History / Study 40 Indefinite Causal Order

Study 40: the cell run is measured — 9 of 9 cells and the Mac, byte-identical The page's last unmeasured sentence is gone. It said the program has no platform-dependent arithmetic, "checkable by reading it -- not that a cell has executed it". A cell has now executed it, and so have the other eight. Cross-compiled aarch64-swift-linux-musl static from the build host and run on all nine live cells (Debian 13 trixie, Linux 6.12.96, aarch64) plus macOS 27. All ten produce a BYTE-IDENTICAL transcript, sha256 10bea22dc61d19ae2cb07d85634f44f6c9b444 ee72682b6a3a757c24cc7a198c, carrying the same internal seal 7e5d40d5. Two operating systems, two C libraries -- musl static on the cells, Darwin on the host -- one source file, not one differing byte. That is the property the whole page rests on. Every other claim on it was already a number; this was the one that was argued, and the founder's instruction was that a study is not finished while it still carries one. The program is unchanged, so the seal and all 29 pins are untouched and still pass. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

rg78803 committed Sep 9, 2026

Restore Study 40 (reverted by 26e5d1f), rebased onto current origin 26e5d1f was built from a snapshot predating several pushes and carried Study 40 back to its first draft: the page from 16,650 to 15,858 bytes and the program from 68,422 to 25,886. It clobbered at least three sessions -- Study 41 and its program were deleted outright, and c60cddc was a third session repairing its own concurrent overwrite. The live defect this left was NOT the one first reported. The page and the served program were mutually CONSISTENT: the page cited transcript seal ecdcb1ce, and the 25,886-byte v1 program does produce ecdcb1ce. bb9dfe37 is that FILE sha256, a different object from the transcript seal, so "the served program matches neither seal" was a false alarm. The real inconsistency was one level down: validate.sh at origin still pinned seal 7e5d40d5 -- 29 pins, all of which survived -- against a page that no longer contained it, so the harness was pinning figures the published page did not carry. Restored from 3b20e3c / 25190c5: the page, the program, the white-paper row 40 title (the index row had survived, so index and white paper disagreed about the study title), and the sidebar line. Rebased onto whatever origin held at this second rather than onto my last local state, and built HEAD-plus-only-my-lines through a temporary index so nothing of Study 41 or of the third session rides along. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

rg78803 committed Sep 8, 2026

The library reads in order now, and a quoted figure stopped counting as evidence Cures first, as a five-step reading order. Twelve unreachable pages brought back into the sidebar. The front door carries seven medicines and both registry-wide maps. The registry-specificity slot named empty on 2026-09-07 is entry 7, seal 2d74f5c6. 39 new pins on The-Order-Of-The-Bases, checkable with no corpus. And the defect those pins exposed: the admission law's E4 and E5 were crediting figures a run had QUOTED rather than computed, so a screen that ran nothing was graded MEASURED against its own published seal. Both clauses test for a refusal first now; three arms hold the repair, including the reverse control. 65/65 arms. Libraries: 18 entries, 14 admitted, 4 held, 0 refused, seal b9b2ac76. This clone now carries library/ and the admission law, so the wiki can grade its own libraries — it never could. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

rg78803 committed Sep 8, 2026

Study 40: close every escape from the float result, and say why the correct way matters more The page had the measurements and not the argument. Three things were missing and all three are now measured or stated. 1. WHY THE CORRECT WAY MATTERS MORE THAN THE WRONG WAY THAT GOT THE ARTICLE. The bench found this effect because nature computes exactly -- there is no floating point in an interferometer. A simulation does not report an effect below its horizon as uncertain or marginal. It reports it as ABSENT, in the same clean 0.0 it would return if the effect did not exist. You cannot survey for what your instrument returns as zero. So the published result is one effect that happened to be reachable on a bench, and an unknown number of others a float-based search retires as null before anyone builds the apparatus. 2. NO WIGGLE. New ARM 10 measures the four defences shut rather than arguing them: - "more bits": half fails at z = 1 - 10^-4, single 10^-8, double 10^-16. Thirteen more mantissa bits bought 4 decades, twenty-nine more bought 8. Nothing in that sequence terminates; exact integers have no horizon at any rung. - "safe regime": failures are non-monotone (10^5 and 10^7 clean, 10^4/10^6/10^8 not). - "rescale": the loss is cancellation, not magnitude. At the crossing the double holds ONE number, 0.5, for two rationals differing by exactly 1/19999999999999999. The asymmetry carrying the effect is not approximated. It is absent. - "just rounding error": it fails in BOTH directions and returns an exact 0. Plus one the defence does not think to make: the horizon is not a property of the value. The same z spelled two ways lands on different floats and one of them still sees the effect at k=16. 3. THE THREE ASSUMPTIONS, NAMED TOGETHER. Definite causal order (withdrawn on the bench); finite floating point as a carrier of real quantities (measured to fail at every width, both directions, non-monotonically); and a total order over events that exists to be agreed on (nine origins, nine sequences, one orientation). Each leg measured to fail on its own terms. And it is seated in the programme rather than presented in a vacuum: the same claim as Studies 34/35/36/38/39, reaching physics for the first time -- with 38/39 named explicitly because there the two arithmetics AGREED to fourteen digits and the board published that as the finding. The instrument is not tuned to indict float. Two defects of my own found while building ARM 10 and fixed: the rescale paragraph asserted the populations were still distinct while the line beneath it printed them equal, and ARM 10 first disagreed with ARM 9 by a decade because the two built z differently -- now aligned, with the disagreement kept as its own measurement. New seal 7e5d40d5. 29 pins in validate.sh, 7 of them control arms. 75 of 75 exact figures independently re-derived, 0 divergences. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

rg78803 committed Sep 8, 2026

Study 40 rewritten to lead with the finding, not the rubric Same measurements, same seal 7f14484b, 40% fewer words. The previous version opened with a grading rubric and carried a grade banner into every section; the numbers were in there, buried under scaffolding. Nobody drops what they are doing to read a compliance document. It now opens with the thing a physicist can act on: their prediction is a fraction, we computed it (-1/18 at z = 1/2, 9 of 9 settings against 0 of 9), and computing the same fraction in double precision returns 0 above z = 1 - 10^-16 -- not small, zero, on 7 of 22 rungs. Then a section titled for the reader rather than for us: three places where the arithmetic and not the physics is deciding what you find, each one a defect that can be sitting in working code today. Grades are kept where they belong -- one word per row in the evidence table -- instead of a banner per section. Every limit is kept and stated in place as a sentence: we did not do the experiment, this does not make quantum hardware unnecessary, we did not identify the geometry of the universe, and no cell has executed the program. All 16 page-pinned figures verified present after the cut. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

rg78803 committed Sep 8, 2026

Study 40 deepened: run their ICO experiment in whole numbers, and measure where a double deletes it The first version measured that order-dependence is a property of the arithmetic. That was true and it was not the claim. This version makes the claim. WE RAN THEIR EXPERIMENT. The quantum switch of two thermalising channels (Felce & Vedral 2020, realised photonically by Xue et al., PRL 2026) computed EXACTLY: at 9 of 9 reservoir settings the switch moves energy that NEITHER definite order moves (0 of 9), exactly -1/18 at z = 1/2. No qubit, no interferometer, no shot noise, no visibility, no error bar -- nothing was sampled, the state was carried as integers and read off. Control at z = 1 (infinite temperature) moves exactly 0, so the arm is not always-red. Cold system against two hot reservoirs lands below both definite orders by exactly 16336650/166148483. AND WHERE A DOUBLE LOSES IT. The same physics in double precision returns EXACTLY ZERO -- no effect at all, no flag, no warning -- on 7 of 22 rungs, first at z = 1 - 10^-16, where the exact arithmetic returns a non-zero fraction on all 22. THE MODEL CLAIM THE FIRST VERSION BURIED. Computation imported its causality from relativity via Lamport (1978), whose own paper says the total extension is ARBITRARY. Two new arms measure what is left of that premise: the commutator is HOLONOMY, and a float-walked closed loop invents curvature where the manifold is flat (4 of 6) and erases it where it is curved (1 of 6), non-monotone, so there is no scale to stay under; and on a compact phase coordinate there is no total order to find at all -- nine origins give 9 sequences from one unchanged event set while the cyclic orientation over 84 triples is 1 vector under all nine, against 1 sequence on a line. Int128 REMOVED. The build host has it and the cells do not, and one law in two types is two laws. Exact integers are now unbounded decimal-string-backed, which is also why ARM 9 reaches 10^-22: under a fixed width this program trapped at the third rung of its own ladder while every answer was small, and a ceiling inside an instrument that measures where floating point runs out is the same defect wearing a different width. Control: the rewrite reproduced every earlier arm byte-for-byte. 75 of 75 exact figures re-derived by an independent arbitrary-precision implementation, 0 divergences, its own two control arms correctly absent. Marker ORDER_IS_AN_ARTEFACT_OF_THE_ARITHMETIC, seal 7f14484b. 24 figures pinned in validate.sh, 8 of them control arms proving each detector fires. NOT KNOWN, stated on the page: this does not make quantum hardware unnecessary -- a two-level system under two fully thermalising channels is classically writable, which is why it can be a fraction, and nothing here is evidence either way about a state space that is not. Nor does it single out one manifold. And a cell execution is not yet measured: the Linux host here carries no Swift toolchain, so the claim is that the program has no platform-dependent arithmetic, not that a cell ran it. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

rg78803 committed Sep 8, 2026

Study 40: order-dependence is a property of the arithmetic, not of the events Nine cells fold one identical 64-event set in nine arrival orders and reach 1 distinct result in exact rationals, 6 in double — a spread of 84% of the true value from identical input. Asked whether two operations commute, a double is wrong in both directions: it manufactures an order-dependence that is exactly zero on 3 of 9 exactly-commuting pairs, and erases a real one on 3 of 9 that never commute. Non-monotone both ways — failures at 10^4/10^8/10^10 with clean rungs between — so there is no scale you can stay under. The exact arm answers the closed form 18 of 18 and separates both populations; a second control on a population float can hold exactly returns 1/1, so ARM 4 is measuring rounding and not the permutation generator. Marker ORDER_IS_AN_ARTEFACT_OF_THE_ARITHMETIC, seal ecdcb1ce. No corpus, no network, no key, no argv — every constant in the program, reference figures on every exit path, 14 registered in validate.sh. First page on the board placed beside an outside physics result (Xue et al., PRL 2026, photonic quantum switch). It is graded apart on its face: the physics is REPORTED with its own limits stated — proof-of-principle, and the second law is not bypassed, the control qubit coherence is the resource — the mapping onto this architecture is ARGUMENT, and universe-geometry is NOT KNOWN. Index row 40, sidebar, readers guide reach line, White Paper v4.5 (Date/Version header had drifted at 4.2 while doc-control carried 4.4; corrected, not carried). Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

rg78803 committed Sep 8, 2026