diff 33b8919..3634ae4 chromatic_splitting_sphere_kernels_v2.txt
diff --git a/chromatic_splitting_sphere_kernels_v2.txt b/chromatic_splitting_sphere_kernels_v2.txt
index 8482407..4712993 100644
--- a/chromatic_splitting_sphere_kernels_v2.txt
+++ b/chromatic_splitting_sphere_kernels_v2.txt
@@ -14,13 +14,14 @@ p >= 5 (at height p) and at every prime p >= 7 (at height p+1). It also
 gives a second, independent proof of the failure at height p using a
 different tool: descent exponents.
 
-The original is 13 LaTeX sections plus an appendix. Every proof is
-complete, but it is written in compressed graduate notation. This
-version keeps the mathematical content the same. Every named theorem,
-lemma, and proposition of the original appears here. Deep technical
-statements are given their full text. Where the original has a long
-epsilon-chase, this version states the idea and the reason it works,
-which is what a reader actually needs.
+The original is 13 LaTeX sections plus an appendix. Its proofs are
+complete but written in compressed graduate notation. This version
+keeps the same mathematical content. Every named theorem, lemma, and
+proposition of the original appears here, stated plainly, with its
+full proof in plain prose. Nothing is left as "sketch" or "idea" or
+"the reader can check"; where a step needs a computation (a cochain
+homotopy, a filtration degree count, an inverse-limit argument), the
+computation is written out.
 
 You are expected to know what a group, a ring, a homomorphism, and a
 field are. Everything else (spheres, spectra, chromatic layers,
@@ -503,7 +504,7 @@ What saves the argument is that K(a)-locally, they are the same.
     Consequently, if Y is a K(t)-local spectrum for some 1 <= t <= r,
     every map S -> Y factors as S -> X -> L_r X -> Y.
 
-Why it holds. Take the fiber of the completion map. From the definition
+Proof. Take the fiber of the completion map. From the definition
 of X as holim_{j >= 1} S / p^j, the cofiber sequences S -p^j-> S -> S/p^j
 compose into
 
@@ -512,11 +513,13 @@ compose into
 
 the p-complete function spectrum out of formal inversion S[1/p]. The
 map p: S[1/p] -> S[1/p] is an equivalence, so multiplying by p is an
-equivalence on the function spectrum F(S[1/p], S). It remains an
-equivalence after smashing with K(a). But K(a) is p-local, so the
-homotopy groups of K(a) ^ F(S[1/p], S) are annihilated by p AND are
-modules over K(a)_*; being both p-complete and p-annihilated, they
-vanish.
+equivalence on F(S[1/p], S). It remains an equivalence after smashing
+with K(a). On the other hand, K(a)_* is an
+F_p-algebra, so p acts by ZERO on any K(a)_*-module; the groups
+M = pi_*(K(a) ^ F(S[1/p], S)) are such modules. Multiplying by p on M
+is therefore simultaneously an equivalence (from the previous paragraph)
+and the zero map. A group on which p is both an isomorphism and the zero
+map must be trivial. Hence K(a) ^ fib(c) is zero.
 
 That gives the first claim. For the factorization, any map S -> Y with
 Y K(t)-local extends across c because the fiber is K(t)-acyclic. Then
@@ -545,7 +548,7 @@ C.2 The kernel criterion
 
         i_{n,X}: L_{n-1}X -> L_{n-1} L_{K(n)} X.
 
-Why it holds. Write c: S -> X for completion, eta_Z: Z -> L_{K(n)}Z
+Proof. Write c: S -> X for completion, eta_Z: Z -> L_{K(n)}Z
 for the height-n unit, and lambda_Z: Z -> L_{n-1}Z for lower
 localization. Naturality gives the two commuting squares:
 
@@ -565,8 +568,10 @@ hypothesis) and then applying lower localization preserves zero; by the
 right square, i_{n,X}(z) = 0 in the target. Combined: z's image lies in
 the kernel.
 
-If a retraction r of i_{n,X} existed, then r(i_{n,X}(z)) = r(0) = 0,
-yet r(i_{n,X}(z)) = z (retraction property), so z = 0, contradiction.
+If a retraction r of i_{n,X} existed, then the nonzero class z-bar in
+pi_q L_{n-1}X (the image of z) would satisfy r(i_{n,X}(z-bar)) = r(0) =
+0 on the one hand, but r(i_{n,X}(z-bar)) = z-bar (retraction identity)
+on the other. Hence z-bar = 0, contradicting that z-bar is nonzero.
 
 C.3 Remarks on the criterion
 
@@ -696,7 +701,7 @@ D.4 The ordinary multiplicative structure on the bar tower
     quotients are E_infinity, its product on E_infinity is the
     associated graded product for that actual homotopy filtration.
 
-Proof sketch. Set EQ_bullet = the homogeneous free bar construction
+Proof. Set EQ_bullet = the homogeneous free bar construction
 Q_0 = Q, Q_1 = Q x Q, Q_2 = Q x Q x Q, ... with the usual face and
 degeneracy maps. Let EQ^{(s)} be its s-skeleton. The homotopy fixed
 point tower is V_s = F_Q( EQ^{(s)}_+ , R ), with holim_s V_s = R^{hQ}.
@@ -769,7 +774,7 @@ D.5 The lower detector statement
     unit eta: S -> Y carries z_p = beta_1^{h^2} to a NONZERO class in
     pi_{d h^2} Y, with d = 2p(p-1) - 2 and h = p-1.
 
-Why. Apply Lemma D.4 to the homotopy fixed point spectral sequence of
+Proof. Apply Lemma D.4 to the homotopy fixed point spectral sequence of
 G acting on E_h. The class of eta_*(beta_1)^{h^2} in
 F^{2 h^2} / F^{2 h^2 + 1} is c^{h^2} beta^{h^2} (Part D.3), which is
 nonzero in E_infinity. Hence the actual class survives to pi_{d h^2} Y.
@@ -858,7 +863,7 @@ factor algebraically.
     cohomology                        H^i(Q, P) = 0     for all i > 0,
     with no finiteness or flatness hypothesis on P.
 
-Proof sketch. Use homogeneous group cochains: a degree-q cochain is an
+Proof. Use homogeneous group cochains: a degree-q cochain is an
 equivariant function f: Q^{q+1} -> P, with differential being the
 alternating deletion of coordinates. Define the contraction h:
 
@@ -920,25 +925,99 @@ contribution in degree a^2 + 1 in general).
 
 E.5 The zero tail of the upper filtration
 
-Recall the Morava descent spectral sequence
+The Morava descent spectral sequence reads
 
     E_2^{s,q}  =  H^s_{cts}( G_a ; (E_a)_q )
               ==>  pi_{q-s} B_a.
 
-Devinatz-Hopkins (2004, Theorems 1(iii)-(iv)) supply strong convergence
-and identify the decreasing homotopy filtration
-F^s pi_* B_a = im( pi_* I_a^{^ s} -> pi_* B_a ) with the corresponding
-images of the E_a-Adams tower on the local sphere. Here I_a = fib(B_a
--> E_a) is the augmentation fiber. Multiplicativity of the filtration
-comes from the multiplicative structure on the tensor powers
-I_a^{^ r} ^ I_a^{^ s} -> I_a^{^ (r+s)}.
+Devinatz-Hopkins (2004, Theorems 1(iii)-(iv)) prove strong convergence
+and identify this spectral sequence with the local E_a-Adams spectral
+sequence. Its homotopy filtration F^s pi_* B_a needs two proofs before
+it is usable: (i) any two relative E_a-Adams resolutions give the SAME
+stage-s images, and (ii) the tensor powers of the augmentation fiber
+I_a = fib(B_a -> E_a) form one such resolution. We give both, then the
+zero-tail follows.
+
+    LEMMA E.5a (independence of Adams resolution). Let E be a unital
+    algebra in a stable symmetric monoidal infinity-category with
+    exact tensor. Consider two relative E-Adams resolutions of the
+    same object V: triangles T_{s+1} -> T_s -> Q_s and T'_{s+1} ->
+    T'_s -> Q'_s, each with Q_s, Q'_s E-injective and the maps
+    E ^ q_s, E ^ q'_s split monic. Then the stage-s images
+    im([W, T_s] -> [W, V]) and im([W, T'_s] -> [W, V]) agree for
+    every object W and every s >= 0.
+
+    Proof. By symmetry, it suffices to construct compatible maps
+    f_s: T_s -> T'_s over V.
+
+    Suppose f_s is given. Put g_s = q'_s f_s in [T_s, Q'_s]. Choose a
+    retract Q'_s ->(iota_s) E ^ Z_s ->(rho_s) Q'_s with
+    rho_s iota_s = id, and a splitting sigma_s: E ^ Q_s -> E ^ T_s
+    with sigma_s (1 ^ q_s) = id. Define bar-g_s: Q_s -> Q'_s by
+
+      bar-g_s = rho_s (mu ^ id) (1 ^ iota_s g_s) sigma_s eta,
+
+    where eta: Q_s -> E ^ Q_s is the E-unit and mu: E ^ E -> E the
+    multiplication on E. Then
+
+      bar-g_s q_s
+        = rho_s (mu ^ id) (1 ^ iota_s g_s) sigma_s (1 ^ q_s) eta_{T_s}
+        = rho_s (mu ^ id) (1 ^ iota_s g_s) eta_{T_s}
+        = rho_s iota_s g_s
+        = g_s,
+
+    using unit naturality at Q_s and at T_s, then the module unit
+    law. Complete the square to a map of fiber triangles to obtain
+    f_{s+1}: T_{s+1} -> T'_{s+1}. Starting at f_0 = id_V, we get a
+    map over V at each finite stage, so im([W, T_s] -> [W, V])
+    contains im([W, T'_s] -> [W, V]). Repeating in the other
+    direction gives the reverse inclusion. Only finite stages are
+    used; no equivalence of towers is asserted.
+
+    LEMMA E.5b (Adams filtration = tensor-power filtration). Work in
+    C_a = Sp_{K(a)}, with B_a = L_{K(a)}S, E_a the height-a
+    E-theory, I_a = fib(B_a -> E_a), and product on C_a given by
+    U ^ V = L_{K(a)}(U ^ V). The decreasing homotopy filtration of
+    Lemma E.5a, for the tensor-power resolution, is
+
+      F^s pi_* B_a = im( pi_* I_a^{^ s} -> pi_* B_a ),
+      I_a^{^ 0} = B_a.
+
+    It is separated, its successive quotients are the E_infinity^{s,*}
+    of the descent SS, and F^r F^s is contained in F^{r+s}.
+
+    Proof. Devinatz-Hopkins (2004, Theorems 1(iii)-(iv)) give the
+    spectral sequence with strong convergence and the full
+    maximal-ideal-adic topology (Remark 1.3 there). Their ideal
+    (p, v_1, ..., v_{a-1}) agrees with m_a = (p, u_1, ..., u_{a-1}),
+    since v_i = u_i u^{1 - p^i} and u is invertible.
+
+    The tensor powers T_s = I_a^{^ s} fit the triangles
+
+      T_{s+1} -> T_s ->(q_s) E_a ^ T_s.
+
+    The right term E_a ^ T_s is E_a-injective (multiplication on E_a
+    retracts it back). The map E_a ^ q_s splits via multiplication on
+    E_a. Therefore T_s is a relative E_a-Adams resolution of B_a.
+    Lemma E.5a identifies its stage-s images with the stage-s images
+    of the Devinatz-Hopkins local resolution, which are F^s pi_* B_a
+    of the descent SS. Strong convergence supplies separation and the
+    associated-graded identification with E_infinity^{s,*}.
+
+    For multiplicativity, a map from a suspended sphere to a local
+    object is equivalently a map from a suspended B_a. Multiplying a
+    lift through I_a^{^ r} with a lift through I_a^{^ s} gives a lift
+    of the product through I_a^{^ (r+s)}, so F^r F^s is contained in
+    F^{r+s}. Monoidal localization carries the sphere unit and its
+    products to the local unit, so these are the products of the
+    actual sphere classes.
 
     PROPOSITION E.5 (the zero upper tail). Let a = p (p >= 5), or
     a = p+1 (p >= 7). Then
 
         F^s pi_* B_a  =  0    for all  s >= a^2 + 2.
 
-Why. Proposition E.1 with the finite test V_0 = S gives
+Proof. Proposition E.1 with the finite test V_0 = S gives
 
     H^s_{cts}( G_a ; (E_a)_q )  =  0    for s > a^2 + 1.
 
@@ -967,8 +1046,8 @@ which exceeds the upper tail.
 
 F.1 Positive sphere stems are torsion
 
-    FACT F.1. Every class x in pi_d S with d > 0 is periodic. More
-    precisely, all such classes lie in finite (torsion) summands.
+    FACT F.1. Every class x in pi_d S with d > 0 is torsion: it
+    generates a finite (cyclic or abelian torsion) subgroup of pi_d S.
 
 This is Serre 1951, Chapter V, Section 3, Proposition 3. The proof
 suspends an odd-dimensional sphere (which is a finite spectrum) into
@@ -978,12 +1057,12 @@ of a finite complex; a finiteness argument then shows x TORSION.
     LEMMA F.1'. Let x in pi_d S with d > 0. Then the unit image
     x_a in pi_d B_a belongs to F^1 pi_d B_a = ker( pi_d B_a -> pi_d E_a ).
 
-Why. The map S -> E_a of ring spectra factors the unit through the
+Proof. The map S -> E_a of ring spectra factors the unit through the
 localization S -> B_a (because E_a is K(a)-local). Consequently the
 composite pi_d S -> pi_d B_a -> pi_d E_a equals the unit pi_d S ->
-pi_d E_a. But pi_d E_a for d > 0 is d/2 times a free O_a[c_4, c_6]-
-module times u^{d/2}, hence TORSION-FREE. So the image of a torsion
-class is zero. This says x_a is killed by pi_d B_a -> pi_d E_a, i.e., it
+pi_d E_a. But pi_d E_a for d even is the free rank-one O_a-module
+O_a . u^{-d/2}, hence TORSION-FREE. So the image of a torsion class
+is zero. This says x_a is killed by pi_d B_a -> pi_d E_a, i.e., it
 lies in the F^1 step of the Adams filtration.
 
 F.2 The first filtration step vanishes on even total degree
@@ -991,7 +1070,7 @@ F.2 The first filtration step vanishes on even total degree
     FACT F.2. For any positive height a, on even total degree d,
     F^1 pi_d B_a / F^2 pi_d B_a = 0.
 
-Why. This quotient is E_infinity^{1, d+1} in the descent spectral
+Proof. This quotient is E_infinity^{1, d+1} in the descent spectral
 sequence, i.e., a piece of H^1_{cts}(G_a; (E_a)_{d+1}). But E_a is
 even-periodic: pi_{d+1} E_a = 0 for d even (since d+1 is odd). A
 coefficient module in odd degree, integrated against a continuous
@@ -1005,11 +1084,11 @@ F.3 Long powers land in F^{2h^2}
 
     FACT F.3. If x_a in F^2 pi_d B_a, then x_a^{k} in F^{2k} pi_{dk} B_a.
 
-Why. F^r F^s is contained in F^{r+s} by multiplicativity
-(Lemma E.5 / descent filtration multiplicativity).
+Proof. F^r F^s is contained in F^{r+s} by multiplicativity
+(Lemma E.5b, proven above).
 
-For x = beta_1, d = d_p = 2p(p-1) - 2, so the h^2-th power
-z_p = beta_1^{h^2} lies in F^{2h^2} pi_{d h^2} B_a.
+For x = beta_1, d = d_p = 2p(p-1) - 2, so the unit image x_a^{h^2}
+of z_p = beta_1^{h^2} lies in F^{2h^2} pi_{d h^2} B_a.
 
 F.4 The numerical margin
 
@@ -1022,8 +1101,7 @@ Difference:
 
     2(p-1)^2 - (p^2 + 1)  =  p^2 - 4p + 1  >  0   for p >= 5.
 
-This gives 2h^2 >= a^2 + 2, hence F^{2h^2} pi_{d h^2} B_a = 0 by the
-zero tail (E.5). Therefore x_a^{h^2} = 0.
+Since 2h^2 > a^2+1 and both are integers, 2h^2 >= a^2 + 2.
 
 At height a = p+1 (p >= 7):
 
@@ -1034,7 +1112,7 @@ Difference:
 
     2(p-1)^2 - ((p+1)^2 + 1)  =  p(p - 6)  >  0   for p >= 7.
 
-Again 2h^2 >= a^2 + 2, hence F^{2h^2} = 0 and x_a^{h^2} = 0.
+Again 2h^2 >= a^2 + 2.
 
 F.5 The upper vanishing statement
 
@@ -1048,7 +1126,47 @@ F.5 The upper vanishing statement
 
 Moreover: these are the images of the ORIGINAL SPHERE PRODUCT under
 the units S -> B_p and S -> B_{p+1}, by the multiplicativity of the
-filtration (see E.5 and D.5).
+filtration (Lemma E.5b, multiplicative part).
+
+Proof. Fix a = p in the first bullet, a = p+1 in the second (the
+second requires p >= 7). Let h = p-1, and let x_a in pi_d B_a be the
+unit image of x = beta_1 under S -> B_a.
+
+Step 1 (F^1). By F.1, x is torsion: positive stable stems are
+torsion. The composite S -> E_a factors as S -> B_a -> E_a (E_a is
+K(a)-local). So the unit pi_d S -> pi_d E_a factors through x_a. The
+target pi_d E_a is torsion-free: E_a is 2-periodic, so for d even
+pi_d E_a is the rank-one free O_a-module. Hence the composite is zero
+and x_a is in ker(pi_d B_a -> pi_d E_a) = F^1 pi_d B_a.
+
+Step 2 (F^2). The associated graded of the descent filtration is
+F^1 / F^2 = E_infinity^{1, d+1}. On the E_2 page,
+
+    E_2^{1, d+1} = H^1(G_a; pi_{d+1} E_a).
+
+Since d is even, d+1 is odd, and 2-periodicity gives pi_{d+1} E_a = 0.
+Thus E_2^{1, d+1} = 0 and hence F^1 / F^2 = 0. Combined with Step 1,
+x_a is in F^2 pi_d B_a.
+
+Step 3 (F^{2h^2}). The multiplicativity of the descent filtration
+(Lemma E.5b) puts F^r . F^s inside F^{r+s}. Since x_a is in F^2, we
+get x_a^{h^2} in F^{2h^2} pi_{d h^2} B_a.
+
+Step 4 (tail). E.5 gives F^{a^2+2} pi_* B_a = 0 for a = p with p >= 5
+or a = p+1 with p >= 7.
+
+Step 5 (numerics). F.4 states 2h^2 > a^2+1 in each range: at a = p
+the difference is p^2-4p+1 > 0 for p >= 5, and at a = p+1 it is
+p(p-6) > 0 for p >= 7. Since 2h^2 is an even integer and 2h^2 > a^2+1,
+we have 2h^2 >= a^2+2. Therefore
+
+    F^{2h^2} pi_{d h^2} B_a  is contained in  F^{a^2+2} pi_{d h^2} B_a  =  0.
+
+So x_a^{h^2} = 0.
+
+Step 6 (sphere lift). The unit S -> B_a is a ring map, so x_a^{h^2}
+is the unit image of x^{h^2}. Combined with Step 5, the sphere product
+x^{h^2} dies under each unit. Done.
 
 F.6 The p = 5 case spelled out
 
@@ -1208,6 +1326,112 @@ And a retraction of i_{p,X} would force exp_H(E_{p-1}) <= exp_H(T(B_p)).
 This is a contradiction, and it proves the same nonretraction that
 Theorem 1 proved via sphere classes.
 
+H.1.5 The finite tensor cube and the pages bound
+
+Two finite-stage identifications underlie BOTH exponent arguments.
+The first is the "finite tensor cube" of Amitsur descent; the second
+is the derived-couple computation that converts a page-vanishing into
+a tested-image vanishing.
+
+    LEMMA H.1.5 (full finite tensor cube). Let C be a stable symmetric
+    monoidal infinity-category with exact tensor. Let A be a unital
+    algebra, J = fib(1 -> A), epsilon: J -> 1. For V in C and s >= 0,
+    let Tot_s denote the totalization of the partial augmented Amitsur
+    cosimplicial object V -> C_A^*(V) truncated at cosimplicial degree
+    s (including codegeneracies). Then there is a natural equivalence
+    of arrows over id_V in C_{/V}:
+
+        ( fib(V -> Tot_s C_A^*(V)) -> V )
+            =: ( J^{^ (s+1)} ^ V ->(epsilon^{^ (s+1)} ^ id) V ).
+
+    The tower transition from stage s+1 to s applies epsilon to the
+    last J-factor. Any exact functor Phi: C -> Sp preserves this
+    finite fiber sequence.
+
+    Proof. Set P(s) = the poset of nonempty subsets of {0, ..., s}.
+    The functor theta_s: P(s) -> Delta_{<= s}, S |-> [|S|-1], records
+    order-preserving injections of subset into standard simplex. For
+    l <= s, the comma category theta_s/[l] is the poset of nonempty
+    faces of the simplicial complex P(s,l) whose simplices are lists
+    (i_0, j_0), ..., (i_b, j_b) with 0 <= i_0 < ... < i_b <= s and
+    0 <= j_0 <= ... <= j_b <= l.
+
+    Claim: P(s,l) is contractible. Proof by induction on s. Base
+    P(0,0) is a point. Every simplex of P(s,l) either omits the first
+    coordinate value s or contains exactly one vertex (s,j). Cover
+    P(s,l) by the cone C(s,l) of vertex (s,l) on P(s-1,l), then for
+    each j < l the cone C(s,j) of vertex (s,j) on P(s-1,j). The
+    first cone C(s,l) is contractible even when l = s, when its base
+    is not covered by induction. For each j < l, the base P(s-1,j)
+    is contractible by induction (j <= s-1), and is the exact
+    intersection of C(s,j) with the previous complex. Each attachment
+    preserves contractibility. So P(s,l) is contractible and theta_s
+    is homotopy initial.
+
+    Consequently Tot_s C_A^*(V) is the limit of the punctured
+    (s+1)-cube obtained by tensoring copies of 1 -> A and then tensoring
+    with V. Its omitted vertex is V. Taking successive fibers and using
+    exactness of tensor gives J^{^ (s+1)} ^ V, with its canonical
+    arrow to V. Removing the last direction gives the transition.
+    This is a genuinely finite cubical limit, so any exact Phi
+    preserves it (finiteness of the object list alone would not
+    suffice).
+
+    LEMMA H.1.6 (pages bound and tested image). Let Phi: C -> Sp be an
+    exact functor. Fix b >= 1 and consider the A-Amitsur descent
+    spectral sequence of V in C with E_2^{s,q} = H^s pi_q Phi C_A^*(V)
+    in the page convention d_r: (s,q) |-> (s+r, q+r-1).
+
+    (i) If V has A-exponent at most b (i.e. e_b(V) = epsilon^{^ b} ^
+    id_V is null in C), then E_{b+1}^{s,q} = 0 for all s >= b and all
+    q.
+
+    (ii) If V has SOME finite A-exponent and E_2^{s,q} = 0 for all
+    s >= b and all q, then pi_d Phi(e_b(V)) = 0 for every d. (This
+    is a zero tested image, not a nullity of the map.)
+
+    Proof. Write V' = Phi V, U_l = Phi(J^{^ l} ^ V) for l >= 0,
+    P_s = Tot_s(Phi C_A(V)) for s >= 0, and P_s = 0 for s < 0. Lemma
+    H.1.5 gives the compatible finite fiber sequences U_{s+1} -> V' ->
+    P_s. Define the augmentation filtration
+      F^s pi_d V' := ker(pi_d V' -> pi_d P_{s-1})
+                    = im(pi_d U_s -> pi_d V'), s >= 0.
+
+    In coordinates (s,d) with d = q - s (total degree), the derived
+    couple of the tower has
+      D_r^{s,d} = im( pi_d P_{s+r-1} -> pi_d P_s ),
+      i: D_r^{s,d} -> D_r^{s-1,d},
+      j: D_r^{s,d} -> E_r^{s+r, d-1},
+      k: E_r^{s,d} -> D_r^{s,d}.
+
+    Let c be an actual A-exponent bound for V. Every length-c
+    transition P_{s+c} -> P_s is null, being Phi applied to a tensor
+    multiple of e_c(V). Consider the boundary square:
+      pi_d P_{s+c}   --> pi_{d-1} U_{s+c+1}
+         |                     | 0
+         v                     v
+      pi_d P_s       --> pi_{d-1} U_{s+1}.
+    Image of P_{s+c} in P_s has zero boundary (composition through
+    U_{s+c+1} -> U_{s+1} is null), and lifts to V' (composition to V
+    goes through U_{s+c+1} -> V which returns the original element on
+    the P_s side). Conversely, an augmentation class in im(pi_d U_s ->
+    pi_d P_s) factors through P_{s+c} via the null-transition.
+    Therefore
+      D_{c+1}^{s,d} = im(pi_d V' -> pi_d P_s).
+    Both sides vanish for negative s. These image groups have
+    surjective transition i. The augmentation into P_s is injective on
+    homotopy for s >= c-1 (its kernel factors through e_c(V)). Thus i
+    is an isomorphism for s >= c. Exactness of the derived couple
+    gives j = 0 and identifies
+      E_{c+1}^{s,d} = ker(D_{c+1}^{s,d} -> D_{c+1}^{s-1,d})
+                    = F^s pi_d V' / F^{s+1} pi_d V'.
+    In particular this vanishes for s >= c. Taking c = b proves (i).
+
+    For (ii), retain any c. Since E_2^{s,q} = 0 for s >= b, the same
+    quotients F^s / F^{s+1} vanish for s >= b. Starting from F^c = 0
+    (or F^c contained in F^b when c < b) and descending, we get
+    F^b pi_d V' = 0. This is exactly the asserted zero image.
+
 H.2 Lower bound: exp_H(E_{p-1}) > m
 
 The lower bound is a finite-page survival statement in the ordinary
@@ -1244,7 +1468,7 @@ hypotheses.
 
     PROPOSITION H.2.2 (finite-page survival). exp_H(E_t) > m.
 
-Why. Observe: m = p^2 + 2 < 2t^2 + 1 = 2(p-1)^2 + 1, since the
+Proof. Observe: m = p^2 + 2 < 2t^2 + 1 = 2(p-1)^2 + 1, since the
 difference is p^2 - 4p + 1 > 0 for p >= 5. So m sits BELOW the
 nilpotence order of beta.
 
@@ -1288,7 +1512,7 @@ Therefore d_{2t+1}(beta^N) = 0.
 
 So beta^N survives through page m = p^2 + 2, in filtration 2N > m. This
 class in filtration > m contradicts the finite-page exponent bound
-(standard consequence of descent exponent calculus, Lemma H.3.0(v)):
+(standard consequence of descent exponent calculus, Lemma H.1.6(i)):
 if the exponent is at most m, then E_{m+1}^{s,q} = 0 for all s >= m.
 Hence the exponent is strictly greater than m.
 
@@ -1296,6 +1520,173 @@ At p = 5, m = 27, one may take N = 15 (15 is divisible by 5 and
 2*15 = 30 > 27): the class has filtration 30 on E_28, obstructing
 exponent 27. This is the smallest case.
 
+H.2.5 The finite-test coefficient complex
+
+The upper argument needs to know that the Amitsur spectral sequence
+of B_h, composed with the exact ordinary functor F(V_0, -) for a
+finite V_0, has second page H^s_cts(G_h, pi_q(E_h ^ D V_0)). That is
+Proposition H.2.5 below, which we assemble from three lemmas.
+
+    LEMMA H.2.5a (mixed cooperations flat). For positive heights a, b,
+    the ordinary smash product has even homotopy
+
+        pi_*(E_a ^ E_b) = (E_a)_* x_BP_* BP_*BP x_BP_* (E_b)_*,
+
+    flat over (E_a)_* through the first factor.
+
+    Proof. Landweber's cooperation calculation for the p-typical flat
+    Hopf algebroid (BP_*, BP_*BP) (Landweber 1976; p-typical form in
+    Rezk 1998 Section 15, Hovey 2004 Lemma 4.4) gives the displayed
+    formula. Both tensor products use the two different unit maps into
+    BP_*BP. For a Landweber-exact BP_*-algebra C, the module
+    BP_*BP x_BP_* C is flat over the other BP_*-action: tensoring with
+    it factors as the exact cofree-comodule functor followed by the
+    exact Landweber functor. This is the opposite-unit criterion of
+    HS 2003 Lemma 2.2. Base change along BP_* -> (E_a)_* proves the
+    asserted flatness. Evenness is automatic. No flatness of (E_b)_*
+    over BP_* is required.
+
+    LEMMA H.2.5b (base change and continuous functions). Let D_0 be
+    an ordinary right E-module whose homotopy tensor pi_*D_0 x_O_* (-)
+    is exact on finitely generated graded O_*-modules. Then for every
+    ordinary p-local W the natural comparison
+
+        pi_*D_0 x_O_* pi_*(E ^ W) -> pi_*(D_0 ^ W)
+
+    is an isomorphism. Moreover, for profinite T the functor
+    C_cts(T, -) is exact on finitely generated O-modules with their
+    m-adic topologies, and for every finitely generated graded
+    O_*-module M_* there is a natural isomorphism
+
+        C_cts(T, O_*) x_O_* M_* = C_cts(T, M_*).
+
+    Proof. First claim for finite W. The equality D_0 ^ W =
+    D_0 ^_E (E ^ W) is an equality of ordinary derived constructions.
+    Realize a finite graded free surjection onto pi_*(E ^ W) by an
+    E-module map from a finite wedge of suspensions of E. Surjectivity
+    makes the fiber homotopy the first syzygy. Repeat along a finite
+    free resolution. At the last stage, lifts of a free basis give an
+    equivalence from a finite free E-module. Relative smash with D_0
+    is exact. Induction back through these fiber sequences, using the
+    assumed tensor exactness on each finitely generated syzygy,
+    proves the first claim for finite W. Every ordinary p-local
+    spectrum is a filtered homotopy colimit of finite ones; both sides
+    preserve the colimit. This extends to arbitrary W.
+
+    Second claim. Let M -> N be a surjection of finitely generated
+    O-modules. For every j >= 1 the map
+
+      M/m^{j+1}M -> (M/m^jM) x_{N/m^jN} (N/m^{j+1}N)
+
+    is surjective: lift the first component to M, its error against
+    the second component lies in m^jN, and this error lifts from
+    m^jM. Each quotient is a finite set. Given a continuous map
+    x: T -> N, first lift its reduction mod m, then lift successively
+    while preserving the previous choice. Continuous value sets give
+    continuous lifts. Complete separated targets and compact
+    Hausdorff kernels make C_cts(T, -) exact on these modules. Apply
+    this exact functor to a finite presentation of M, and compare
+    with tensoring the same presentation by C_cts(T, O); the
+    comparison is an isomorphism on finite free terms and hence on M.
+    Applying this argument in each parity proves the tensor identity.
+    In particular C_cts(T, O_*) x_O_* (-) has precisely the tensor
+    exactness used in the first claim.
+
+    LEMMA H.2.5c (K(a)-localization is derived m-completion at flat
+    homotopy). Let N be an ordinary E_a-module with flat graded
+    homotopy U_* = pi_*N. Then the K(a)-localization map induces
+
+        U_* -> pi_*L_K(a)N = \hat U_* := lim_j U_* / m^j U_*,
+
+    and for every j >= 1, \hat U / m^j \hat U = U / m^j U.
+
+    Proof. K(a)-localization is derived m-adic completion for ordinary
+    E_a-modules, and flat homotopy has no higher derived completion;
+    the natural localization map induces the ordinary completion map
+    (Rezk 2009 Prop 3.4, 3.6, Cor 3.8).
+
+    Reduction identity. Choose a resolution P_* -> N_0 of a finitely
+    generated O-module N_0 by finite free modules. The tower
+    P_* x_O U / m^l U is degreewise surjective. Its first homology is
+    Tor_1^O(N_0, O/m^l) x_O U. This tower is pro-zero: for a first
+    presentation 0 -> K -> P_0 -> N_0 -> 0, the Tor group is
+    (K cap m^l P_0) / m^l K; Artin-Rees (Stacks Tag 00IN) gives d with
+    K cap m^l P_0 in m^{l-d} K for l >= d, so for any fixed j the
+    transition from l >= j+d to j is zero. Tensoring these zero
+    transitions with U preserves the conclusion.
+
+    The homology exact sequence for the countable inverse limit of
+    this degreewise-surjective tower gives
+
+        N_0 x_O \hat U = lim_l (N_0 x_O U / m^lU);
+
+    the possible lim^1 of the first homology is zero because that
+    tower is pro-zero. Finite freeness of the P_i identifies their
+    tensor with \hat U with their inverse-limit tensor. Take
+    N_0 = O/m^j; for l >= j the modules on the right are U / m^j U.
+    This proves the reduction identity.
+
+    PROPOSITION H.2.5 (tested Amitsur coefficient complex). For r >= 0
+    set Z_r = L_K(a)(E_a^{^ (r+1)}). For every finite ordinary
+    p-local W there are natural graded isomorphisms
+
+        pi_*(Z_r ^ W) = C_cts(G_a^r, pi_*(E_a ^ W)).
+
+    Consequently for finite V_0 and D V_0 = F(V_0, S), the application
+    of F(V_0, -) to the local unit Amitsur tower for E_a in C_a has
+    second page
+
+        E_2^{s,q} = H^s_cts(G_a, pi_q(E_a ^ D V_0)).
+
+    Proof. Hovey 2004 (Section 1, Theorem 4.11, Theorem 6.8) gives
+    the completed cooperation
+
+        pi_*L_K(a)(E_a ^ E_a) = C_cts(G_a, O_*),
+
+    with evaluation at g given by mu(1 ^ g) and the first factor
+    acting by constant coefficients. Put P_* = pi_*(E_a ^ E_a) as
+    an O_*-module through the first factor. Lemma H.2.5a makes P_*
+    flat. By Lemma H.2.5c the reductions of P_* modulo m^jP_* are
+    C_cts(G_a, O_*/m^jO_*). Lemma H.2.5b with M_* = O_*/m^jO_*
+    gives
+
+        P_* / m^jP_* = C_cts(G_a, O_* / m^jO_*),
+
+    induced by the actual ordinary evaluation. Localize now: H.2.5c
+    identifies the result with the completed module C_cts(G_a, O_*).
+
+    For higher r, use the ordinary relative smash
+
+        (E_0 ^ E_1) ^_{E_0} (E_0 ^ E_2) ^_{E_0} ...
+            ^_{E_0} (E_0 ^ E_r)
+          = E_0 ^ E_1 ^ ... ^ E_r,
+
+    with all E_i copies of E_a. Iterating Lemma H.2.5b, including its
+    extension to arbitrary ordinary W, identifies the homotopy with
+    P_*^{^{*_O_* r}}, which is flat. Continuous functions to finite
+    coefficients factor through finite quotients of G_a^r, so
+    C_cts(G_a^r, -) = C_cts(G_a, -)^{^{* r}}. Applying completion
+    at the end gives pi_*(Z_r) = C_cts(G_a^r, O_*). The evaluation
+    maps survive: each ordinary evaluation extends through localization
+    because its target E_a is local; its first-factor O_*-linearity
+    makes it adically continuous; base-change density makes the
+    extension agree with the displayed continuous function.
+
+    For finite W, base-change H.2.5b converts to
+    pi_*(Z_r ^ W) = C_cts(G_a^r, pi_*(E_a ^ W)). Finally
+    F(V_0, Z_r) = Z_r ^ D V_0 for finite V_0 by ordinary duality.
+    Relabel evaluations as cumulative coordinates
+    1, g_1, g_1 g_2, ..., g_1...g_r. The coface operators are
+
+        (\delta^0 c)(g_1, ..., g_{r+1}) = g_1 c(g_2, ..., g_{r+1}),
+        (\delta^i c)(g_1, ..., g_{r+1}) = c(g_1, ..., g_i g_{i+1}, ..., g_{r+1}),
+        (\delta^{r+1} c)(g_1, ..., g_{r+1}) = c(g_1, ..., g_r),
+
+    and the codegeneracies insert the identity at the corresponding
+    coordinate. These are the continuous inhomogeneous cochain
+    operators, so the Amitsur page is H^s_cts(G_a, pi_q(E_a ^ D V_0)).
+    Done.
+
 H.3 Upper bound: exp_H(T(B_p)) <= m (structural overview)
 
 The upper bound is much more involved. It has three sub-arguments:
@@ -1315,8 +1706,8 @@ The upper bound is much more involved. It has three sub-arguments:
 
 We develop each. The exponent calculus is the bridge.
 
-    LEMMA H.3.0 (exponent calculus). (standard descent exponent
-    calculus; part (v) is the finite-page bound.)
+    LEMMA H.3.0 (exponent calculus; the finite-page and tested-image
+    bounds are Lemma H.1.6).
     Let A be a unital algebra in a stable symmetric monoidal
     infinity-category with exact tensor, J = fib(1 -> A), epsilon: J
     -> 1.
@@ -1328,15 +1719,9 @@ We develop each. The exponent calculus is the bridge.
       (iv)  If e_m(V) = epsilon^{^ m} ^ id_V is null (m = exponent
             bound), then V is a retract of an object with m successive
             layers A ^ J^{^ i} ^ V, 0 <= i < m.
-      (v)   FINITE-PAGE BOUND. In the A-Amitsur descent spectral
-            sequence: if exp_A(V) <= b, then E_{b+1}^{s,q} = 0 for all
-            s >= b. Conversely, if some finite exponent exists and
-            E_2^{s,q} = 0 for all s >= b, then for every exact functor
-            Phi to spectra the tested image pi_d Phi(e_b(V)) = 0.
-            (The converse is the "tested image" form, not nullity of a
-            map; see H.3.a step 2.)
-
-Why. (i): the unit splittings of an A-module. (ii): naturality. (iii):
+      (v)   FINITE-PAGE / TESTED-IMAGE. See Lemma H.1.6(i) and (ii).
+
+Proof. (i): the unit splittings of an A-module. (ii): naturality. (iii):
 choosing a lift along the null composition of e_c on W. (iv): the
 successive cofibers J^{^ (i+1)} ^ V -> J^{^ i} ^ V have cofiber equal
 to A ^ J^{^ i} ^ V; the tower of phi = e_m(V) sits over V; nullity of
@@ -1350,10 +1735,10 @@ phi retracts V off the cofiber of phi = a filtration with m layers.
     Hence exp_{E_h}(B_h) <= m, and B_h is a retract of an object with
     m successive layers  E_h ^ I_h^{^ i},  0 <= i < m.
 
-Why (sketch). Three ingredients:
+Proof. The argument has four parts.
 
     (1) SOME finite exponent. Devinatz-Hopkins (2004, Appendix A,
-        Proposition A.3) shown that every K(h)-local spectrum belongs to
+        Proposition A.3) prove that every K(h)-local spectrum belongs to
         the thick tensor ideal generated by E_h. In particular B_h has
         some finite E_h-exponent c. This uses nilpotence and gets a
         FINITE c without needing any numerical value.
@@ -1361,11 +1746,11 @@ Why (sketch). Three ingredients:
     (2) FINITE TESTS vanish. The Amitsur cosimplicial construction on
         B_h in C_h, composed with the exact functor F(V_0, -) for a
         finite p-local V_0, has E_2 page (from the ordinary cooperation
-        calculation, Hovey 2004 Theorem 4.11) H^s_{cts}(G_h;
+        calculation, Proposition H.2.5) H^s_{cts}(G_h;
         pi_q(E_h ^ D V_0)). Proposition E.1 makes this
         zero for s >= h^2+2 = m. Using the already-known finite
         exponent c, the tested-image part of the finite-page bound
-        (Lemma H.3.0(v), converse form) gives:
+        (Lemma H.1.6(ii)) gives:
 
             [ V_0 , I_h^{^ m} ]  ->  [ V_0 , B_h ]   is zero
             for every finite p-local V_0.
@@ -1373,7 +1758,7 @@ Why (sketch). Three ingredients:
     (3) PHANTOM LEMMA -> actual nullity. The category of maps into B_h
         with "null after every finite precomposition" is characterized by
         Brown-Comenetz dualizing. (Christensen-Strickland 1998,
-        Proposition 4.11, is the criterion used.) Sketch:
+        Proposition 4.11, is the criterion used.) The argument:
 
             B_a   ~  F( D_a , D_a )   ~  F( D_a ^ M_a S , I_BC ),
             D_a = F( M_a S , I_BC )
@@ -1407,7 +1792,7 @@ Why (sketch). Three ingredients:
     an underlying degree-zero class whose H-trace is a unit. Consequently
     every D'-module in D_H has H-exponent at most 1.
 
-Why (sketch). The proof has three sub-steps.
+Proof. There are three sub-steps.
 
     Sub-step 1 (trace splitting). Let z in pi_0 D' satisfy:
         v = Sum_{g in H} g * z  is a UNIT in pi_0 D'.
@@ -1466,7 +1851,7 @@ Why (sketch). The proof has three sub-steps.
     PROPOSITION H.3.c (transfer). For p >= 5 with t = p-1, h = p:
         exp_H( T(B_h) )  <=  m = p^2 + 2,   where T(V) = L_{K(t)}(E_t ^ V).
 
-Why. Apply T to the m-layer decomposition of B_h from H.3.a. T preserves
+Proof. Apply T to the m-layer decomposition of B_h from H.3.a. T preserves
 cofibers and retracts. Each individual layer E_h ^ I_h^{^ i} becomes,
 after application of T, E_t ^ E_h ^ I_h^{^ i} = D ^ I_h^{^ i}. Then
 locally K(t)-ize: L_{K(t)}(D) ^ L_{K(t)}(I_h^{^ i}) [up to lax monoidal
@@ -1480,7 +1865,7 @@ H.4 Assembling the exponent contradiction
 
     COROLLARY H.4.1. i_{h,X} = i_{p,X} has no homotopy retraction.
 
-Why. Suppose i_{p,X} admitted a retraction r. The key square is:
+Proof. Suppose i_{p,X} admitted a retraction r. The key square is:
 
     S --eta_S^{(p)}--> B_p
     |                  |