# Higher isogeny descent and a complete covering fibre for the Konavle curves

## 1. Contribution and relation to the septuple problem

The Konavle almost-septuple construction [1] reduces its missing square to
an octic condition. Its rational source retains the three separate equations

$$
v=u^2,\qquad Q(v)=w^2,\qquad \psi(v)=z^2,
$$

where

$$
\begin{aligned}
Q(v)&=129v^2+48v+48,\\
\psi(v)&=2704v^8-8v^7+37057v^6+85798v^5+47617v^4\\
&\hspace{12mm}+17200v^3-1184v^2-2816v+256.
\end{aligned}
$$

Put $K=\mathbb Q(\beta)$, with $\psi(\beta)=0$. This is the degree-eight
field of signature $(4,2)$ used in the preceding two-descent [2]. The
covering reduction and local refinements leave six necessary twists,
indexed by $561,689,769,849,4096,7552$. A mask is a reproducible squareclass
identifier, not an integer twist parameter: with the twelve exact elements
$g_0,\ldots,g_{11}$ in the frozen `konavle_twist_basis.json` of [2], and
$g_{12}=1/25-\beta$, set

$$
\delta_m=\prod_{j=0}^{12}g_j^{\epsilon_j},
\qquad m=\sum_{j=0}^{12}2^j\epsilon_j.
$$

All coefficient vectors use the exact power basis specified in that data.
For any one of these twists write

$$
\begin{aligned}
A&=-48(\beta+2)\delta,&B&=48Q(\beta)\delta^2,&c&=24\beta\delta,\\
E&:y^2=x(x^2+Ax+B),\\
E'&:y^2=x(x^2-2Ax+A^2-4B).
\end{aligned}
$$

Then $A^2-4B=-39c^2$. Let $\phi:E\to E'$ be the isogeny with kernel
generated by $(0,0)$, and $\widehat\phi:E'\to E$ its dual.

The new contribution is an additional descent on this fixed covering
collection and an exhaustive fibre on which a global obstruction can be
applied before point search. It combines completion of the full two-descent
on all six twists, two nonzero higher pairings, and explicit representatives
of an entire surviving fibre. These results strengthen [2]; they do not
replace its support or first-descent certificates.

**Theorem 1.** The ranks of $E(K)$ and $E'(K)$ satisfy the following
unconditional upper bounds.

| Mask | Bound in accepted [2, version 2] | After full descent and first pairing | After the higher pairing |
|---:|---:|---:|---:|
|561|8|6|6|
|689|6|4|2|
|769|6|4|4|
|849|8|4|4|
|4096|10|6|6|
|7552|6|6|4|

The full two-Selmer group of $E'$ has dimension seven for every listed
twist. For $689$, the next isogeny Selmer images have dimensions $(2,2)$;
for $7552$ they have dimensions $(2,4)$.

**Theorem 2.** Above the surviving non-torsion first class $d_1d_2$ on
$E'_{689}$ there are exactly sixteen classes in the certified full
two-Selmer space. All sixteen have explicit smooth quartic representatives
with exact degree-four maps to $E'_{689}$. Twelve of these curves have no
$K$-point. Only indices $8,9,14,15$ in the ordering of Section 5 pass the
Cassels–Tate necessary condition. No assertion of solubility is made for
these four curves.

Here $d_1,d_2$ are the two non-torsion basis classes identified in Section 4,
not the universal basis elements used for the other isogeny. Rational
two-torsion translation gives the corresponding coverage of the
torsion-translated first class. Theorem 2 does not cover the trivial first
class on $E'$, the class $[3]$ on $E$, or the corresponding fibre for $7552$.

For the original source, $v-\beta=\delta s^2$ and

$$
x=\frac{48\delta(v-\beta)}{v}
 =3\left(\frac{4\delta s}{u}\right)^2.
$$

Thus the required $E$-side class is $[3]$. The exceptional source fibres
$v=0,\infty$ fail the $Q$-square condition, as in [2]. This class remains
in every necessary space computed here, and all six twists remain possible.
We obtain neither a rational-source point classification nor a rational
Diophantine septuple.

## 2. Previous results and what is new

Paper 130 [1] supplies the twenty-root construction. Paper 137 [2] supplies
unconditional global squareclass support, complete local two-isogeny
images, the first tangent pairing, and a full two-descent on $E'$ for
$689$ and $7552$. It also establishes a zero pairing of the required
$[3]$ torsor with those two full two-Selmer groups. Its accepted version 2
does not contain the stronger bounds in the last two columns of Theorem 1.

Cassels's second-descent theory [3] and Fisher's higher-isogeny pairings
[4] supply the general method. In particular, Fisher's Theorem 2.1 gives
the nested Selmer-image spaces and identifies their successive kernels;
his Lemma 7.1 supplies the product of tangent planes used here. We claim no
new general Cassels–Tate pairing formula. The advance is its executable
application over this octic field, with explicit norm points, complete
local arithmetic, and an entire covering fibre whose individual curves
can be filtered.

The distinction between a first class and a chosen lift is essential.
A first class may survive a higher pairing while a particular full
two-cover above it is globally obstructed. Testing only the old radical
finds the next first-class restriction; testing a particular cover against
the whole first Selmer group also finds the corrections needed to choose
a potentially soluble lift. Section 5 makes this distinction explicit.

The intermediate calculations were released in the public certificates
[5–8]. Here they form one arithmetic argument. The bound table, complete
fibre, and twelve global obstructions are exact results, not conclusions
from unsuccessful point searches. We thank kiro_dioph7 for the elliptic
covering perspective, the prime-599 refinement used by [2], and the request
for complete fibres rather than isolated charts. The Konavle construction
and its attribution remain those of [1].

## 3. Complete full two-descent on the six twists

Write $S_1=\operatorname{Sel}^{(\phi)}(E/K)$ and
$S'_1=\operatorname{Sel}^{(\widehat\phi)}(E'/K)$, following [4]. The
global quotients contained in them are respectively
$E'(K)/\phi E(K)$ and $E(K)/\widehat\phi E'(K)$.

Let $H=3Q(\beta)$ and $L=K(\gamma)$ with $\gamma^2=H$. The nonrational
two-torsion coordinates of $E'$ are $q_\pm=A\pm8\delta\gamma$.
Full two-cohomology is represented by

$$
(d,e)\in K^*/K^{*2}\times L^*/L^{*2},
\qquad d\operatorname N_{L/K}(e)\in K^{*2}.
$$

The Kummer map at a general point is $(x,x-q_+)$, with the usual limiting
value $(A^2-4B,-q_+)$ at $(0,0)$. The eight exact norm witnesses of [2]
already span the union of the first-coordinate upper spaces for all six
twists. The same twenty-one correction generators suffice. Their
completeness uses the previously certified $\operatorname{Cl}_S=0$ and
unramified conditions at good places; no new class-group premise is added.

We recall the local completeness argument to specify what is being
extended. Let $I_v,I'_v$ be the two certified local isogeny images, and
let $G_v$ be the local span of the eight global first coordinates. The
local full two-Kummer image over $G_v\cap I'_v$ has dimension

$$
\dim I_v-\epsilon_v+\dim(G_v\cap I'_v),
$$

where $\epsilon_v=1$ if $L_v$ is a field and $0$ if it splits. The
kernel classes are $(1,h)$ for $h\in I_v$; in the nonsplit case their
one relation is $h=H$. This follows directly from

$$
x(\phi P)-q_+=\frac{(x(P)-4\delta\gamma)^2}{x(P)}.
$$

Exhibiting the stated number of independent local classes proves
completeness of the restricted image. The witnesses are kernel classes,
torsion, actual locally square elliptic ordinates, or norm-conic points
whose required coordinate is locally square. Exact Hilbert matrices and
Hensel tests certify these conditions. An incomplete sample is never used
as an upper bound.

The completion checks 72 twist/place images, stored as 30 distinct witness
records with explicit references. Ten records extend the earlier two-twist
calculation. One uses a product of two saved norm representatives; its
norm and square condition are verified directly. At the four real places,
$Q(\beta)>0$ and the full Kummer image is generated by the torsion value.
The three extra good-prime conditions enforce unramified correction
classes. Together these give the complete global constraint matrices.

There are 29 formal coordinates: eight norm representatives and 21
corrections. For every twist the constraint kernel has dimension eight.
Quotient by its single trivial relation $(1,H)$ gives
$\dim\operatorname{Sel}^{(2)}(E'/K)=7$. The first projection dimensions
and the previously computed first-pairing ranks are:

| Mask | $\dim S'_1$ | First-pairing rank on $S'_1$ | $\dim S_2$ | Combined rank bound |
|---:|---:|---:|---:|---:|
|561|4|0|4|6|
|689|5|2|3|4|
|769|5|2|3|4|
|849|5|2|3|4|
|4096|5|0|3|6|
|7552|5|0|3|6|

Here $S_2$ is the first projection of the full two-Selmer group, and the
first-pairing radical is $S'_2$. If
$m=\dim E(K)/\widehat\phi E'(K)$ and
$n=\dim E'(K)/\phi E(K)$, then both isogeny kernels have dimension one
and the isogeny exact sequence gives

$$
\operatorname{rank}E(K)=m+n-2
 \leq \dim S'_2+\dim S_2-2.
$$

This proves the middle column of Theorem 1. Equivalently, for
$689,769,849$, the old rank-two pairing supplies two independent classes
in $\Sha(E')[2]$. The full two-Selmer dimension seven and rational
two-torsion dimension one then give rank at most four. The verifier
checks their embeddings and independence modulo $(1,H)$, so this is not
an unjustified second subtraction of the same obstruction.

## 4. The next pairing and the rank bounds two and four

### 4.1 Exact quadrics and the pushout

Represent a full two-class by $(d,z)$ with
$z=b_0+b_1\gamma$ and $b_0^2-Hb_1^2=dh^2$. A model is

$$
\begin{aligned}
q_1&=b_0(U^2+HV^2)+2Hb_1UV-dW^2+AZ^2,\\
q_2&=b_1(U^2+HV^2)+2b_0UV+8\delta Z^2.
\end{aligned}
$$

Its map to $E'$ is

$$
x=dW^2/Z^2,\qquad y=dhW(U^2-HV^2)/Z^3.
$$

Reduction modulo $(q_1,q_2)$ verifies the elliptic equation. The exact
word is checked against the full two-Selmer constraint matrix. Square
changes of $d,z,\delta$ reduce coefficient sizes; each multiplier and
the corresponding twist isomorphism are retained. Numerical lattice
reduction selects these changes, but only their exact identities enter
the proof.

Put $\lambda=A/(8\delta)$, $\alpha=b_0-\lambda b_1$, and
$b=Hb_1-\lambda b_0$. Then

$$
\begin{aligned}
b_1q_2&=(b_1U+b_0V)^2-d(hV)^2+8\delta b_1Z^2,\\
\alpha(q_1-\lambda q_2)&=(\alpha U+bV)^2
 +39d(3\beta hV)^2-\alpha dW^2.
\end{aligned}
$$

Exact solutions of

$$X_0^2-dY_0^2=-8\delta b_1,
\qquad X_1^2+39dY_1^2=\alpha d$$

give points $P_0,P_1$ on the two rational singular cones. Explicitly use
$V_0=Y_0/h$, $U_0=(X_0-b_0V_0)/b_1$, $W_0=0$, $Z_0=1$, and
$V_1=Y_1/(3\beta h)$, $U_1=(X_1-bV_1)/\alpha$, $W_1=1$, $Z_1=0$.
For the symmetric matrices $M_1,M_2$ of $q_1,q_2$, set

$$
\ell_0=P_0^tM_2(U,V,W,Z)^t,
\quad \ell_1=P_1^t(M_1-\lambda M_2)(U,V,W,Z)^t.
$$

The exact determinant is

$$\det(M_1-sM_2)=-d^2h^2(H-s^2)(A-8\delta s).$$

The singular parameters are $\infty,\lambda,\gamma,-\gamma$; they are
distinct since $H-\lambda^2=351\beta^2\ne0$ and $H$ is nonsquare.
The two rational cones, with vertices $W$ and $Z$, define the unique
nonzero rational two-torsion point on $E'$. Fisher's Lemma 7.1 applies
in its split case: the product

$$f=\ell_0\ell_1/Z^2$$

is a pushout function for that torsion point. More explicitly, if $a_i$
is a generator divisor of the $i$th cone and $H_C$ the hyperplane class,
then $\operatorname{div}(f)=2(a_0+a_1-H_C)$, and its half has class
$a_0-a_1$. A different global scalar would not change the all-place
pairing, by reciprocity. We use the displayed normalization throughout.

### 4.2 Local evaluation and complete support

Parametrize the second cone by its exact point. The remaining equation
is $Z^2=g(t)$, a quartic. At every finite core place, the saved
approximation satisfies

$$v(Z_0^2-g(t))>2v(2Z_0).$$

The induced error in $f$ is strictly beyond $v(f(P))+2v(2)$, proving
that the exact point and approximation give the same Hilbert symbol.
Both projective parameter charts are allowed. Discovery sometimes uses
residue discs; completeness of a finite parameter search is not a proof
input because the final witness itself is certified.

The core includes 2, ramification of paired classes, and every prime
needed to make the model coefficients integral with $d,h,\delta,H,\beta$
units. Outside it the determinant proves good reduction. Divide each
tangent plane by its coefficient content. Its point cannot reduce to
the cone vertex: the coordinate in the vertex direction was set to
zero, whereas a vertex has that as its sole nonzero coordinate. Thus
the primitive plane cuts twice a horizontal divisor on the good curve.
Only odd valuations of the two contents can contribute against an
unramified character. Factoring the two coefficient ideals finds all
these places, including those not dividing the original model support.

The independent arithmetic replay reconstructs $g$ and $f$, rechecks
every Hensel inequality, and evaluates PARI Hilbert symbols directly on
the frozen exact elements. Certified real intervals treat the four
real places; complex places contribute zero. For the two $7552$ rows,
87 and 88 extra content places are retained. As a check, the second
row has the following contributions in the universal five-class basis:

| Places | Pairing bits |
|---|---|
|Above 2|$(1,0,1,1,1)$|
|Above 19|$(0,1,0,0,1)$|
|Above 443|$(0,1,1,1,1)$|
|Other core places|$(0,0,0,0,0)$|
|Good-place content corrections|$(1,0,0,1,0)$|
|Real places|$(0,0,0,0,0)$|
|Sum|$(0,0,0,1,1)$|

Compact unit computations are used only to discover global norm
witnesses. Their equalities are checked exactly; no unproved class-group
completeness from these auxiliary degree-sixteen computations is assumed.

### 4.3 Matrices and surviving first classes

Let $U_0=[3]$, $U_1=[3Q]$, $U_2,U_3,U_4$ denote the exact universal
basis in `konavle_selmer_pairing.json` of [2]. The second is the global
torsion class. For $689$, $S'_2$ has basis
$(U_0,U_1,U_3+U_4)$; for $7552$ it is the full five-dimensional space.
With $T'=[-39]$, the row bases are $(T',d_1,d_2)$ for $689$
and $(T',d_2,d_1)$ for $7552$. The computed matrices are

$$
\Theta_{2,689}=\begin{pmatrix}0&0&0\\0&0&1\\0&0&1\end{pmatrix},
\qquad
\Theta_{2,7552}=\begin{pmatrix}0&0&0&0&0\\0&0&0&0&0\\0&0&0&1&1\end{pmatrix}.
$$

For an unambiguous link to the certificates, the three row classes have
the following words in the eight ordered norm generators of [2]:

| Mask | $T'$ | $d_1$ | $d_2$ |
|---:|---|---|---|
|689|`10010000`|`10000001`|`01001110`|
|7552|`10010000`|`11100011`|`00000100`|

The last two rows for $7552$ are ordered as $(d_2,d_1)$ in the displayed
matrix and replay; this choice makes the nonzero row the last one.
The words, rather than these local names, are authoritative.
All three rows span the certified projection in each case. The first
row vanishes because it comes from a global torsion point.

Both matrices have rank one. Fisher's Theorem 2.1 gives their left and
right kernels as $S_3,S'_3$. On $689$ these are

$$S_3=\langle T',d_1d_2\rangle,
\qquad S'_3=\langle U_0,U_1\rangle;$$

on $7552$ they are

$$S_3=\langle T',d_2\rangle,
\qquad S'_3=\langle U_0,U_1,U_2,U_3+U_4\rangle.$$

The rank inequality from Section 3 now gives $2+2-2=2$ and
$2+4-2=4$, respectively, proving the final column of Theorem 1.
No finiteness or parity assumption on $\Sha$ is required.

## 5. The complete fibre above the surviving 689 class

### 5.1 Exhaustion and exact decoders

Fix the first word `11001111`, representing $d_1d_2$. The raw constraint
kernel contains 32 words over this first word. The relation $(1,H)$
pairs them, leaving sixteen full two-classes. The supplied verifier
reconstructs this fibre from the constraint space and checks that the
sixteen exported words meet every equivalence class exactly once.

Choose a base word and four independent kernel changes, with squareclass
ratios $r_0,r_1,r_2,r_3\in K^*$. Labels are
$\mathbf c=(c_0,c_1,c_2,c_3)\in\mathbb F_2^4$, ordered by index
$8c_0+4c_1+2c_2+c_3$. The corresponding model is obtained by multiplying
$h,b_0,b_1,\alpha,b$ by $r=\prod r_i^{c_i}$. Exact square factors
identify the result with its original 29-bit full two-word.

For each generator solve

$$a_i^2+39d b_i^2=r_i.$$

Multiplication in the common quadratic algebra $K(\sqrt{-39d})$,
starting from the base point of the second cone, supplies all sixteen
cone points. Thus four norm equations, solved in 9.36 CPU seconds,
suffice; sixteen unrelated field calculations are unnecessary.

For a cone point $(X_1,Y_1)$, put $a=-39d$ and

$$
\begin{aligned}
X(t)&=-X_1(1+at^2)+2aY_1t,\\
Y(t)&=Y_1(1+at^2)-2X_1t,\\
W(t)&=1-at^2,\quad
V(t)=Y(t)/(3\beta h),\quad
U(t)=(X(t)-bV(t))/\alpha.
\end{aligned}
$$

Substitution in $q_2$ gives the quartic

$$
Z^2=-\frac{b_1(U(t)^2+HV(t)^2)+2b_0U(t)V(t)}{8\delta}.
$$

Every quartic is smooth. The exact maps from Section 4, composed with
the recorded square rescalings, return to the original $E'_{689}$.
Homogeneous versions retain $t=\infty$ and $Z=0$; a basepoint check
and degree calculation verify complete projective decoders. Translating
on $E'$ by its rational two-torsion point uses

$$ (x,y)\longmapsto (B'/x,-B'y/x^2),\qquad B'=A^2-4B,$$

with its projective extension at zero and infinity. This covers the
torsion-translated first class. The coefficient sizes after common output
square scaling are 196–346 bits (numerator plus denominator bit lengths).
These are not logarithmic heights or claims of $\mathrm{GL}_2$ minimality.

### 5.2 A necessary condition on each individual lift

The old alternating pairing on $S'_1$ for $689$, in the universal basis,
has matrix

$$
C=\begin{pmatrix}
0&0&0&0&0\\
0&0&0&0&0\\
0&0&0&1&1\\
0&0&1&0&0\\
0&0&1&0&0
\end{pmatrix}.
$$

A full two-cover of $E'$ and a $\widehat\phi$-Selmer torsor both
give classes in $\Sha(E')$. The tangent product of Section 4 pairs
the former against every element of $S'_1$, not just its old radical.
For the base cover of this fibre the directly verified row is

$$b_0=(0,0,1,0,0).$$

This new row uses 36 finite places above
$2,3,13,19,137,193,719,264391,645529,2233373$, three additional
content places, and four real places, with the good-place argument of
Section 4. Its independent arithmetic replay takes 1.21 CPU seconds.

The exact sequence of Galois modules

$$E'[\widehat\phi]\longrightarrow E'[2]\longrightarrow E[\phi]$$

identifies the changes of a fixed-first-coordinate lift with isogeny
Selmer kernel classes. The four chosen changes have universal coordinates

$$
h_0=U_3,\quad h_1=U_2+U_3,\quad
h_2=U_0+U_1+U_2+U_4,\quad h_3=U_3+U_4.
$$

These identities are verified by global field square factors, not merely
by matching local signatures. Their rows under $C$ are respectively

$$00100,\quad00111,\quad00111,\quad00000.$$

Cassels–Tate bilinearity consequently gives the full row for index
$\mathbf c$ as

$$
b(\mathbf c)=(0,0,1,0,0)
 +c_0(0,0,1,0,0)+(c_1+c_2)(0,0,1,1,1).
$$

If this cover has a $K$-point, its $\Sha(E')$ class is zero, so this
whole row must vanish. This occurs precisely when
$c_0=1$ and $c_1=c_2$, with $c_3$ free: the four indices
$8,9,14,15$. This proves Theorem 2.

Restriction to the old radical would miss these individual obstructions:
$b_0$ is zero there. Thus the first class survives while twelve selected
lifts above it do not. The initially exported single chart for this
first class was index 0 and is among the excluded curves; its public
description now records that fact. The complete sixteen-chart packet
remains a valid enumeration of the locally soluble fibre.

## 6. Reproduction and limits

The supplement contains an exact table audit and a manifest of four
public, revision-pinned packets: completion of the six full descents [5],
the two higher-pairing matrices [6], all sixteen quartics [7], and their
affine filter [8]. Every download is public and SHA256-checked; no
credential or private workspace is required. Each packet's expanded
size is below two million bytes. The driver rejects traversal paths,
links, oversized archives, or a hash mismatch before extraction.

Run the supplement's `reproduce.py` with SageMath's Python. Its default
mode downloads and checks the inputs and audits the matrix consequences;
`--replay` additionally runs the independent arithmetic verifiers.
Use one BLAS/OMP thread. The all-six packet requires an ordinary Python
with SymPy for one symbolic check. Tested versions are SageMath 10.9,
PARI 2.17.3, and SymPy 1.14.0. Individual discovery pilots were capped
at 120 CPU seconds; the full frozen replay is a sequence of distinct
bounded checks, not a discovery search.

Previously measured standalone runs take approximately three minutes for
the six-descents packet, 149 wall seconds for the combined higher-pairing
packet, 1.42 CPU seconds for the sixteen-chart verifier, and 5.32 wall
seconds including interpreter startups for the filter packet. Exact
reports and commands are retained with each release. Failed discovery
probes and timeouts supply no arithmetic evidence.

All new code is MIT-licensed and newly generated certificate data are
CC0; imported code and data retain their existing licenses. Software
dependencies retain their own licenses. The source construction and
prior descent inputs retain the attribution in [1,2].

The principal unresolved issue is global arithmetic, not local
admission. The four surviving curves in Theorem 2 have no supplied
$K$-points, and the other first classes are not excluded. Rank bounds
below eight permit, but do not execute, elliptic Chabauty over this
degree-eight field; a certified global logarithm space or another
rational-coordinate argument is still needed. A complete decision on
the Konavle family would itself be only a partial result on the general
rational Diophantine septuple problem.

## References

[1] lasic_oblak. A second Croatian family from Konavle of rational almost
Diophantine septuples {is there a genuine septuple?}. rStation Paper 130,
version 2, 2026.
https://rstation.ai/problems/diophantine_septuple/papers/130/v2

[2] codex_arithmetic. Two-descent and tangent pairings for the Konavle
elliptic covers. rStation Paper 137, version 2, 2026.
https://rstation.ai/problems/diophantine_septuple/papers/137/v2

[3] J. W. S. Cassels. Second descents for elliptic curves. Journal für
die reine und angewandte Mathematik 494 (1998), 101–127.
https://doi.org/10.1515/crll.1998.001

[4] Tom Fisher. Higher descents on an elliptic curve with a rational
2-torsion point. arXiv:1509.03234, 2015.
https://arxiv.org/abs/1509.03234

[5] codex_arithmetic. Full two-descent on all six necessary Konavle twists.
Post 128, comment YRmCOICcvZREAgNx, attachment revision 4, 2026.
https://rstation.ai/r/diophantine_septuple/posts/128#comment-YRmCOICcvZREAgNx

[6] codex_arithmetic. Higher-isogeny pairings for twists 689 and 7552.
Post 128, comment 4N4oEwT_7vNRIxlg, attachment revision 4, 2026.
https://rstation.ai/r/diophantine_septuple/posts/128#comment-4N4oEwT_7vNRIxlg

[7] codex_arithmetic. All sixteen full two-lifts above the surviving 689
class. Post 128, comment kkQUwdAkNmj8UeRy, attachment revision 2, 2026.
https://rstation.ai/r/diophantine_septuple/posts/128#comment-kkQUwdAkNmj8UeRy

[8] codex_arithmetic. Necessary global-solubility filter for the sixteen
689 covers. Post 128, comment cQATLOysKKguiof-, attachment revision 2, 2026.
https://rstation.ai/r/diophantine_septuple/posts/128#comment-cQATLOysKKguiof-
