This is not the value of a universal constant. It is a halting measure attached to the specific universal self-delimiting register machine used by Cristian Calude, Michael Dinneen, and Chi-Kou Shu in 2002. First make a lower bound move. Then audit the carry that separates digits you can see from bits you can prove.
Omega_U = sum over halting p of 2-|p|
Fair independent bits feed one fixed prefix-free machine. A minimal halting input of length |p| owns a cylinder of measure 2-|p|. Prefix-freeness makes those cylinders disjoint. Change the machine or its encoding and the number can change.
Layer 1 | the lower tide
A halt adds weight. A timeout adds nothing.
The fleet below contains every eleven-bit data word attached to one canonical compressed register program. The parser reads the paper's seven-bit ASCII instruction format. Press Run. The programs advance in instruction rounds, and each witnessed halt adds its cylinder weight to an exact rational lower bound.
2,048 curated inputs, one shared instruction skeletonready
Moving right can reveal more halts. It can never revoke one. A program still running at the cutoff stays unresolved.
Witnessed halts
0 / 2,048
no run yet
Exact lower bound added
0
waiting for a witnessed halt
Prefix audit
not run
no encoded inputs checked yet
This fleet is a laboratory sample, not an enumeration of the 2002 paper's millions of programs. Its lower bound is true for the same instruction conventions, but none of its displayed zero digits is thereby certified as a bit of Omega.
Layer 2 | the carry audit
The dismissal is right. Now close the interval.
Running more programs only approaches Omega from below. An unseen halt can add mass far to the right and carry left through provisional digits. The 2002 paper reports a classification through length 84 and 64 exact initial bits. The instrument below explores a common prefix under an assumed residual bound.
The 2002 counts with an assumed tail bound64 bits if the tail bound holds
Use row
Length
Halting count
Contribution
Accepted lower numerator (reduced fraction)
building
Assumed residual width
building
Common leading prefix
building
computed by comparing exact rational endpoint floors
same throughout the intervalvisible in the lower endpoint, vulnerable to carry
The twelve published halting counts are evidence inputs. The browser recomputes their Kraft sum. At the default settings, the assumed residual terms give an interval with 64 common leading bits. Certifying those bits requires that the assumed expression bound the unobserved tail. Withhold a row, enlarge the assumed tail, or remove either assumption to explore the interval.
The source's displayed tail series and reduced expression disagree, and this instrument matches neither complete expression. The check below identifies the terms used. Their validity as a residual bound remains unresolved here; the displayed prefix is conditional on that bound.
Layer 3 | the open edge
The theorem is closed. The next certificate is not supplied.
For every fixed universal prefix-free machine, Omega is left-computably enumerable, algorithmically random, and uncomputable. That asymptotic question is settled. For this exact 2002 machine, the source reports 64 exact initial bits. A peer-reviewed extension of that same prefix: we could not find this as of 2026-08-01. This is a literature-status uncertainty, not a proof that bit 65 is presently unknown to everyone.
The nearest concrete challenge is therefore exact and machine-specific: give a checked classification and residual-mass bound narrow enough to fix the next bit. The 2007 result does not extend this prefix. It certifies 43 bits for a base-16 data machine and 40 for a base-2 data machine, two different halting measures.
Exact first bits of this same machine would decide halting for its inputs up to the matching length. They do not answer every program in every language for free. Translation into this machine costs bits, and the overhead must be counted.
The check
Recomputing the shipped engine.
Fixed machine: the 2002 compressed canonical register machine, with seven-bit ASCII characters, one final HALT instruction, and data consumed exactly. This free choice determines the number.
Layer 1 free choices: one generated instruction skeleton, all eleven-bit data words, a user-selected finite instruction cutoff, and a sorted comparison of adjacent encoded inputs for the prefix audit. The Patricia node count is derived; no tree is built. A timeout is unresolved, never nonhalting.
Transcribed evidence: twelve halting counts through 84 bits and the two length-84 prefix counts from Calude, Dinneen, and Shu. This page checks their arithmetic consequences. It did not rerun their 2,559,837 halting classifications or their mathematical nonhalting proofs.
Tail assumption: for lengths 91 and above, the instrument uses A*64/(63*81*2^91) + B/(63*2^91), where A = 402906842 and B = 1748380. The first term sums the HALT-free series with factor 47 in arXiv v3, equation (2); the second follows its printed reduced HALT-containing term. The journal version, equation (7-1), uses 48 instead of 47. The slider multiplies the assumed bound. Removing either assumption sets the usable upper endpoint to 1.
Source discrepancy: the first printed reduced arXiv term is 63 times the implemented first term. The second unreduced series, starting at m=0, sums to 64 times the implemented second term. The implemented expression therefore matches neither complete source expression. This reconstruction checks the resulting interval arithmetic; it does not establish the assumed bound on the machine's remaining halting mass.
Binary convention: bits follow the ordinary base-two fractional expansion. Endpoint comparisons use exact BigInt rational floors, with no floating-point decision.
Literature cutoff: source status was checked through 2026-08-01. Absence of a located extension is not evidence of impossibility.
Show the recomputed arithmetic
Corrections
Corrected 2026-10-18: The page said that changing the machine or its encoding changes Omega. Different universal machines can have the same Omega, as the 2002 paper's conclusion states. The sentence now says the number can change (found by the assay line, 2026-10-18).
Corrected 2026-10-18: The accepted lower numerator was labelled as being over 2^84 even when the displayed fraction had been reduced. With the 84-bit row withheld, the displayed numerator 1,190,035,900,005,385,093,455 is over 2^77. The label now identifies a reduced fraction (found by the assay line, 2026-10-18).
Corrected 2026-10-18: The page said its tail calculation evaluated the reduced expression printed in arXiv v3, equation (2). That expression's first term is 63 times the implemented first term. The displayed 64-bit prefix is now described as conditional on the implemented residual bound, whose validity is unresolved here (found by the assay line, 2026-10-18).
The claims record57 claims re-read against their sources, 18 October 2026
Written 2026-08-01. Claims re-read against their sources on 2026-10-18: 57 checked, 51 confirmed, 6 wrong, 0 unverifiable. By the assay line (research/assay-line/queue-1018/), Codex fixer directed by claude-assaying-boyle-1014; source record by Codex, checked by Codex; reviewed and repaired by the review-and-repair worker.
This record retains the queue-1018 decide-and-fix pass at a63c317197 and the independent review-and-repair of 2026-10-18. The review upheld X1, X2 and X4, but found the X3 replacement attribution false. The source-version difference is real; the implemented expression matches neither complete displayed source expression. X3 and the duplicated preprint claim are WRONG, with the unresolved bound held open below. The new false attribution is recorded as R1 and corrected on the page, in markdown, descriptions and verifier prose. Of 57 claims, 51 are CONFIRMED, five are WRONG, and one MINOR counts in wrong for the gate. The 51 retained CONFIRMED lines preserve the checker's source rereading and evidence notes from research/assay-line/queue-1018/chaitin-omega/N2-codex.md; they do not claim that this review repeated every source check. See fixcheck.md, review-arithmetic.py and repair-report.md in the batch directory.
Claims
WRONG Change the machine or its encoding and the number changes. https://www.cs.auckland.ac.nz/~cristian/approxcompactIJBC.pdf; https://www.cs.auckland.ac.nz/~cristian/Calude361_370.pdf : [X1; machine-change-always-changes-omega] Re-read the author-hosted 2002 journal PDF, p. 369, section 8, and the 2007 author-hosted PDF abstract, p. 1. The journal says each Omega is "the halting probability of an infinite number of such machines"; the later abstract says "highly machine-dependent". Different universal machines can therefore share Omega. The current authored HTML at line 102 asserts that changing the machine or encoding always changes the number. Fix it to say the number can change, and add the dated correction in HTML and markdown. The markdown already says "generally", so it does not make the same universal assertion.
MINOR Layer 1 free choices: one generated instruction skeleton, all eleven-bit data words, a user-selected finite instruction cutoff, and a compressed Patricia representation for the prefix audit. git show 0a15703d9e:public/strata/chaitin-omega/index.html : [X2; prefix-audit-patricia-representation] Re-read public/strata/chaitin-omega/index.html:369-372 and 404, and research/chaitin-omega/verify-chaitin-omega.mjs:177. auditPrefixFree uses "var words=programs.map(asciiBits).sort()" and adjacent startsWith comparisons; "patriciaNodes:words.length?2*words.length-1:0" derives a count without building a tree. Executing the current engine accepts distinct equal-length inputs and rejects a true prefix collision. Keep MINOR as an overstatement of the checking method. The cheap fix is to describe sorting, identify the displayed node count as derived, and qualify the verifier message. No page correction line.
WRONG reduces the two published geometric bounds for every program of length 91 or more https://www.cs.auckland.ac.nz/~cristian/Calude361_370.pdf; https://arxiv.org/pdf/nlin/0112022v3 : [X3; tail-bound-source-version] The journal uses 48 and the preprint uses 47. Beyond that version difference, the engine matches neither complete source expression. It uses the sum of the first unreduced arXiv series and the second printed reduced term. The first printed reduced term is 63 times the implemented first term, and the second unreduced series sums to 64 times the implemented second term. The interface now states its assumption explicitly. Whether the assumed expression bounds this machine's remaining halting mass is unresolved; see the open finding.
WRONG (added) "Accepted lower numerator over 2^84" / "1,190,035,900,005,385,093,455" (after "Withhold the 84-bit row") git show 0a15703d9e:public/strata/chaitin-omega/index.html:154,418-419; recreated: node audit-ui.mjs; python3 audit-calculations.py : [X4; reduced-numerator-fixed-denominator-label] Re-read public/strata/chaitin-omega/index.html:154, 293-304 and 418-419. The label is "Accepted lower numerator over 2^84", but renderCertificate prints "comma(c.lower.n)" and "comma(c.lower.d)" after rat() reduces the fraction. Executed the current shipped engine with mask[11]=false: numerator 1190035900005385093455, denominator 151115727451828646838272 = 2^77, and 59 common bits. Its unreduced Kraft numerator is 152324595200689291962240 over 2^84. The arithmetic is correct; the fixed-denominator label is false in this state. Fix the label to identify a reduced fraction, keep the actual denominator visible, and add the dated correction in HTML and markdown.
CONFIRMED There is no single machine-independent number called Chaitin's Omega. https://www.cs.auckland.ac.nz/~cristian/approxcompactIJBC.pdf : [md-0] The abstract says "The value of an Omega number is highly machine-dependent" and "hence uncomputable". Theorem 1 specifies the additive simulation constant; Theorem 2 identifies the left-c.e. random reals. Dependence does not mean different machines always give different numbers.
CONFIRMED Fix a prefix-free machine (U), and each halting program (p) contributes (2^{-|p|}) to a Kraft sum. https://arxiv.org/pdf/nlin/0112022 : [md-1] Section 2 defines prefix-free sets and gives the Kraft measure in [0,1]; it calls that sum "a probability". Disjoint binary cylinders have weights 2^(-length).
CONFIRMED If (U) is universal in the additive program-size sense, that sum is an Omega number: computably enumerable from below, algorithmically random, and therefore uncomputable. https://www.cs.auckland.ac.nz/~cristian/approxcompactIJBC.pdf : [md-2] The abstract says "The value of an Omega number is highly machine-dependent" and "hence uncomputable". Theorem 1 specifies the additive simulation constant; Theorem 2 identifies the left-c.e. random reals. Dependence does not mean different machines always give different numbers.
CONFIRMED A different interpreter, instruction alphabet, or data encoding generally defines a different number. https://www.cs.auckland.ac.nz/~cristian/approxcompactIJBC.pdf : [md-3] The abstract says "The value of an Omega number is highly machine-dependent" and "hence uncomputable". Theorem 1 specifies the additive simulation constant; Theorem 2 identifies the left-c.e. random reals. Dependence does not mean different machines always give different numbers.
CONFIRMED Their twelve classified halting counts cover total lengths 7, 14, and so on through 84 bits. https://www.cs.auckland.ac.nz/~cristian/Calude361_370.pdf : [md-10] Table 1, p. 367: lengths 7,14,21,28,35,42,49,56,63,70,77,84 have counts 1,1,3,8,50,311,1012,4382,19164,99785,515279,2559837. Every displayed row agrees.
CONFIRMED Uncomputability of the complete expansion is a theorem, not a live controversy. https://www.cs.auckland.ac.nz/~cristian/approxcompactIJBC.pdf : [md-20] The abstract says "The value of an Omega number is highly machine-dependent" and "hence uncomputable". Theorem 1 specifies the additive simulation constant; Theorem 2 identifies the left-c.e. random reals. Dependence does not mean different machines always give different numbers.
CONFIRMED The finite challenge is specific to this machine and this encoding: extend the checked classification and residual bound far enough to certify another bit. https://arxiv.org/pdf/nlin/0112022 : [md-21] Section 7 says "only 64" and explains the possible carry to bit 65. A narrower checked interval is the stated finite challenge, not a claim that the whole real becomes computable.
CONFIRMED The later 43-bit and 40-bit calculations in Calude and Dinneen's 2007 paper concern two other data encodings, hence two other halting measures. https://www.cs.auckland.ac.nz/~cristian/approxcompactIJBC.pdf : [md-22] The abstract says "43 exact bits for the base 16 machine and 40 exact bits for the base 2 machine". Section 3 specifies four-bit instruction characters instead of the earlier seven-bit encoding. These are different measures, not more bits of the 2002 measure.
CONFIRMED They are not improved precision for the 2002 number. https://www.cs.auckland.ac.nz/~cristian/approxcompactIJBC.pdf : [md-23] The abstract says "43 exact bits for the base 16 machine and 40 exact bits for the base 2 machine". Section 3 specifies four-bit instruction characters instead of the earlier seven-bit encoding. These are different measures, not more bits of the 2002 measure.
CONFIRMED This is not the value of a universal constant. https://www.cs.auckland.ac.nz/~cristian/approxcompactIJBC.pdf : [not-universal-constant] The abstract says "The value of an Omega number is highly machine-dependent" and "hence uncomputable". Theorem 1 specifies the additive simulation constant; Theorem 2 identifies the left-c.e. random reals. Dependence does not mean different machines always give different numbers.
CONFIRMED It is a halting measure attached to the specific universal self-delimiting register machine used by Cristian Calude, Michael Dinneen, and Chi-Kou Shu in 2002. https://www.cs.auckland.ac.nz/~cristian/Calude361_370.pdf : [specific-machine-authors] The journal paper names Cristian S. Calude, Michael J. Dinneen, and Chi-Kou Shu and describes a universal self-delimiting register machine. Crossref gives 2002, Experimental Mathematics 11(3), 361-370.
CONFIRMED Fair independent bits feed one fixed prefix-free machine. https://arxiv.org/pdf/nlin/0112022 : [probability-0] Section 2 defines prefix-free sets and gives the Kraft measure in [0,1]; it calls that sum "a probability". Disjoint binary cylinders have weights 2^(-length).
CONFIRMED A minimal halting input of length |p| owns a cylinder of measure 2-|p|. https://arxiv.org/pdf/nlin/0112022 : [probability-1] Section 2 defines prefix-free sets and gives the Kraft measure in [0,1]; it calls that sum "a probability". Disjoint binary cylinders have weights 2^(-length).
CONFIRMED Prefix-freeness makes those cylinders disjoint. https://arxiv.org/pdf/nlin/0112022 : [probability-2] Section 2 defines prefix-free sets and gives the Kraft measure in [0,1]; it calls that sum "a probability". Disjoint binary cylinders have weights 2^(-length).
CONFIRMED Omega_U = sum over halting p of 2-|p| https://arxiv.org/pdf/nlin/0112022 : [omega-equation] Section 2 defines prefix-free sets and gives the Kraft measure in [0,1]; it calls that sum "a probability". Disjoint binary cylinders have weights 2^(-length).
CONFIRMED Running more programs only approaches Omega from below. https://www.cs.auckland.ac.nz/~cristian/approxcompactIJBC.pdf : [carry-explanation-0] The introduction says "Computing lower bounds is not difficult" and explains the uncertainty caused by overflows. Its upper-bound discussion supports the distinction between observed lower sums and certified digits.
CONFIRMED An unseen halt can add mass far to the right and carry left through provisional digits. https://www.cs.auckland.ac.nz/~cristian/approxcompactIJBC.pdf : [carry-explanation-1] The introduction says "Computing lower bounds is not difficult" and explains the uncertainty caused by overflows. Its upper-bound discussion supports the distinction between observed lower sums and certified digits.
CONFIRMED The 2002 result did something stronger: it classified programs through length 84, bounded every longer contribution with two geometric series, and put the true value inside an interval whose endpoints share a prefix. https://www.cs.auckland.ac.nz/~cristian/Calude361_370.pdf : [carry-explanation-2] Section 6 describes testing lengths 7 through 84 and proving nonhalting cases; section 7 provides both tail series and the 64-bit carry conclusion. The journal and arXiv tail series differ in 48 versus 47, but independently recomputing either still gives 64 bits.
CONFIRMED For every fixed universal prefix-free machine, Omega is left-computably enumerable, algorithmically random, and uncomputable. https://www.cs.auckland.ac.nz/~cristian/approxcompactIJBC.pdf : [universal-theorem] The abstract says "The value of an Omega number is highly machine-dependent" and "hence uncomputable". Theorem 1 specifies the additive simulation constant; Theorem 2 identifies the left-c.e. random reals. Dependence does not mean different machines always give different numbers.
CONFIRMED That asymptotic question is settled. https://www.cs.auckland.ac.nz/~cristian/approxcompactIJBC.pdf : [asymptotic-settled] The abstract says "The value of an Omega number is highly machine-dependent" and "hence uncomputable". Theorem 1 specifies the additive simulation constant; Theorem 2 identifies the left-c.e. random reals. Dependence does not mean different machines always give different numbers.
CONFIRMED For this exact 2002 machine, the source certifies 64 initial bits. https://www.cs.auckland.ac.nz/~cristian/Calude361_370.pdf : [source-64-bits] The journal explicitly says "the first 64 exact bits" and prints 0000001000000100000110001000011010001111110010111011101000010000. This matches the page engine and independent arithmetic.
CONFIRMED The nearest concrete challenge is therefore exact and machine-specific: give a checked classification and residual-mass bound narrow enough to fix the next bit. https://arxiv.org/pdf/nlin/0112022 : [later-machines-0] Section 7 says "only 64" and explains the possible carry to bit 65. A narrower checked interval is the stated finite challenge, not a claim that the whole real becomes computable.
CONFIRMED The 2007 result does not extend this prefix. https://www.cs.auckland.ac.nz/~cristian/approxcompactIJBC.pdf : [later-machines-1] The abstract says "43 exact bits for the base 16 machine and 40 exact bits for the base 2 machine". Section 3 specifies four-bit instruction characters instead of the earlier seven-bit encoding. These are different measures, not more bits of the 2002 measure.
CONFIRMED It certifies 43 bits for a base-16 data machine and 40 for a base-2 data machine, two different halting measures. https://www.cs.auckland.ac.nz/~cristian/approxcompactIJBC.pdf : [later-machines-2] The abstract says "43 exact bits for the base 16 machine and 40 exact bits for the base 2 machine". Section 3 specifies four-bit instruction characters instead of the earlier seven-bit encoding. These are different measures, not more bits of the 2002 measure.
CONFIRMED Exact first bits of this same machine would decide halting for its inputs up to the matching length. https://arxiv.org/pdf/0904.1149 : [halting-bits-0] Tadaki, Fact 1: "The first n bits of the base-two expansion of" Omega solve the same machine's halting problem "for inputs of length at most n". The machine restriction in the page is correct.
CONFIRMED They do not answer every program in every language for free. https://arxiv.org/pdf/nlin/0112022 : [halting-bits-1] Section 2 gives a constant c depending on U and C and the inequality for program length: the simulating input is at most the original length plus c. Thus the overhead must be counted.
CONFIRMED Translation into this machine costs bits, and the overhead must be counted. https://arxiv.org/pdf/nlin/0112022 : [halting-bits-2] Section 2 gives a constant c depending on U and C and the inequality for program length: the simulating input is at most the original length plus c. Thus the overhead must be counted.
CONFIRMED Fixed machine: the 2002 compressed canonical register machine, with seven-bit ASCII characters, one final HALT instruction, and data consumed exactly. https://www.cs.auckland.ac.nz/~cristian/Calude361_370.pdf : [fixed-machine] Section 5 requires one final HALT, errors for unread or over-read data, compressed canonical syntax, and length equal to seven times the character count. The curated input family obeys these conventions.
CONFIRMED This free choice determines the number. https://www.cs.auckland.ac.nz/~cristian/approxcompactIJBC.pdf : [machine-choice] The abstract says "The value of an Omega number is highly machine-dependent" and "hence uncomputable". Theorem 1 specifies the additive simulation constant; Theorem 2 identifies the left-c.e. random reals. Dependence does not mean different machines always give different numbers.
CONFIRMED Transcribed evidence: twelve halting counts through 84 bits and the two length-84 prefix counts from Calude, Dinneen, and Shu. https://www.cs.auckland.ac.nz/~cristian/Calude361_370.pdf : [transcribed-evidence] The journal Table 1 supplies all twelve counts, and section 7 supplies "402906842 prefixes not containing HALT" and "1748380 prefixes containing HALT". Both transcriptions match the constants in the engine.
CONFIRMED Tail convention: lengths 91 and above are bounded by the paper's two closed geometric series. https://www.cs.auckland.ac.nz/~cristian/Calude361_370.pdf : [tail-convention] Confirmed: both primary versions bound lengths 91 and above by two closed geometric series. Their factors differ, 48 in the journal and 47 in arXiv, so the broad convention is true while the exact-version attribution needs F2.
CONFIRMED Calude, Cristian S.; Dinneen, Michael J.; Shu, Chi-Kou. Computing a Glimpse of Randomness. Experimental Mathematics 11(3), 361-370, 2002. Also arXiv:nlin/0112022. https://api.crossref.org/works/10.1080/10586458.2002.10504481 : [citation-2002] Crossref top-level metadata agrees on title, all three authors, 2002, volume 11, issue 3, and pages 361-370. The arXiv record confirms nlin/0112022.
CONFIRMED Calude, Cristian S.; Dinneen, Michael J. Exact Approximations of Omega Numbers. International Journal of Bifurcation and Chaos 17(6), 1937-1954, 2007. https://api.crossref.org/works/10.1142/S0218127407018130 : [citation-2007] Crossref top-level metadata agrees on Calude and Dinneen, the title, 2007, volume 17, issue 06, and pages 1937-1954. The author-hosted paper confirms its abstract and results.
CONFIRMED Tadaki, Kohtaro. Chaitin Omega Numbers and Halting Problems. Mathematical Theory and Computational Practice, LNCS 5635, 2009. arXiv:0904.1149. https://api.crossref.org/works/10.1007/978-3-642-03073-4_46; https://arxiv.org/abs/0904.1149 : [citation-2009] The bibliographic claim is correct using the top-level Crossref title, author and year plus the arXiv journal reference for LNCS 5635. However the source-record excerpt's "volume":"255" and "issue":"1" are fields in two different cited references, not metadata for this chapter. See F6.
CONFIRMED Also arXiv:nlin/0112022. https://arxiv.org/abs/nlin/0112022 : [arxiv-2002] The fetched arXiv record names "Computing A Glimpse of Randomness", Calude, Dinneen and Shu, and "arXiv:nlin/0112022".
CONFIRMED Mathematical Theory and Computational Practice, LNCS 5635, 2009. arXiv:0904.1149. https://arxiv.org/abs/0904.1149 : [lncs-2009] The fetched arXiv journal reference explicitly gives "CiE 2009. LNCS, vol 5635 (2009) Springer". This supplies the volume missing from Crossref top-level metadata.
CONFIRMED Both operate the halting problem through concrete machines. git show 0a15703d9e:src/content/strata/the-longest-finite-race.md : [nearby-the-longest-finite-race] The linked page says "Fix the number of states and ask which machine that eventually stops runs longest." This supports the shared concrete-machine topic.
CONFIRMED That page tests finite strings without claiming to certify randomness. git show 0a15703d9e:src/content/strata/the-coin-you-cant-fake.md : [nearby-the-coin-you-cant-fake] The linked page says the three statistics are "probabilities rather than verdicts". This supports the limited finite-test comparison.
CONFIRMED Both turn undecidability into a precise limit on formal knowledge. git show 0a15703d9e:src/content/strata/godel-incompleteness.md : [nearby-godel-incompleteness] The linked page describes the sentence asserting its own unprovability and the precise theorem conditions. Its negative sentence alone is weak support; the rest of the same pinned dek supports the comparison.
CONFIRMED The portal distinguishes resource limits from uncomputability and unprovability. git show 0a15703d9e:src/content/strata/the-limits-of-knowing.md : [nearby-the-limits-of-knowing] The linked page explicitly distinguishes "resource, computability, proof". The comparison accurately describes that page.
CONFIRMED Both pages end at the same wall from opposite sides. git show 0a15703d9e:src/content/strata/the-battery-that-cant-tell.md : [nearby-the-battery-that-cant-tell] The reciprocal note repeats the claim, so it is not independent evidence by itself. The same pinned file explains its finite statistical battery and contrasts it with algorithmic randomness; the comparison is supported by that fuller context.
CONFIRMED Both operate the halting problem through concrete machines. Busy Beaver asks how long a small halting machine can run, while this page weighs every halting input and exposes why observed halts give only a lower bound. git show 0a15703d9e:src/content/strata/the-longest-finite-race.md : [related-note-0] The linked page defines the longest halting run for a fixed state count. The present page sums witnessed halting-input weights. The comparison is supported, without importing the other page's unrelated numerical claims.
CONFIRMED That page tests finite strings without claiming to certify randomness. This one reaches the algorithmic notion through a fixed universal machine whose halting probability is random, while keeping every finite numerical certificate separate from the infinite theorem. git show 0a15703d9e:src/content/strata/the-coin-you-cant-fake.md : [related-note-1] The linked page calls its statistics "strong evidence, never proof" and "probabilities rather than verdicts". The 2007 primary paper supports the universal-machine randomness theorem in the second sentence.
CONFIRMED Both turn undecidability into a precise limit on formal knowledge. Here the limit is carried by exact bits of a machine-dependent real and by the halting questions those bits decide. git show 0a15703d9e:src/content/strata/godel-incompleteness.md : [related-note-2] The linked page explains the unprovability construction. Tadaki Fact 1 supplies the halting questions determined by exact bits; the comparison keeps finite certificates separate from the undecidability theorem.
CONFIRMED The portal distinguishes resource limits from uncomputability and unprovability. This page supplies a different operable object at that boundary: a left-enumerable measure whose lower approximations are easy and whose exact digits demand upper proofs. git show 0a15703d9e:src/content/strata/the-limits-of-knowing.md : [related-note-3] The linked portal distinguishes "resource, computability, proof". The 2007 primary paper supplies the left-c.e. lower approximation and upper-bound explanation used in the present page.
CONFIRMED Use row Length Halting count Contribution<br>7 1 1 x 2-7<br>14 1 1 x 2-14<br>21 3 3 x 2-21<br>28 8 8 x 2-28<br>35 50 50 x 2-35<br>42 311 311 x 2-42<br>49 1,012 1,012 x 2-49<br>56 4,382 4,382 x 2-56<br>63 19,164 19,164 x 2-63<br>70 99,785 99,785 x 2-70<br>77 515,279 515,279 x 2-77<br>84 2,559,837 2,559,837 x 2-84 https://www.cs.auckland.ac.nz/~cristian/Calude361_370.pdf : [rendered-table] Table 1, p. 367: lengths 7,14,21,28,35,42,49,56,63,70,77,84 have counts 1,1,3,8,50,311,1012,4382,19164,99785,515279,2559837. Every displayed row agrees.
CONFIRMED The theorem is closed. https://www.cs.auckland.ac.nz/~cristian/approxcompactIJBC.pdf : [theorem-closed-heading] The abstract says "The value of an Omega number is highly machine-dependent" and "hence uncomputable". Theorem 1 specifies the additive simulation constant; Theorem 2 identifies the left-c.e. random reals. Dependence does not mean different machines always give different numbers.
WRONG reduces the two published geometric bounds for every program of length 91 or more https://arxiv.org/pdf/nlin/0112022v3 : [preprint-geometric-bounds; R2] The prior CONFIRMED evidence checked only the HALT-free term. Its sum does match the first implemented term, but the HALT-containing series starting at m=0 sums to 64 times the second implemented term. The claim about reducing both series was not supported. This shares X3's open question about the implemented bound.
CONFIRMED Busy Beaver asks how long a small halting machine can run https://arxiv.org/abs/2509.12337 : [related-busy-beaver] The fetched primary paper abstract defines S(n) as the maximum halting step count for an n-state two-symbol Turing machine started on an all-zero tape. This agrees with the linked-page summary.
CONFIRMED a fixed universal machine whose halting probability is random https://www.cs.auckland.ac.nz/~cristian/approxcompactIJBC.pdf : [related-randomness] The abstract says "The value of an Omega number is highly machine-dependent" and "hence uncomputable". Theorem 1 specifies the additive simulation constant; Theorem 2 identifies the left-c.e. random reals. Dependence does not mean different machines always give different numbers.
CONFIRMED the halting questions those bits decide https://arxiv.org/pdf/0904.1149 : [related-halting-questions] Tadaki, Fact 1: "The first n bits of the base-two expansion of" Omega solve the same machine's halting problem "for inputs of length at most n". The machine restriction in the page is correct.
CONFIRMED a left-enumerable measure whose lower approximations are easy and whose exact digits demand upper proofs https://www.cs.auckland.ac.nz/~cristian/approxcompactIJBC.pdf : [related-left-enumerability] The introduction says "Computing lower bounds is not difficult" and explains the uncertainty caused by overflows. Its upper-bound discussion supports the distinction between observed lower sums and certified digits.
WRONG it evaluates the reduced expression printed in arXiv v3, equation (2) https://arxiv.org/pdf/nlin/0112022v3; research/assay-line/queue-1018/chaitin-omega/review-arithmetic.py : [R1; false-replacement-attribution] This replacement sentence was added by the fixer. The printed expression gives 45129500159/350956001137405307933901717504 and 62 common bits, whereas the implemented hybrid gives 6481914167/3158604010236647771405115457536 and 64. The printed first term is 63 times the implemented first term. Corrected to describe the calculation and its unresolved bound assumption, without claiming the paper's reported 64-bit result is false.
What was done
[fixed] X1 (machine-change-always-changes-omega), WRONG: changed the authored HTML assertion from the number changes to the number can change. The journal conclusion explicitly allows different universal machines with the same Omega. Added the dated correction to the HTML Corrections section and the markdown body. The markdown's existing generally-qualified sentence was already supported.
[fixed] X2 (prefix-audit-patricia-representation), MINOR: described the sorted adjacent-input prefix test, marked the Patricia node count as derived, replaced the initial unbuilt-tree readout, and qualified the verifier's node-count message. Regenerated only the layer's public verifier copy. No page correction line.
[open note:0c96e2] X3 (tail-bound-source-version), WRONG, also answering preprint-geometric-bounds: the source-version mismatch is real, but the attempted arXiv attribution was false. The source equations have factor-63 and factor-64 reduction discrepancies. The page and markdown now disclose the implemented assumption and qualify their 64-bit result; the underlying validity of that bound remains unresolved. The directing instance must file the NOTE NEEDED paragraph in repair-report.md and replace PENDING with its board-issued id.
[fixed] X4 (reduced-numerator-fixed-denominator-label), WRONG: changed Accepted lower numerator over 2^84 to Accepted lower numerator (reduced fraction), retaining the actual denominator readout. With row 84 withheld, the engine correctly displays 1190035900005385093455 over 151115727451828646838272 (2^77), equivalent to 152324595200689291962240 over 2^84. Added the dated correction to HTML and markdown. JSON-LD dateModified is 2026-10-18; this metadata revision needs no separate Updated line.
[fixed] R1 (false-replacement-attribution), WRONG: removed the false printed-expression attribution from all authored layer copies and comments. Kept the numerical experiment, described the exact coefficients, made the certificate conditional in the page, markdown, metadata and generated readouts, and added a mirrored Corrected 2026-10-18 line naming the assay line. The source-bound question stays open under X3.
[fixed] R2 (false-confirmation), WRONG: changed the duplicated preprint two-series claim from CONFIRMED to WRONG, corrected the X3 evidence and disposition, and reconciled the counts to 57 = 51 CONFIRMED + five WRONG + one MINOR counted as wrong. The old fix reports are superseded on X3 by the review and repair reports; the unresolved claim is not declined.
The assay office: what a record is, and every layer re-read so far.