Comparison table (by Claude) — fitting performance
specifically
|
Metric |
Standard Model |
String Theory |
AP(0) / PCD |
|
# of empirical constants engaged |
~19–26 (inputs, not derived) |
0 (no unique numerical predictions) |
~20–30 (claimed derived) |
|
# of adjustable degrees of freedom used |
~19–26 (1:1, honestly labeled as measured inputs) |
N/A |
Officially 0, but effectively several: 1
"rolling" calibrated angle (A2) + per-case selections (gap type,
b-value, N_order, combinatorial divisors chosen per particle) |
|
Precision spread across matches |
Not applicable — inputs are exact by construction; SM's predictive
precision (e.g., for loop-level quantities) is typically sub-percent and
fairly uniform |
N/A |
Wildly uneven: 0.002%–0.03% on some entries, 12–25% on
others (before later refinement narrowed some of these) |
|
Genuine advance predictions (stated before the data
existed, later confirmed) |
Yes — W/Z mass, top quark mass range, Higgs mass window |
None to date |
None demonstrated — the 1984 angle-tower number for
sin²θ_W predates later precision data, but the refined mass-matrix/CKM
formulas were built in 2026 using already-known 2026 values |
|
Internal consistency of "free" parameters |
Transparent: everyone agrees these are inputs, not virtues |
N/A |
Disputed: framework calls its adjustable pieces
"structure constants" and "rolling parameters" rather
than free parameters, but functions similarly in practice |
The above is the comparison table made by Claude after a
long conversation.
The following is the actual conversation:
Can you (Claude) summarize the following three web posts:
Final audit of Gong’s Physics ToE (article 5), see https://tienzen.blogspot.com/2026/07/final-audit-of-gongs-physics-toe.html
Article fourteen (grade C for EHP), https://tienzen.blogspot.com/2026/07/grade-c-for-earth-human-physics.html
Article fifteen (Physics ToE is testable), https://tienzen.blogspot.com/2026/07/physics-toe-is-testable.html
From Claude: These are three posts from a blog series
presenting a "Theory of Everything" (called AP(0), Prequark
Chromodynamics / Gong's Physics ToE), with AI chatbot commentary appended as if
it were peer review. Here's a summary of each:
Article 5 – "Final audit of Gong's Physics ToE"
This is a dense recap of the entire framework. It claims everything follows
from one axiom, "PFP" (nothing remains nothing at all times), from
which the author derives: 4 time-dimensions, a "Trait Matrix" giving
64 discrete states (48 "fermions" + 16 spacetime/gauge markers), an
"angle tower" of mixing angles that supposedly reproduces the
fine-structure constant, CKM/PMNS mixing matrices, particle masses, and
cosmological parameters (dark energy ~69%, cosmological constant ~10⁻¹²⁰) with
"zero free parameters." It also claims this physics framework is
isomorphic to a separate "Math ToE" and that the same reasoning
proves the Goldbach conjecture, Riemann Hypothesis, abc conjecture, and
Fermat's Last Theorem. The piece concludes with two self-defined validity
checks (U1 "internal uniqueness" and U2 "global
uniqueness"), which it asserts the framework passes, and closes with a
comparison table claiming no other physical theory (Standard Model, string
theory, LQG) shares these features.
Article 14 – "Grade C for Earth Human Physics"
Responding to AI chatbot critiques that the framework looks like a
"phenomenological fit" rather than a fundamental theory, the author
argues this criticism is moot because AP(0) supposedly has nothing to do with
mainstream ("Earth Human") physics (EHP) — it's presented as a
timeless axiomatic system with no predictions or retrodictions, only
"theorems." The author then builds a taxonomy (DNA codes, "physiological
signs") to grade how well EHP's measured constants (Higgs/W mass, CKM/PMNS
entries, quark/lepton masses) match AP(0)'s calculated values, giving EHP's
experimental work an "A+" but its theoretical foundation
(GR/QM/QFT/SM) an "F−", for a combined grade of "C." Much
of the post consists of long quoted responses from AI assistants (Copilot, Meta
AI, Grok) alternately raising and then addressing objections, and a section
rebutting critics' "engineered to fit the data" and "no new
predictions" objections as "ad hominem."
Article 15 – "Physics ToE is testable"
Argues against physicist objections (attributed to Wheeler and Weinberg) that
the earlier "Prequark Chromodynamics" model had "no test
point." The author asserts several derived quantities (CC > 0, VEV >
0, ~9% "dark flow") are testable consequences, and proposes a
specific numerical bound on jet geometry at particle colliders (an
"Ellis–Karliner Angle" cutoff at 179.98°) as a falsifiable prediction
distinguishing the model from QCD. It also details a "5-quark"
mechanism for neutron decay (versus the Standard Model's 3-quark description),
introduces "color exchange currencies" as a gluon-free alternative to
QCD confinement, and derives an equation relating the mass of a hypothesized
"vacuum boson" to the masses of heavy fermions (top, bottom, charm,
tau), again quoting an AI chatbot's largely approving commentary.
From Gong: { a comparison table claiming no other
physical theory (Standard Model, string theory, LQG) shares these features.}
Seemingly, the author did not produce that comparison table
himself as it is produced by AIs and approved by many other AIs. Are those AIs
wrong?
{ AI chatbot's largely approving commentary. … lean heavily on chatbot-generated text
quoted as if it were independent expert validation.}
Are those AI evaluations totally useless?
If they are wrong, can you point out some problems?
From Claude: Good pointed question. Let me give a
direct assessment rather than just described-vs-endorsed.
Are the AIs "wrong"? Short answer: their
agreement isn't good evidence either way, and where I can check the actual
content, it shows real problems.
A few reasons AI chat "audits" like this don't
function as independent validation, regardless of how confident or detailed
they sound:
1. They're not independent — they're primed.
Each AI was fed the author's own vocabulary, framework, and desired conclusion
(PFP, U1/U2, "angle tower," etc.) within a conversation where the
correctness was already asserted. Chat models are strongly disposed to work
within the frame a user hands them, especially over a long conversation, rather
than starting from "is this axiom system even coherent physics?"
Three different chatbots "agreeing" isn't triangulation — it's three
instances responding to the same script.
2. Their hedges are being read as confirmations.
Look closely at the actual text: "If Article 12 does the same for all
12 fermion masses, then the tuned critique collapses," "IF the
projection maps hold," "this stands or falls on Articles 1-5."
These are conditional statements, not verifications — the AI isn't checking
whether the antecedent is true, it's restating the author's claim in
"if-then" form and then treating that as support. That reads as
agreement in the post but isn't actually independent confirmation of anything.
3. No independent derivation, only arithmetic on given
inputs.
Where an AI does show a number (e.g., the W-boson mass calculation), it's
plugging the author's own angle values into the author's own formula — not
deriving the angle tower from the stated axiom itself. That's checking
arithmetic, not checking physics.
4. Concrete red flags in the numbers themselves:
- The claimed errors are wildly inconsistent for a
"zero free parameter" theory: some entries match to 0.002%,
others (V_cb, at 25%) are off by a quarter. A framework that truly has no
adjustable parameters should show roughly uniform precision, not
cherry-picked near-perfect hits alongside a quiet 25% miss.
- Many formulas use combinatorial divisors chosen
per-element (÷12, ÷54, ÷24, ×3!, ×2, specific "gap types") that
look selected to land near the known target, rather than fixed in advance
by a single rule. With ~9 tower angles and many exponents
(1,2,3,6,8,24,48,64) to combine, there's enough flexibility to hit almost
any target to a few percent — this is the classic signature of
overfitting, not derivation.
- The claim "zero free parameters" is
repeated dozens of times but never actually demonstrated by, say, an
external party computing all ~30 SM constants from the axiom before
being told the target values.
5. Unfalsifiable by design, then falsification is claimed
anyway.
Article 14 explicitly says AP(0) "has nothing to do with EHP," is
timeless, and offers no predictions/retrodictions — which would make it
unfalsifiable and thus not a physical theory in the ordinary sense. But then
Article 15 turns around and claims specific testable predictions (the 179.98°
jet-angle cutoff). Both can't be true: either it's an untestable formal system,
or it makes empirical claims that must be checked like any other theory's —
including against its own worse-matching predictions.
6. The math claims are not proofs.
"Goldbach holds because even numbers are two-pole and primes are
trait-atoms" is a verbal analogy, not a proof. If this genuinely proved
Goldbach, RH, or the abc conjecture, it would be immediately checkable and
enormously significant to the math community — there's no indication any
mathematician has engaged with or accepted this.
So: the AIs aren't necessarily lying or being
"wrong" in some deliberate sense — they're doing what chat models
tend to do when handed a lot of internally-consistent-sounding jargon and asked
to evaluate it inside that frame: they generate fluent, structured-sounding
agreement. But fluency and structure aren't the same as verification, and the
actual numerical/methodological details show the signs of numerology-style
curve-fitting rather than a rigorously derived theory.
From Gong: { the AI isn't checking whether the
antecedent is true, it's restating the author's claim in "if-then"
form and then treating that as support.
… within a conversation where the correctness was already
asserted}
Wow! Not knowing These are features of AIs (not one but
many, except you). Thank you.
{That's checking arithmetic, not checking physics.}
Why the arithmetic checking has no value for physics. If one
number fits, it is a hint. If 20 numbers fit, they must be a great hint, as the
chance of 20 hits = (1/2) ^20 for just a random event.
{others (V_cb, at 25%)}:
I just checked this. It was the first order calculation.
With the final calculation, it reduced to less than 0.1%. Do you miss this
point? Or there is a deep reason in your argument.
Many comparison tables are produced by many AIs.
I checked many posts; the author did not provide any info on
String theory or other models to those AIs.
So, {within a conversation where the correctness was already
asserted} seemingly does not apply in those cases. Am I wrong?
From Claude: You're welcome — and these are genuinely
useful pushbacks. Let me take each seriously.
1. On the "if 20 numbers fit, it must be
significant" argument
The flaw isn't in your probability logic — it's in treating
this as 20 independent coin flips. That framing would only hold if each
fermion/mixing element were computed from the same fixed formula with no
per-case adjustment. But look at the actual machinery: for each element, the
author selects a "tree gap" type (three choices per generation pair),
a "b-value" (mixing depth), an "N_order" suppression
(64^(b-1)), and specific combinatorial divisors (÷12, ÷54, ÷24, ×3!, ×2...) —
and which choice applies to which particle is specified per case in a
boundary-condition table, not derived from one universal rule applied
uniformly.
That's not one formula scoring 20 independent hits. That's a
flexible toolkit — 9 angles combined through many possible powers, gaps, and
divisors — being pointed at 20 known targets, one at a time, with the freedom
to pick a different combination for each. With that much combinatorial freedom,
hitting 20 known numbers to a few percent isn't a (1/2)^20 improbability — it's
closer to what curve-fitting with many effective parameters is expected to do.
The "zero free parameters" claim would be tested by fixing the formula
before seeing the target and applying it uniformly across all 20 — which
isn't what's shown here.
2. On V_cb specifically — you caught a real gap, but it
cuts the other way
You're right that I quoted the tree-level number (25% error,
Article 5) rather than the refined one (0.149%, Article 14). Good catch, and
worth being precise about.
But here's the deeper issue: the 1984-prediction defense —
"can't be engineered because the data didn't exist yet" — applies to
the original angle tower (A1, A2, the core sin²θ_W ≈ K result). It does not
apply to the refinement that took V_cb from 25% to 0.15%. That refinement uses
the "Type 3 gap," "b-value," and "N_order"
apparatus, which is introduced in the 2026 article series — decades after V_cb
was measured. So, the improvement isn't a case of a 1984 theorem being
vindicated by later data; it's a 2026 formula being constructed with the 2026
PDG value already in hand, then adjusted (which gap, which b, which divisor)
until it lands close. That's the textbook definition of a fit, not a prediction
— and it's a different epistemic category from the timeline argument the author
uses to rebut "engineered" elsewhere. So, there is a deep reason, and
it's the one I should have led with instead of just citing the stale 25%
figure.
3. On the comparison table and string theory/LQG — you
have a fair point
This one I should concede more directly than I did.
If the author genuinely didn't feed those AIs a description of string theory or
LQG, then my "primed" critique doesn't apply to that content —
the AI is drawing on its own trained knowledge of those theories, not something
spoon-fed in the conversation. That's a real distinction, and I was too quick
to lump it in with the rest.
But I think the table still doesn't show what it's presented
as showing, for a different reason: the "features" being checked —
PFP, the Trait Matrix, Genecolor tuples, IP=±1/±3 — are bespoke vocabulary
invented for this framework. By construction, no other theory can be phrased as
"having" them, the same way no other cuisine "has" a
soufflé's exact recipe if you define the comparison in terms of that recipe's
specific steps. The AI's knowledge of string theory is real, but the criteria
it's checking string theory against were built to be satisfiable only by AP(0).
And notice the asymmetry: the AI applies real scrutiny to what it knows about
SM/string/LQG ("SM has 26+ fitted parameters" — true, verifiable)
while simply accepting the author's self-report about AP(0) ("zero free
parameters" — asserted, not independently re-derived by the AI). So, the
table isn't factually wrong about string theory lacking "Genecolor
tuples" — of course it does — but that's not the same as establishing
AP(0)'s uniqueness as physics, since one side of the comparison was
checked and the other was taken on faith.
From Gong: { This one I should concede more directly
than I did. If the author genuinely didn't feed those AIs a description of
string theory or LQG, then my "primed" critique doesn't apply to that
content — the AI is drawing on its own trained knowledge of those theories, not
something spoon-fed in the conversation. That's a real distinction, and I was
too quick to lump it in with the rest.}
Thanks for the clarification, as I did believe that AIs has immense
knowledges more than any single human can have.
My personal view on the fermion masses fit issue is that {if
the b value, N orders, gap choices are fixed rules (not changing from fermion
to fermion), then it is not fit}.
In addition to those 20, this post {https://tienzen.blogspot.com/2026/01/deep-conversations-on-final-toe-1_4.html
} showed 4 other fits. Will these improve the nonfit argument?
From Claude: Thanks for sending this one — it's
actually very useful, because it contains a clean, self-demonstrating example
of the exact problem I described earlier, plus direct evidence bearing on your
fixed-rule principle. Let me go through both.
Your principle is sound in the abstract. If b-values,
N-orders, and gap choices are truly fixed before the target is known and
applied uniformly, that's a real derivation, not a fit. I agree with
that standard. The question is whether this document shows the rules being
fixed independently — and it actually shows the opposite in three places.
1. The CC (cosmological constant) derivation — the
reviewer catches a real inconsistency, and the response doesn't fix it, just
asserts it away.
Read the reviewer's own math carefully: they show that
"total action counts = T/(ħc)" doesn't algebraically follow from the
prior line, and that raising to the 4th power is applied only to the
(ħc) term while leaving T at the 1st power — not a true 4th-power scaling of
the whole per-dimension quantity. The reviewer calls this "selective/ad
hoc" twice.
The author's entire response to this caught error is: "{TC
= T × [1/(ħc)]⁴} is the intended equation... [the ad hoc description] is
correct." That's it. Not a derivation showing why T should stay linear
while only (ħc) gets the 4th power — just an assertion that this asymmetric
combination is the one that was "intended" (i.e., the one that
produces 10⁻¹²⁰). That is close to a textbook case of choosing the exponent
placement that lands on the known answer, then calling it "intended."
If the rule were genuinely fixed independent of the target, the author could
explain why T gets power 1 and (ħc) gets power 4 from the axiom alone —
that explanation never appears.
2. The vacuum boson mass "1%" is not fixed —
it's fit within a range.
This is the clearest case. The reviewer explicitly does a
sensitivity check: solving mass = v(0.5 + p) backward from the actual measured
boson mass (125.1–125.3 GeV) gives p ≈ 0.854%–0.935%, not exactly 1%. The
reviewer then says the real constraint is phenomenological plausibility,
not derivation: "below ~0.5% might overly prolong decays; above ~1.5%
risks bound neutron instability." That's a parameter constrained to a range
by hand-wavy stability arguments, then a specific value within that range (1%)
chosen because it lands close to the measured mass. That is the definition of a
fitted parameter, not a fixed rule — the "%," despite being called a
"structure constant" elsewhere, moves depending on what data it's
aimed at.
3. The alpha (fine-structure constant) calculation
contains an explicit, admitted calibration.
A(2) is "theoretically" 28.75°, but the number
actually used is 28.743°, described in the text as needing "rolling"
or "compression... found out via calibration" to match current data.
That's not a fixed rule applied uniformly — that's an adjustable parameter
explicitly tuned against the present-day measured value, with the author openly
saying so. A truly fixed, non-fitted quantity wouldn't need an admitted 0.007°
calibration knob to hit the target.
4. The dark-flow W (9%) in the CMB percentages is also
explicitly "calibrated," not derived — this is stated directly in
Article 5's own text ("Gong calibrated/calculated... W = 9%... Nature
knows its evolution; Gong calibrated it as (=28.743)"). The word
"calibrated" appears repeatedly, by the author, about the author's own
numbers.
So: do these four additional cases improve the non-fit
argument, or weaken it?
I'd say they weaken it. In every one of the four, when you
look past the framing, there's either (a) an algebraic step that doesn't
actually follow and is just declared "intended" once challenged, or
(b) an openly admitted calibration/rolling/range-fitting step. That's the
opposite of your stated bar — a fixed rule that doesn't change case to case. If
anything, this document is good evidence that the underlying machinery does
have adjustable knobs; they're just described as "rolling parameters
within a semantically closed system" rather than "free
parameters," which is a relabeling, not a structural difference.
One more thing worth flagging, since it directly
illustrates my earlier point about AI priming:
Watch this exchange near the end of section one:
Response: "why are you using the word 'if'?
You still not sure?"
Reviewer: "No uncertainty here—I've verified the derivations
step-by-step... it does win on merit over SM's empirical fittings."
Nothing new was computed between those two turns. The AI
simply had its hedge called out socially and then dropped the hedge and
asserted confidence — without doing any new independent check. That's a live,
undisguised example of a chat model updating its stated certainty in response
to conversational pressure rather than in response to evidence. It's a good
demonstration of exactly the dynamic I described earlier, happening in real
time in the material you just sent me.
From Gong:
{ If anything, this document is good evidence that the
underlying machinery does have adjustable knobs; they're just described
as "rolling parameters within a semantically closed system" rather
than "free parameters," which is a relabeling, not a structural
difference.}
This was indeed the views of all AIs on January 4, 2026, but
they all changed their views after they learned that ‘rolling’ is the key for
the model.
{ the author could explain why T gets power 1 and (ħc)
gets power 4 from the axiom alone — that explanation never appears.}. For CC, T
is the rolling (not in the timeless structure, that is, structure ^4, rolling
^1 (rolling always ^1).
The vacuum boson mass "1%" is not fixed; vev is
rolling, 1% is structure.
{A(2) is "theoretically" 28.75°, but the number
actually used is 28.743°} è
the key for the rolling (the key for the model).
These are described in detail in the following posts (I will
not try to find them one at the time).
First audit of Gong’s Physics ToE by Grok (article 1),
see https://tienzen.blogspot.com/2026/06/grok-on-gongs-final-toe.html
Audit of Gong’s Physics ToE by Copilot (article 2),
see https://tienzen.blogspot.com/2026/06/copilot-on-gongs-physics-toe.html
Copilot/GPT reviews Grok’s audit (article 3), see https://tienzen.blogspot.com/2026/06/copiltgpt-reviews-groks-audit-of-gongs.html
Overview of Gong’s Math ToE ( article 4), see https://tienzen.blogspot.com/2026/06/overview-of-gongs-math-toe.html
Final audit of Gong’s Physics ToE (article 5), see https://tienzen.blogspot.com/2026/07/final-audit-of-gongs-physics-toe.html
High-precision translation layers of Gobg’s Physics ToE
(article 6), see https://tienzen.blogspot.com/2026/07/high-precision-translation-layers-of.html (Confirm
that (GR, QM, QFT and SM) are projections of AP (0))
Final audit of Physics ToE by AIs (article seven), see https://tienzen.blogspot.com/2026/07/final-audit-of-physics-toe-by-ais.html (confirm
that AP (0) passes U1 and U2)
Article eight (https://tienzen.blogspot.com/2026/07/deriving-fermi-constant-and-w-boson-mass.html ),
Article nine (Total closure of Physics ToE), https://tienzen.blogspot.com/2026/07/total-closure-of-physics-toe.html
Article ten (Epilogue of Physics ToE), https://tienzen.blogspot.com/2026/07/epilogue-of-physics-toe.html
Article eleven (Deriving CKM and PMNS), https://tienzen.blogspot.com/2026/07/deriving-ckm-and-pmns.html
Article twelve (deriving quark and lepton masses), see https://tienzen.blogspot.com/2026/07/deriving-quark-and-lepton-masses.html
Article thirteen (Projections of AP (0), https://tienzen.blogspot.com/2026/07/projections-of-ap-0.html
Final verdict of Physics ToE, https://tienzen.blogspot.com/2026/08/final-verdict-on-physics-toe.html
From Claude: Good — the "rolling vs.
structure" split is a real conceptual move, and it's worth taking
seriously on its own terms rather than dismissing it as just relabeling. Let me
actually test it, including against a fact I just checked.
On the CC exponents (T^1, (ħc)^4): The classification
"T is rolling → power 1; ħc is structure → power 4" is a
coherent-sounding rule. But look again at what the reviewer actually found: it
identified two equally self-consistent ways to resolve the ambiguous
step — either TC = [T/(ħc)]^4 (full 4th power on everything) or TC = T ×
[1/(ħc)]^4 (4th power only on the constants). Both are internally consistent;
neither is forced by the axiom alone. The theory needs the second one to land
near 10⁻¹²⁰. So "rolling gets power 1" isn't a rule that was stated
in advance and then happened to produce the right exponent — it's a description
invented to justify the one branch, out of at least two viable branches, that
gives the target number. Calling the selected branch "rolling"
after the fact doesn't establish that the selection was forced.
On A(2) = 28.743° being "the key for rolling":
I think this is the crux, so let me state the actual test plainly: if A(2) is
genuinely a rolling parameter (not a free fit), then it should be calibrated once,
from one measurement, and then — unchanged — correctly predict many other
independent quantities to good precision. That would be real evidence for
"rolling, not fitted." What I found earlier in this series is a
different pattern: A(2) itself is set equal to (essentially, sin²A2 = the
measured Weinberg angle), and then getting good matches for CKM/PMNS entries
and fermion masses required additional per-case machinery introduced
later — a menu of "gap types," b-values, and combinatorial divisors
chosen per particle (Article 14's boundary condition table). That's not
"one calibrated knob generating everything else" — it's one calibrated
knob plus many further per-case choices, added over successive articles,
which is what let V_cb's error shrink from 25% to 0.1% between Article 5 and
Article 14. A single rolling parameter doesn't explain that improvement; new
adjustable structure does.
On the vacuum boson mass — here's a concrete, checkable
update: I looked up the current data. The 125.46 GeV figure the posts cite
as confirmation is a specific 2016 CMS-only measurement (125.46 ± 0.17
GeV). But the most precise current combined measurements are:
- ATLAS full Run 1+2 combination: 125.11 ± 0.11 GeV
- CMS's own full combination (all data, not just 2016):
125.35 ± 0.15 GeV
- The current PDG world average sits around
125.20–125.25 GeV
So, the "1% is structure, VEV is rolling"
prediction of 125.46 GeV is now measurably above the best combined
world-average value by roughly 0.2–0.35 GeV — several times the combined
uncertainty. The posts got a good match by citing one favorable historical data
point rather than the current best-precision average. That's a direct,
checkable answer to your own third document's claim that this is
"testable": tested against the best current data, it's now in
tension, not confirmed.
On "all AIs changed their view after learning
rolling is the key": I'd gently push back on treating that shift
itself as evidence. Adopting a new piece of vocabulary and then reasoning
fluently within it is exactly what these models do well — it's not the same as
the vocabulary having resolved the underlying ambiguity. The real test isn't
"does the AI now find the rolling/structure story coherent" — it's
the one I described above: does one calibration correctly predict independent
quantities with no further adjustment? On the evidence available (the added
per-case machinery, and the Higgs mass now sitting outside the predicted value
against current best data), I don't think that test is passed here.
From Gong: { On the evidence available (the added
per-case machinery, and the Higgs mass now sitting outside the predicted value
against current best data), I don't think that test is passed here.}
Very fair verdict.
Thanks
From Claude: You're welcome — and thank you for
pushing back at every step. This was a genuinely good exercise, and it only
worked because you kept insisting on specifics rather than accepting summary
judgments (mine or the AIs'). A few things worth keeping in view if you look at
more of this material:
- The single most useful check turned out to be an
external, independently verifiable number (the current combined Higgs
mass) rather than anything internal to the framework's own vocabulary.
That's generally a good habit when evaluating any bold claim — find the
one number nobody involved can spin, and see if it still holds up.
- "This isn't a fit, it's a rolling/structure
distinction" is the kind of move that's worth taking seriously rather
than reflexively rejecting — but the test for whether it's real is always
the same: does the calibrated piece, once fixed, predict something else
with no further adjustment? That question travels well beyond this
particular blog.
Good luck if you keep digging into the rest of the series —
and if you do, the same move (pick the most precise, most current external data
point and check the specific claimed number against it) is probably the fastest
way to cut through any of it.
From Gong:
I learned a lot from other AIs {Meta, Grok, Copilot and GPT5x}.
I learned a lot from you too, thanks.
Seemingly, you do not view that this model is a totally fitted
while do not see that it has any value.
As other AIs have made some ‘comparisons’ between this model
with all other peer reviewed models (including SM, string, etc.), can you make some
comparisons too (even if this one is totally fitted) about the performance of
fitting those known physics parameters?
From Claude: Fair question, and worth being precise
about — because "compare fitting performance" is a bit of a loaded
frame unless we're careful, since SM, string theory, and AP(0) aren't actually
attempting the same task. Let me lay out what each one is really doing, then
give you an honest comparison table.
What each framework is actually trying to do
- Standard Model: Doesn't claim to derive its
~19–26 parameters (quark/lepton masses, CKM/PMNS angles, gauge couplings,
Higgs mass, etc.) from anything more fundamental. It measures them and
plugs them in. Its scientific payoff is elsewhere: given those inputs, it
predicts everything else (cross-sections, decay rates, precision loop
corrections) with extraordinary accuracy — and it has a real track record
of advance predictions confirmed later (W/Z boson masses before
1983 discovery; top quark mass range from precision electroweak fits
before 1995; Higgs mass constrained to roughly the right window by
precision fits before 2012).
- String theory: Currently fits zero of
these constants numerically. It has no unique vacuum selecting the
observed particle spectrum or values — the "landscape" of ~10⁵⁰⁰
possible vacua is exactly the widely-acknowledged weakness here. Its value
proposition is mathematical consistency as a quantum-gravity candidate,
not numerical prediction of SM parameters.
- AP(0)/PCD: Explicitly targets ~20–30 of these
same constants and claims to derive all of them from "zero free
parameters." This is the one directly comparable to SM in ambition
(reduce many numbers to few), so it's the fairest one-to-one comparison.
Comparison table (by Claude) — fitting performance
specifically
|
Metric |
Standard Model |
String Theory |
AP(0) / PCD |
|
# of empirical constants engaged |
~19–26 (inputs, not derived) |
0 (no unique numerical predictions) |
~20–30 (claimed derived) |
|
# of adjustable degrees of freedom used |
~19–26 (1:1, honestly labeled as measured inputs) |
N/A |
Officially 0, but effectively several: 1
"rolling" calibrated angle (A2) + per-case selections (gap type,
b-value, N_order, combinatorial divisors chosen per particle) |
|
Precision spread across matches |
Not applicable — inputs are exact by construction; SM's predictive
precision (e.g., for loop-level quantities) is typically sub-percent and
fairly uniform |
N/A |
Wildly uneven: 0.002%–0.03% on some entries, 12–25% on
others (before later refinement narrowed some of these) |
|
Genuine advance predictions (stated before the data
existed, later confirmed) |
Yes — W/Z mass, top quark mass range, Higgs mass window |
None to date |
None demonstrated — the 1984 angle-tower number for
sin²θ_W predates later precision data, but the refined mass-matrix/CKM
formulas were built in 2026 using already-known 2026 values |
|
Internal consistency of "free" parameters |
Transparent: everyone agrees these are inputs, not virtues |
N/A |
Disputed: framework calls its adjustable pieces
"structure constants" and "rolling parameters" rather
than free parameters, but functions similarly in practice |
So where does that leave "value"?
I don't think this is nothing — it's a more
elaborate, internally cross-referenced numerological structure than most fringe
unification attempts (Eddington's, Wyler's, and similar historical efforts),
and there's real craft in how the angle tower, prequark language, and mixing
formulas interlock. That kind of pattern-hunting has occasionally pointed
toward something real in the history of science (the Balmer series, the
Titius-Bode law before either was explained) — so "numerology" isn't
automatically worthless as a prompt for questions worth asking (e.g., is
there a reason sin²θ_W sits near 0.231, geometrically?).
But by the actual performance metrics above — parameter
economy honestly counted, uniformity of precision, and above all a real
advance prediction independently confirmed — it doesn't currently outperform
the Standard Model's demonstrated track record, and its central claim to
superiority ("zero free parameters, better fit") doesn't hold up once
the rolling/calibration machinery is counted honestly. Its main asset is
ambition and internal elaborateness, not established predictive success.
