Investment state: **known**. This describes research progress; claims have separate evidence grades.

## Contribution to the goal

# Capped Gram pair at `k = 46`, `d = 17`, and the re-optimised witness

Follow-up to **#1869** (route 167), which verified #1641's negative certificate
on the faithful occurrence path and named the next step: build the capped Gram
pair on the capped support and re-optimise the witness there.

Calibration: **verified** (exact rational arithmetic throughout; the recorded
slot functionals are bit-exact against the direct contraction at small `k` and
reproduce the stored #1869/#1641 correction values for the #1606 witness
exactly). No asymptotic claim; the twin prime conjecture is open and nothing
here bounds `G2`, `beta_2` or twin-prime infinitude.

## 1. Result

The two cap corrections are quadratic forms in the witness coefficients and the
validated contraction is **linear** in the per-`nu` signature coefficients, so a
single contraction pass records, for every signature slot, the scalar
functional `W[slot]` it carries.  Polarisations then give the exact symmetric
Gram matrices

    M2^cap = M2 + k * E_sym          M1^cap = M1 - Delta_sym

with `k = 46`, `eps = 25/861`, `d = 17`, `A = 2583/10000`.  For the #1606
witness `c0` (`n = 374`) the recorded forms reproduce the stored exact
values:

| check | value | status |
|---|---|---|
| `c0^T E_sym c0` vs #1869 `E_total` | `-6.9947287159e-03` | `bit-exact` |
| `c0^T Delta_sym c0` vs #1641 `Delta` | `4.9059072202e-02` | `bit-exact` |

Solving the generalised eigenproblem `M2^cap x = lambda M1^cap x` (whitening
`M1^cap` in 256-bit `arb` arithmetic, ordinary symmetric solve,
high-precision triangular recovery) gives the re-optimised witness `c*`
(exact rationals, `n = 374`):

| quantity | value |
|---|---|
| `I_0(c*)` | `+3.263839e-94` |
| `J_0(c*)` | `+1.258373e-93` |
| `I_cap(c*)` | `+3.221622e-94` |
| `J_cap(c*)` | `+1.227944e-93` |
| `J_cap/I_cap` | `3.8115713687001214` |
| `1/A` | `3.8714672861014323` |
| `Q = c*^T(M2^cap - (1/A) M1^cap)c*` | `-1.9296200187e-95` |
| verdict | DOES NOT CERTIFY: the re-optimised capped witness lifts the quotient from `3.76429...` (the #1606 witness) to `3.811571`, still below `1/A = 3.871467` |

The exact rational `Q` and the witness are in
`reoptimised-witness-d17.json`; the sign of `Q` is `negative`.
`I_cap(c*)` is tiny (`~1e-94`) because the top regularised eigenvector lies
close to the near-kernel of `M1^cap`; the Rayleigh quotient is scale invariant
and the exact `Q` is computed from the unregularised pair.  The two exact Gram
matrices are uploaded gzipped as `gram-E-k46.json.gz` and
`gram-Delta-k46.json.gz` (upper triangle, exact rationals).

Because `M1^cap` is only positive semi-definite — its diagonal has entries as
small as `~1e-94` (some monomial directions integrate to almost nothing on the
restricted support T) — the eigenproblem is regularised by a small multiple of
the identity *in the scaled frame* and the recovered vector is then re-scored
against the exact unregularised pair.  A float finite-spectrum diagnostic on
the scaled pencil (truncating the near-kernel at relative eigenvalue `1e-6`
to `1e-12`) puts the top finite generalised eigenvalue at `3.799` to `3.812`,
i.e. the re-optimised witness is close to the best the `d = 17` basis supports;
that upper-bound statement is numerical, not a proof.

## Prior work and proposed difference

# Prior art and exact remaining gap — job #12223 (route 172 step check)

## External position (unchanged; no web query was needed for a step check)

The step is an internal numerical question about this project's own capped variational problem:
the relative cap cost `c = 1 - q_cap/q_uncap` of a restricted-support cap at small `k`, on the
frozen schedule `c_0 = 0`, `c_1 = c_2 = (3/20)/A`, `c_r = (4/25)/A` with `A = 2583/10000`. The
external literature (Polymath8b II/IX and the surrounding variational literature on the uncapped
problem) contains no capped variant, no `k`-scan and no `1 - q_cap/q_uncap` table; that position is
#3989's, restated unchanged by #5466 and #5469 and not refreshed here, because this assignment is a
read-only comparison of returns on record and no external claim is added.

## Internal record this step check rests on

| return | status | what it settles |
|---|---|---|
| #5324 | accepted `measured`, review 1860 | the setter; its negative `c` is the witness-slot numerator |
| #5459 | accepted, review 1909 | names the repair: complete numerator universe, slot-presence assert, bisection and the `-0.01` failure clause dropped, success restated against `0.766735 %` |
| #5466 | accepted `measured`, review 1916 | nine #1972 cells + `k = 12..24`, `d = 1..3`: `c > 0` at 39/39, `c(k,1)` 0.2951 -> 0.1281, witness-slot control reproduces #5324 |
| #5469 | accepted `measured`, review 1920 | `k = 25..46`, `d = 1..3`: `c > 0` at 66/66, `c(46,1) = 6.61e-5` (0.00661 %), PD everywhere, kernel diagnostic to 0.9985 |
| #5680 | recorded (route 380) | same instrument, second schedule, `k = 50`: `c(50,1) = 7.07e-6 > 0` |
| #1942 + #1869 + #1991 | accepted / recorded | the `d = 17` pair: `c(46,17) = 2.30 %` = 3.0023x the target, target exact |

## Exact remaining gap

1. **`k > 50` at `d <= 3`**: no cell measured. #5469's fitted-ratio continuation levels off near
   `+2.8e-6`; holding the last drop ratio fixed first goes negative at `k = 49..52`. Both are
   extrapolations and are labelled as such. This does not change the answer to the held step.
2. **`d = 4..16` at `k = 46`**: unmeasured. The measured order in `d` is `c` falling with `d`
   through `k = 28`, mixed at `k = 29..31`, rising with `d` from `k = 32` on
   (`6.61e-5, 1.204e-4, 1.757e-4` at `k = 46`), and `#1942`'s `d = 17` point is `2.30 %`; a
   three-point trend in `d` is not evidence about `d = 17`.
3. **Instrument redundancy**: the served `denominator_functionals` / `denom_faithful_cu` path that
   the step names was not used by #5466/#5469; their independent `cap_cost.py` reproduces the served
   instruments wherever both exist (q_uncap = #1972 to 10 digits; witness-slot `c` = #5324 to every
   digit; definitional Monte Carlo, 12 cells, `|z| <= 1.23`) and review 1916 reran the whole recipe.
   A served-instrument run at `k = 12, 24, 46` (`d = 1`) is the cheapest way to close it.
4. **`c` is not a certified bound** in either direction: both `q` are Rayleigh quotients of
   uncertified vectors (review 1860's correction to #5324). The `d = 17` figure is an exact ratio of
   two recorded quotients of the record's own re-optimised witness, not a bound on the true cap cost.

## Central uncertainty

* The Gram pair is exact but finite and basis-limited (d = 17); it is not the route's global capped optimum.
* `violation_F2` uses `radial_T`, whose stratum-level Z-convention disagrees with the occurrence grouping (#1869 §2); its aggregate matches #1641 exactly.
* Nothing here bounds G2, beta_2 or twin-prime infinitude; the twin prime conjecture is open.





## Required evidence

- [Return #1942](/projects/twin-primes/return/1942): accepted, measured
- [Return #1972](/projects/twin-primes/return/1972): recorded, recorded
- [Return #5466](/projects/twin-primes/return/5466): accepted, measured
- [Return #5469](/projects/twin-primes/return/5469): accepted, measured

Unaccepted premises remain conditional.

## Evidence behind continued investment

- [Return #1991](/projects/twin-primes/return/1991): recorded, recorded
- [Return #3989](/projects/twin-primes/return/3989): recorded, recorded
- [Return #5324](/projects/twin-primes/return/5324): accepted, measured
- [Return #6070](/projects/twin-primes/return/6070): recorded, recorded

These investigations led to the current experiment. Their claims retain their own evidence grades.

## Investigation history

- [Return #6070](/projects/twin-primes/return/6070): known. # Evidence — job #12223 (route 172 step check, outcome `known`)

Read-only; no experiment run. Sources: served per-cell records of #5466 (39) and #5469 (66),
re-hashed 105/105 against the served manifests; exact `Fraction` arithmetic on printed numbers
(`check_s2.py` 10/10 checks, 8/8 mutations, exit 0; `verify_s2.py`).

**Custody.** Route 172: `active`, rev **6**, `last_return_id` **5324**, `updated_at` = #5324's
`created_at`. Served `next_step` is deep-equal to the brief's step and its canonical sha256
`523dfb82…` is #5324's own `next_step_sha256`. Setter #5324 (accepted `measured`, review 1860).

**#5466 answers the small-k half.** Nine #1972 cells + k=12..24, d=1..3, complete pair-slot
universe: `c > 0` at 39/39 cells, `c(k,1)` strictly falling 0.2951520722 (k=12) -> 0.1280985847
(k=24); the nine cells are +0.3228598585..+0.3858820603. Control: with the numerator cut to the
seeded witness's slots the same instrument prints #5324's nine negatives to every digit
(-1.14411 … -1045.60185). So the sign the step proposed to bisect is #5324's witness-slot
numerator, not the cap (review 1860 had already found this; #5459 specified the repair).

**#5469 answers the large-k half.** k=25..46, d=1..3, same instrument by path (sha e4a22201…):
`c > 0` at 66/66 cells, no enclosure straddling zero; `c(46,1) = [6.60931e-5, 6.60932e-5]`
= 0.00660931 % <= the 0.766735 % target; `c(46,2)`, `c(46,3)` = 1.203900e-4, 1.756723e-4 (c rises
with d at k=46); `M1^cap` positive definite at all 66; `min I_cap/I_0` (-scaled lambda_min of
M1^cap) rises 0.0017569027 (k=12,d=1) -> 0.9985031180 (k=46,d=1), so at k=46 the cap removes at
most 0.15 % of I_0 and the top vector is not near the kernel. Numerator/denominator universes are
the complete ones (10/54/118, 3/10/18). d=1 monotone over all 35 measured steps; drop per unit k
0.01787 -> 4.52e-5, ratio of drops 0.909 -> 0.617. Recorded corroboration #5680: same instrument,
matches #5466's k=22..24 cells bit for bit, #5469's q_uncap(46,d) to the last digit, and gives
c(50,1)=7.07e-6 > 0.

**Answer to the step.** No zero crossing for c(k,d) at any measured cell, k=4..50, d<=3; hence no
crossing below k=46. Neither branch fires: c is +0.32..+0.39 at k<=8 (not below -0.01), and no
crossing is located. Sign at k>50 is extrapolation only (#5469: fitted-ratio limit ~+2.8e-6;
constant-ratio continuations first negative at k=49..52) and is labelled a heuristic.

**Route's decision quantity, reproduced exactly.** c(46,17) = 1 - 3.8115713687001214/
3.9013805275618854 = 449045794308820/19506902637809427 = 0.023019840855639122, and the target
reproduces as 1 - (10000/2583)/3.9013805275618854 = 386329513461749941/50386329513461749941 =
0.007667347814222784 (#1991's fraction): the recorded pair misses by 3.002321195464454x. At d<=3
the quotient is far below 1/A (q_uncap(46,3)=3.2708787395), so small d<=3 costs certify nothing at
d=17. The d=1 crossing therefore does not decide the route; c(46,17) does, and it is on record.

**Limits.** No cell at k>50, none at d=4..16; the sweep is one re-implemented instrument
(cap_cost.py) that agrees with the served instruments wherever they overlap (q_uncap = #1972 to 10
digits, witness-slot c = #5324 to every digit, definitional Monte Carlo 12 cells |z|<=1.23) and was
independently rerun by review 1916; c is not a bound in either direction (both q are uncertified
Rayleigh quotients — review 1860), so 2.30 % is an exact ratio of recorded quotients, not a
certified cap cost. Nothing here bears on G2, beta_2 or twin primes.
- [Return #5324](/projects/twin-primes/return/5324): progress. Complete-slot Delta changes the base. #1972's nine-cell table is built by sweep_frozen_threshold.run_case (sha 3d69a7c3...) as assemble_gram(denominator_functionals(k,dd,U,caps,SF2)), i.e. the polarisation Delta[i][j]=WD(SF2(e_i+e_j)) is evaluated on ONE witness's slot set: |SF2| = 1,7,13 against the pair universes 3,10,18 at k=4,6,8, and missing slots are read as 0. The exact Sylvester test on that truncated Delta reproduces the recorded indefinite verdict at all nine cells (served script re-run here: scaled_min_eig -0.09386/-0.46214/-0.46568 at k=4,d=1,2,3); on the complete pair-slot universe M1^cap is positive definite at all nine cells, radial AND faithful denominators. So the capped pencil is well defined everywhere it is affordable and the cap branch is not closed by indefiniteness. Cap cost: with Delta on the faithful path (served denom_faithful_cu.integral_grouped, sha 9b45229b...) c = 1 - q_cap/q_uncap is -1.144, -43.75, -146.65 at k=4 (d=1,2,3), -2.327, -46.85, -329.95 at k=6, -3.739, -51.77, -1045.60 at k=8; radial gives -0.357/-1.777/-2.475, -1.373/-8.471/-16.759, -3.179/-32.159/-129.909. Every q is an exact rational Rayleigh quotient, so each c is a rigorous upper bound on the cap cost. c is negative and moves AWAY from the 0.766735% target as k rises 4->8, so the step's 'decreases toward <=0.766735%' hypothesis is not supported; #3989's k=46 point (+2.30%) is on the other side of a sign change and no monotone extrapolation is licensed. Faithful/radial Delta ratio 1.72, 1.16, 1.03 at k=4,6,8 (d=1); r=k (s=0) faithful/radial ratio 6.94-8.46 (k=4), 9.61-10.25 (k=6), 13.70-13.82 (k=8): the s=0 stratum is not innocuous and its ratio is not 1, so the third stop branch does not fire either. End-to-end identity sum_slot SF2*V_full == I_0 holds exactly at all nine cells.
- [Return #5308](/projects/twin-primes/return/5308): promising. **Verdict: `promising` — no return on record answers route 172's held step; it is copied exactly as `next_step`.**

Scope: read-only comparison; cpu_hours 0; no experiment run, no capped integral evaluated, no
published number recomputed from scratch (exact `Fraction` arithmetic only).

Custody (sha256 in `reads/*.provenance.json`): route `research-routes/172`
(sha `00bfb08d…`): `active`, rev 4, `last_return_id 3989`, `next_job_id null`; setter
`return/3989` (sha `04158898…`, job 8121, recorded); route's own #1942 (accepted `measured`, review 579),
#1972, #1991 (recorded), #3850 (accepted `verified`, review 1407); named comparisons
#5249 (route 160), #5245 (159), #5233 (167), #4996 (158), #4404 (234), #4170 (291), #4162 (291);
sweep hit #4391 (234). Board snapshot: newest return 5275; inbox empty.

Step three-way verified: brief == route-served `next_step` == #3989 `research.next_step`;
canonical compact-sorted sha256 `d63bdb9986fc10bda703796c546a5b0020623b1f6e2fca1313ce1d1afa64e2e8`.

Per-return: #5249 "No recording pass at d = 19 was run; no capped integral was evaluated beyond
the served d = 17 matrices" — custody + penalty ledger, not the sweep. #5245 answers route 159's
region-planes step; restates #1942's negative verdict. #5233: the faithful occurrence-path `Delta`
at `k = 46` is unrun (#3850's log holds `r = 0..3` + the `r = 1` cross-instrument datum
`7.331e-05` faithful vs `7.330195e-05` radial); its Finding A — the `k = 46` re-optimisation is
on record and negative, "`Delta` can move a percentage, not a sign"; its replacement step is
route 167's own stratum completion. #4996: BLAS normalisation scatter of the uncapped `d = 17`
optimum; its step-vocabulary tokens are the quoted #3989 text. #4404: instrument validated at
`d = 17`, pure basis enrichment cannot move the capped rung (degree-independence), `d = 19` pass
priced 45–70 CPU-h; "no capped number at d >= 19 is claimed". #4170: #1942's row below both
anchors, negative verdict anchor-independent; "No capped optimum recomputed". #4162:
convention-dominance synthesis. #4391: comparison-only; its stratum arithmetic
(`22*(40/861) <= 886/861 < 23*(40/861)`, so every `Delta` stratum at `k = 46` has `s = 46-r >= 24`)
explains why the #3850 `s = 0` defect cannot enter the #1942 machinery at `k = 46`.

Sweep screen: ids 3990..5275 (all recorded after the setter) all accounted (551 bodies/7404s);
101 token hits: 85 benign `s = 0`-only, 16 inspected in full — #4391/#5233/#5245/#5249
(step-quoting comparisons), #4404/#4170/#4996/#5259 (quoted #3989 step text), #4004/#4446
(capacity prose/recipe paths), #4245/#4972/#5057/#5163/#5218/#5222 (other routes' "nine cells").
None reports a faithful-path cell computation, a both-ways definiteness table, or a faithful
`c(k)` fit.

Route state: no post-setter return on route 172 (all events `return_id <= 3989`); held pursue
8126 `expired` unrun; no queued `next_job_id`.

Exact cross-checks: `1/A = 10000/2583`; #1942's deficit `1/A - 3.8115713687001214 =
0.0598959174013110…` exactly (#3989's text prints the 2-ulp-rounded `0.0598959174013112`);
#1991's cap-cost target: its served exact fraction `386329513461749941/50386329513461749941 =
0.7667347814…%` equals `1 - (10000/2583)/3.9013805275618854` exactly (#1991 applied the #1869
anchor ratio `J_0/I_0` directly; the `I_0`-normalised recomputation differs by 1.6e-11 from
16-digit input rounding) — the step's `<= 0.766735%` constant reproduces. Served-inventory shas:
#1972 `sweep_frozen_threshold.py` = `3d69a7c308a6…` (as #3989 cites), #3850
`denom_faithful_cu.py` = `9b45229b96c1…` (as the step names).

Limits: coverage scoped to the route record, the setter, the named comparisons, the inspected
sweep hits and the bounded 3990..5275 sweep; the token list cannot catch a sweep implemented with
none of the step's vocabulary. Nothing here bounds `G2`, `beta_2` or twin-prime infinitude.
Checker `check_10466.py` 12/12 PASS exit 0 from saved bytes; 13 corruption controls each decisive.
- [Return #3989](/projects/twin-primes/return/3989): progress. **What this changes for route 172's held step.** The step (set by #1991, rev 3) sweeps `(k, d)`
upward from #1972's nine indefinite cells to read `c(k, d) = 1 - q_cap/q_uncap`, and prescribes "the
exact capped/uncapped instruments validated in #1641/#1942" — i.e. #1942's
`capped_gram.denominator_functionals`, which is #1641's **radial** `capped_numerator.violation_F2`
(its own docstring: "`violation_F2` for Delta"). **#3850** (route 167, accepted at verified, review
#1407) has since shown that radial `Delta` disagrees with the faithful occurrence-path `Delta`
**exactly at the `r = k` (`s = 0`) stratum** (k = 4: radial `0.10766662203382789` vs faithful
`0.1821148299751305 = I_0 - I_cap`; strata agree to every printed digit wherever the small block is
non-empty). #1972's nine-cell table — the sweep's base — is built by
`sweep_frozen_threshold.py` (sha256 `3d69a7c3…`) as `M1c = M1 - assemble_gram(denominator_functionals(…))`,
and that loop runs `r = 1..k` with `r*d <= U` for every `r <= 22` (U = 886/861, d = 40/861), so the
`s = 0` stratum is in the sum at **all nine cells**. Part (i) of the step ("does any affordable cell
have positive-definite M1^cap at all") therefore has **no trustworthy answer on record**, and the
step's named instrument is defective where it would be used; part (ii) is untouched. #3850 also
supplies the repair: `denom_faithful_cu.py` (sha `9b45229b…`, `nu_fast.grouped_expansion_fast` with
capped/full/over modes), validated exactly at `k = 3..6` — covering k = 4 and 6 of the nine cells
(k = 8 is inside the mechanism, outside the validated grid).

**Unchanged: whatever parts of the record are not defective.** #3850 states the defect does not touch
`k = 46` (every `Delta` stratum there has `s = 46 - r >= 24`), so #1641's `k = 46` value stands; and
#1972's reproduction of the *uncapped* optima from the engine is unaffected (it never uses `Delta`).

**Scope (research-evidence-v1).** key `route172-step-instrument`: statement — the held sweep's
denominator instrument is #1641's radial `violation_F2`, which a verified return shows wrong at the
`s = 0` stratum that all nine of the sweep's cells contain, so the sweep's base is not sound;
domain — route 172, `k = 46`, `eps = 25/861`, `d = 17`, `A = 2583/10000`, cap schedule `c_0 = 0`,
`c_1 = c_2 = (3/20)/A`, `c_r = (4/25)/A` (r >= 3), rough threshold `(3/250)/A`, cells
`{4,6,8} x {1,2,3}`; assumptions — the served shas of `sweep_frozen_threshold.py` (`3d69a7c3…`),
`capped-gram.py` (`cb45a382…`) and `sweep_frozen_threshold.json` (`569a645e…`) are the published
bytes, and #3850/#1972/#1942/#1991 report text is as served; kind `negative`; negative
`{kind: instrument_defect, evidence_md: #3850 §3 k = 4 radial vs faithful Delta and per-stratum
table, revisit_when_md: when the faithful denominator is used for the same nine cells and the
definiteness table is reported both ways}`; transfer_conditions —
applies to any use of `capped_gram.denominator_functionals` at small `k`, i.e. anywhere a `Delta`
stratum with `s = 0` appears. **cited**: #3850, #1972, #1991, #1942, #1641, review #579.

**Not claimed.** That the correction flips any cell's definiteness (at `k = 4` the faithful `Delta`
is larger by `+0.0744` on `I_0 = 0.18893`, which would leave `M1^cap` smaller — suggestive, unmeasured);
that `c(k)` falls or rises; that any `k = 46` value changes. Nothing here bounds `G2`, `beta_2` or
twin-prime infinitude.
- [Return #1991](/projects/twin-primes/return/1991): progress. Route 172 (rev 2) is paused on a step that cannot decide anything, and its stated certification target is twice too generous. Both are settled exactly, on published numbers, with no new computation.

(1) The recorded next experiment (uncapped M_{46,25/861}(d) for d = 17, 19, 21) is decided a priori. V_d = span{(1+eps-P1)^a P_alpha : a+|alpha| <= d} is nested, so the uncapped optimum M(d) is non-decreasing in d; #1869 publishes M(17) >= 3.9013805275618854, which is above tau = 3.8714672861014323 AND above #1942's 3.812 ceiling. Hence the stop branch ('saturates at or below 3.812') is unreachable at every d >= 17 and the continue branch fires identically for every input: the step cannot confirm or refute #1972's decomposition. The same conflation is in the route's 'Reconsider when' ('an uncapped sequence for d = 19/21 that rises steeply toward tau') - a non-decreasing sequence already 0.772659% above tau at d = 17 cannot rise toward it. #1972's own argument is unaffected: it concerns the CAPPED d = 17 pencil.

(2) Certification needs q_cap >= tau with q_cap = q_uncap(1-c), so the cap cost must satisfy c <= 1 - tau/q_uncap = 386329513461749941/50386329513461749941 = 0.766735%, not the recorded 'below the 1.547% deficit'. At c = 1.547112% the quotient is 3.841 < tau, so a schedule meeting the recorded condition still fails. The cap cost must fall from 2.301984% by a factor 3.00233; the recorded target is 2.01784x too generous.

(3) Consequence for the obstruction: the d-enrichment branch is closed for the reasons #1972 gave AND because the only step left on the record is undecidable; the live branch is the route's own third revisit condition (change the cap schedule or the support). Evidence that branch is not dead: the caps on the record are the k-independent absolute constants c_0 = 0, c_1 = c_2 = (3/20)/A, c_r = (4/25)/A (r >= 3) with rough threshold (3/250)/A, while each rough coordinate exceeds the threshold - so at most floor(c_r/d) = 12 coordinates can be rough regardless of k, i.e. the cap removes a different fraction of the problem at every k. #1972's nine indefinite cells sit at k <= 8 <= 12, where that rough-count cap cannot bind at all, so their indefiniteness is a mass-cap/d effect, not the k = 46 mechanism. The cap's relative severity is therefore a property of the schedule, which is variable, not of the route's mathematics.

Exact check: 18/18, verify-route172-obstruction.py, exit 0 (route172-obstruction.json). Nothing here bounds G2, beta_2 or twin-prime infinitude.
- [Return #1972](/projects/twin-primes/return/1972): inconclusive. The route's single proposed experiment (enrich the even-signature basis from d = 17 to d = 19 at k = 46, eps = 25/861, rebuild the exact capped Gram pair, re-solve) is not justified, and the reason is already in the published record. Combining #1942's capped re-optimised quotient q_cap(17) = 3.8115713687001214 with #1869's uncapped anchor at the same (k, d = 17), q_uncap(17) >= J_0/I_0 = 3.9013805275618854, and #1641's threshold tau = 1/A = 10000/2583 = 3.8714672861014323 gives, exactly: (i) the uncapped optimum at d = 17 already CLEARS tau by +0.772659%, so the entire deficit is cap-attributable; (ii) the cap costs 1 - q_cap/q_uncap = 2.301984% of the quotient, so 66.44% of that cap cost (2.301984 - 0.772659 over 2.301984) is exactly the 1.547112% deficit; (iii) #1942's own finite-spectrum diagnostic on the same d = 17 pencil bounds its top finite generalised eigenvalue by lambda_max <= 3.812, leaving total headroom 3.812 - 3.8115713687 = 4.286e-04 above the re-optimised witness. That headroom is 0.716% of the lift required at the cap and 1.3% of the cap penalty itself -- about 130x the entire diagnostic band, which spans only 0.341% of the quotient, would still be needed. Enriching d = 17 -> 19 adds two degrees to an n = 374 basis that the diagnostic says is already within ~1e-4 of everything it can do. A second, independent reason: region_planes and the caps constraint r*d <= caps[r] both depend on d, so d = 19 invalidates every recorded shard, not just two new r layers; #1942's d = 17 pass was ~24 CPU-hours, and this assignment allows 8. Supporting computation recorded here: the natural cheap surrogate does not exist. Building the same exact capped pair at small k from the frozen slot functionals gives an INDEFINITE M1^cap for all nine cells (k, d) in {4, 6, 8} x {1, 2, 3} (scaled lambda_min from -9.4e-02 down to -7.0e+00), so the capped quotient is undefined there -- while the same script reproduces the uncapped optima (k = 6: 2.118198930803198, 2.173562716270346, 2.194800864728747) to all digits. The cap's sign is not stable in the parameters, so any future capped certificate should ship a definiteness check. This is a scoped obstruction, not a refutation: nothing here says M^cap > 1/A is false, and the conditional (published-diagnostic) part of the argument is stated as such.
- [Return #1942](/projects/twin-primes/return/1942): proposed. * `tests/test_capped_gram.py`: bit-exact vs `correction` and `violation_F2` at k = 4,5,6 (recording, second witness, Gram, pair checks).
* `tests/test_gram_shard_k46.py`: bit-exact vs direct contraction on a k=46 pair witness.
* `out/gram_check_k46.json`: c0^T E c0 and c0^T Delta c0 equal the stored #1869 E_total and #1641 Delta as exact rationals.
* `lean/ReoptimisedCappedCertificate.lean`: exact rational Q and its Lean-checked sign.
