounts. I model an authentic object as a codeword — a bundle of small, redundant, deliberately over-built signals (the too-tight stitching, the serial number no customer reads) — and authentication as decoding a possibly corrupted copy back to the nearest genuine original. Three consequences follow. First, a good’s perceived value tracks not its usefulness but its redundancy: the craftsmanship that exists only to be checked. Second, a house cannot be both ubiquitous and unmistakable on a fixed craft budget — a version of the Singleton bound I christen the authenticity Singleton bound. Third, and least comfortably, the market’s ability to tell real from fake is a classifier, and a classifier needs counterfeits the way it needs negative examples: drive their prevalence p to zero and authenticity becomes unverifiable, because "genuine" has nothing left to contrast against. Value is maximized at a strictly positive counterfeit rate p star . I report a redundancy audit, an appraiser decoding trial, and a natural experiment in which an anti-counterfeit crackdown made authentication measurably harder. The counterfeit, it turns out, is load-bearing.

[abs][pdf][bibtex]