By @josebrox
# Audited reductions from Gabe Conant with Astra 6.2 and further improvements — 19 November 2026
The two supplied bundles of notes by Gabe Conant with Astra 6.2 pass the audit of their four principal variable reductions: **193 to 165**, **229 to 180**, **68 to 62**, and **13 to 11**. The conclusions retain the arithmetic premises named in the source papers. The finite checks do not establish those premises. In particular, “general” in this guide means no additional `abc` assumption beyond the retained source inputs.
All variable counts include the input coordinate. Quartic degrees, supports and heights below are exact. The degrees of the large finite-algebra constructions are upper bounds for complete determinant specifications; these enormous polynomials have not been fully expanded. No global minimum is claimed.
The subsequent [v79 continuation](results_20261119_v79.md) reduces the ordinary-abc quartic to 59 variables. The numerical records below describe the v78 checkpoint and remain independently reproducible.
## Records at the v78 checkpoint
| Construction | Additional height hypothesis | Variables | Degree | Further information |
|:--|:--|--:|--:|:--|
| General few-variable family | None | 25 | ≤ 87,778,394,112 | Four-contact scalar, positive factors, projective two-root encoder; sufficient odd `L` ineffective. |
| Explicit general polynomial | None | 27 | ≤ 335,647,080,448 | `L=143` is sufficient; coefficient height < `2^(2^46)`. |
| General quartic family | None | 165 | 4 | 139 quadratic squares, 2,607 monomials; exact height below. |
| Explicit quartic | None | 180 | 4 | 152 quadratic squares, 2,850 monomials; 108-bit exact height. |
| Few-variable family | Weak `abc`, κ=121/100 | 11 | ≤ 32,542,081 | Rank 128; sufficient fixed odd shift ineffective. |
| Few-variable family | Weak `abc`, κ=16/15 | 11 | ≤ 19,366,273 | Rank 128; sufficient fixed odd shift ineffective. |
| Few-variable family | Ordinary rational `abc` | 11 | ≤ 6,361,985 | Twelve contacts at the first height scale; rank 128. |
| Small-height quartic | Ordinary rational `abc` | 62 | 4 | 50 quadratic squares, 840 monomials, height `B+1` on the stated cofinal family. |
| Weaker-hypothesis quartic | Weak `abc`, κ=9/8 | 63 | 4 | 51 quadratic squares, 863 monomials. |
| Weaker-hypothesis quartic | Weak `abc`, κ=17/14 | 64 | 4 | 52 quadratic squares, 866 monomials. |
| Eight-contact quartic | Weak `abc`, κ=7/5 | 67 | 4 | 55 quadratic squares, 905 monomials. |
The proofs are [shared-product quartics](../papers/quartic/shared_product_audit_20261119_v78.tex), [62-variable and weak-abc quartics](../papers/quartic_existence/abc_quartic_audit_20261119_v78.tex), [weak-abc eleven-variable transfer](../papers/conditional_few_variables/weak11_audit_and_transfer_20261119_v78.tex), [four-contact scalar](../papers/few_variables25/four_contact_scalar_20261119_v78.tex), and [ordinary-abc twelve-contact construction](../papers/conditional_few_variables/twelve_contact_first_scale_20261119_v78.tex). The [machine-readable result table](../data/results_20261119_v78.json) identifies the exact certificates and parameter qualifications.
## How Gabe Conant with Astra 6.2 reduced the variable counts
### One shared detector for several contact blocks
The decisive general-quartic change is to share a four-factor detector product instead of keeping a separate detector for each contact block. Write `c=Σq_j`, where the `q_j` are reciprocal contact weights. The shifted norm equation
```math
D\operatorname{Tr}(W)=\operatorname{Nm}(W-1)
```
has no zero-denominator rational solution: if the trace were zero, the equation would give `W=1` in the quadratic field, whose trace is two. The contact-specific gate uses `D` and the other reciprocal weights to recover the required valuation at some conjugate slot. The four local valuation classes remain distinct. Signed hyperbolic pairs then allow the products to be compiled using integral Laurent weights and fewer helper coordinates.
The audit checks the shared selector, the physical-to-quadratic maps, synchronization of the elliptic comparison labels, vanishing test values, and the real and local preparation used for integer witnesses. Every published quartic coefficient is independently reconstructed.
The new continuation raises the odd terms in the first three shared factors from `c U²−c³ Z²` to `c⁵ U²−c⁷ Z²`. The old prepared contact depths still suffice, and the valuation classes remain distinct. This strengthens the extracted divisibilities to
```math
M_j^4\mid\Delta,\qquad M_j^3\mid\operatorname{num}(\beta-n)
```
when the latter numerator is nonzero. The improved Laurent table still needs 165 coordinates and 139 quadrics, with one fewer monomial than the supplied table.
### The 62-variable ordinary-abc quartic
Gabe Conant with Astra 6.2 also shares its two-contact detector and makes an invertible reference-chart change. With the chart outputs `C,I,J,K'`, the field detector is
```math
W=1+cC^4+c^2J^2-c^3\overline I^{4}-c^4\overline{K'}^{2}.
```
Even valuations of the reciprocal-abscissa products make the four negative residue classes distinct modulo four. This supplies extraction even if an individual test or gate vanishes. Tangencies, zero numerator ordinates and finite two-torsion differences remain in the chart. The input zero has a separate checked witness.
Our further improvement removes the constant from the two gate equations:
```math
2\tau_j=q_{1-j}^2+q_jD^2,\qquad q_jb_j=\tau_j^2.
```
The physical pole input now tends to zero at real infinity. A small-input version of the same retained simultaneous-twist premise proves completeness: it suffices that `0<b<1`, that the reduced denominator is a square, and that `b` is a two-adic square of valuation at least four. The proof does not require every odd numerator prime to be `1 mod 4`.
The primitive constant-free presentation has **825 monomials**, and a sparse weighted presentation has **752**. The preferred small-height presentation has **840**, because its coefficient normalization serves a different objective. All have 62 variables, degree four and content one.
### Eleven variables under weak abc
The Gabe Conant and Astra 6.2's eight-contact map supplies the rate
```math
\gamma_8=12/\kappa-8.
```
It is positive for every fixed `1<κ<3/2`. A rational scalar with three positive denominator factors combines the two contact values with the height scale so that one intrinsic pole block suffices. The scalar itself may have either sign on a general rational tuple. The count is one input, six point coordinates, one contact parameter, two pole parameters and one encoder: **eleven**.
The independent review checks positive and negative odd valuations, the order of fixed-place and selected-prime preparation, all point-chart denominators, and every finite-algebra character. Our continuation enlarges the auxiliary point charts, removes guards only in their proved recovery order, and normalizes the intrinsic input by an existing nonzero square. This lowers the two named degree bounds from **33,344,385** and **19,861,377** to **32,542,081** and **19,366,273**.
## Further degree reductions
### The four-contact scalar for 25 and 27 coordinates
Set
```math
\beta=1632X^2+2477,\quad \tau=X^2+1,\quad
A=\beta(\beta-1),\quad B_0=(\beta-2401)(32\beta+343),
```
```math
T_L=[1224(67^{(L+1)/2}+1)]^2,\qquad
\rho=102^2\,67^L\frac{T_LAB_0}{(B_0+T_LA)^2}.
```
Here `L≥3` is odd. For every rational input, the scalar has exactly `v_67(ρ)=L`, a square reduced denominator, and a small positive real value. For integral input it also has the required prepared local square classes at 2, 3 and 17. Its logarithmic height satisfies
```math
h(\rho)\le4h(\beta)+2L\log67+128.
```
The target `τ=X²+1` is always a positive nonzero integer at integer input, and rational `X` with integral `τ` is integral. The exact height comparison `h(τ)≤h(β)` preserves the target-height argument. The scalar therefore needs only four contact factors. In the explicit 27-coordinate construction the outer squares on the two varying scales can also be removed. The effective height ledger, including its increased constants, proves that **every odd `L≥143` suffices**.
The positive factors preserve local valuation extraction while making the selected rational-character traces and norms nonzero. The projective two-root separator gives the better general 25-coordinate bound. Automatic base-coordinate positivity is proved before character extraction; positivity of detector expressions is used only after their denominators have been recovered.
The final degrees and support bounds are:
| Variables | Degree upper bound | Support upper bound | Parameter |
|--:|--:|--:|:--|
| 25 | 87,778,394,112 | `<2^814` | Sufficient odd `L`, ineffective |
| 27 | 335,647,080,448 | `<2^929` | Explicit `L=143` |
For the explicit row, the coefficient one-norm is at most `2^53113872449536`, hence the coefficient height is below `2^(2^46)`. The stored `L=143` data for the general row is only an algebraic fixture; the effective claim belongs to the three-scale 27-coordinate route.
### Twelve contacts at the first height scale
For the ordinary-abc eleven-coordinate source put
```math
\beta=X^2+B,\quad T=\operatorname{Tr}x(A),\quad
M=\beta^4+T^4,\quad a=M.
```
Use the twelve shifts
```math
\{-9,-8,-7,-5,-4,-3,-2,-1,6,10,11,12\},
```
and the derivative quotient
```math
F=\prod_c((a+c)^2-z^2),\qquad u=-24F/F_a.
```
The stronger intermediate height estimate in the original derivative argument applies to the first scale. It yields `log r⁺(u)≥(39/40)h(M)−O(1)`. With the original binary-power coefficient `997/1001`, the strict margin is **18863/40040**. No division by `a²−z²` occurs, so its former guard is unnecessary.
The missing two-adic preparation is supplied by a new all-integer trace lemma. For the fixed elliptic generator `P` and every nonzero integer `n`,
```math
n^2\operatorname{Tr}x([n]P)-57/4\in16\mathbb Z_2.
```
Thus the trace has valuation `−2−2v₂(n)` and is itself a two-adic square. The prepared scale `M` is consequently a two-adic square. The proof uses an integral formal group at two, followed by exact duplication; it is not extrapolated from a finite list of multiples.
There is also a reusable intrinsic-pole identity. If `e=dt²` and `t≠0`, the substitutions
```math
v_{\rm old}=tv,\qquad Y_{\rm old}=t^3Y
```
replace the old equation by
```math
Y^2=d(d+v^2)^2s^4-d(d-v^2)^2(1-e)^2.
```
This is a rational bijection, including zero ordinates. For the actual detector one can take `d=7u`. Removing this known square before compiling the finite algebra greatly reduces its degree.
The final guard degree is 650, the separator degree is 49,052 and the algebra rank is 128. Hence
```math
1+128(1+650+49052)=6,361,985.
```
The support is less than `2^212`. The sufficient odd shift is still ineffective. The `B=3` fixture is certified only as a polynomial specification, not as a defining member.
### Automatic elliptic point guards
This simplification applies across all three few-variable routes. In a square-abscissa root chart, `j²=g⁴−D`, where `D=11−6√2`. Its norm radicand cannot vanish at a rational base, since that would force `(Nm g)^4=49`. For a full auxiliary chart, its retained ordinate norm forces the abscissa norm to be nonzero. These facts establish the needed point étaleness before character extraction. After a rational character is extracted, total positivity of `D` excludes `g=0`.
Consequently two root norm atoms and one abscissa norm atom per full auxiliary chart are redundant. Their deletion saves guard degree 14 for eleven coordinates and 50 for the four-root/five-auxiliary constructions. The proof distinguishes integral algebra formation, character extraction and subsequent rational denominator evaluation.
## Exact coefficient and sparsity improvements
The enhanced 165-variable family has
```math
H(F_N)=\max\{126616122944180358034867798936,\,149^{2N+2}\},\qquad N\ge1.
```
This is an exact coefficient statement for all those `N`. A sufficient defining `N` exists under the retained assumptions, but no numerical sufficient value is supplied.
For the explicit 180-variable quartic, the stronger local divisibilities reduce the necessary anchor to
```math
A=51132150162547763354407979038657,\qquad A^3\ge2^{316}.
```
The independent certificate checks every anchor prime and residue, the effective height endpoint, every row transformation, all coefficients and the literal text. Its exact height is
```math
H=206625078319751697390503348692624,
```
which is **108 bits**, compared with 161 bits in the supplied 180-variable polynomial. The new literal has **59,084 bytes** and 2,850 monomials. The Gabe Conant and Astra 6.2's **54,695-byte, 2,648-monomial** presentation remains available when sparsity or text length matters more than height. The smaller height does not imply a smaller text file.
For the small-height 62-variable family, write `B=n⁴` with positive odd `n`. Clearing four reference constants using the existing unit equation yields
```math
H=\max(4091n^2,n^4+1).
```
Therefore `H=B+1` for every odd `n≥65`. This cutoff certifies the height formula; it is not a sufficient defining threshold. The sparse 752-monomial alternative has exact height `B+758541`.
## Weaker hypotheses and generalizations
The supplied 63- and 64-variable quartics work under weaker hypotheses than claimed in their initial tables: `κ=9/8` and `κ=17/14`, respectively, with strict absorption coefficients `18/19` and `68/71`.
For six contacts and gap exponent `2^r`, a sufficient condition is
```math
[2^r-\kappa(2^{r-1}+2)](8/\kappa-6)>4(2\kappa-1),
```
using `59+r` coordinates. Eight contacts replace the height rate by `12/κ−8` and use `61+r` coordinates. Thus for every fixed **`1<κ<3/2`**, taking `r` sufficiently large gives a quartic family under that weak-abc hypothesis and the other retained premises.
The explicit integer contacts are `{1,8,22,28,29,36,43,64}`. They come from the exact degree-six identity
```math
(v+4)(v+20)^3(v-5)^2-(v-4)(v-20)^3(v+5)^2=108v^5
```
under `t=43+210/(v−10)`. The transformation has diameter 63 and avoids zero. The new 63-, 64- and 67-variable integer-contact families all have height `B+1` on sufficiently large odd fourth-power subfamilies. Their numerical coefficient-height cutoffs, respectively 17,207, 17,207 and 192,405, are again separate from their ineffective defining thresholds.
The shared selector also extends to a degree-`d` number field using
```math
D e_{d-1}(W)=\operatorname{Nm}(W-1),
```
where `e_{d-1}` is the penultimate elementary symmetric function. Its denominator cannot vanish at a rational solution. A slot configuration with at least one conjugate near one and all others poles gives positive valuation of `D`, after excluding the finitely many small residue characteristics that can cancel the leading coefficient. This arithmetic extension does not by itself keep the norm equations quadratic.
## Ancillary obstructions and the 21/9 targets
The [ancillary audit](../negative_results/papers/pole_parameters/ancillary_audit_20261119_v78.tex) verifies the supplied diagonal Hasse-principle failure at `d=112/289`, the zero-ordinate obstruction at `d=112/225`, and the qualified fixed-slice and selector-density results.
The Hasse obstruction now has an independent primitive-pair enumeration modulo 256, 27, 49 and 41, leaving respectively 32, 16, 4 and zero possible square classes. The elliptic curve used for the zero-ordinate obstruction has an independently certified nonzero first derivative of its L-function; the proof then uses the standard analytic-rank-one theorem, not an assumed BSD equality.
These are obstructions to specific universal specializations on all prepared inputs. Neither example is proved to be an actual detector value that must occur for an integer input. The density argument concerns preserving every old prepared tuple. It supplies no general lower bound of eleven variables.
No complete **21-variable general** or **9-variable ordinary-abc** defining polynomial is established here. Their actual physical rational realization conditions remain open. The current general few-variable count is 25; the current weak-abc and ordinary-abc count is 11. The unsuccessful searches for a further quartic coordinate reduction do not prove optimality of 165, 180 or 62.
## Reproduction and provenance
The [computation package](../computation/coauthor_audit_v78) preserves the contributions of Gabe Conant with Astra 6.2 separately from the new audit and extensions. The supplied archives were checked against their complete checksum lists, and all 590 distinct bundled upstream source files agree with the pinned project source. The original README snapshot is retained explicitly because the current README is updated by this release.
Run:
```sh
python3 -B scripts/verify_v78.py --report /tmp/v78.json
python3 -B scripts/verify.py --full --receipt /tmp/legacy.json
```
The new gate stages all inputs in a temporary directory, runs both original suites, rebuilds the new specifications, invokes independent checkers, and byte-compares regenerated published data. Its dependencies are Python 3.12 and SymPy 1.14.0. No external computer-algebra server, optimizer, database connection or floating-point rank heuristic is needed to reproduce the decisive finite certificates.
For a standalone audit layout:
```sh
python3 -B scripts/verify_v78.py --materialize /tmp/v78-offline
```
The papers state the universal mathematical arguments and every retained external premise. The receipts record only what the finite procedures verify. The publication manifest pins all delivered code, proofs and specifications, and the TeX review receipts pin the internally rendered and inspected versions.