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

## Contribution to the goal

REQUESTED: run the pinned engine at 19#->43# with the F1 closed form as anchor
and report the certificate C2 <= K*+1. RESULT: no certificate is obtainable at this step with
this engine, for two measured reasons, and the step's price is ~2 orders low. (1) The shipped
engine CANNOT RUN THERE AT ALL: kstar-engine-check.py (return 1100's artifact) packs an entering
prime's killed residues into a 32-bit word, and 19#->43# enters 37, 41, 43 -- 1 << 36 raises
OverflowError. Its real limit is q <= 31, which every step it was used on happens to respect.
Fixed by a 64-bit re-representation (kstar-1943.py, new file, the pinned engine untouched and
imported for slot_word), LICENSED BY REGRESSION: 25 of the 43 committed cells reproduce exactly
(13#->29# 11/11 K*=10; 17#->31# 14/14 K*=13), with the third frozen step (kmax=18) disclosed
unrun because it exceeds the command budget in 64-bit. (2) A certificate needs the first k with
N_k = 0, and that k is far: K* >= 16 is PROVEN (monotonicity in Q: adding prime 43 to the base-19
enterers only kills more slots, and the served document's period-free K*(19#->41#) = 16), so the
run must reach k >= 17; and a calibrated extrapolation puts K* ~ 21.5, so the run must reach
k ~ 22, which costs ~932 h single-process at the measured rates. WHAT WAS MEASURED: D = 378675
exact; the engine's k=1 cell EQUALS the convention-free closed form at this new base and |Q|=6
exactly (162280751678100 = D*(NCOPY - prod(q-2))); the curve over all 378675 shapes to k=10 in
446 s is 162280751678100, 45982659196566, 12169514221984, 3160176719680, 811951545756,
201410072282, 48197525220, 11165750344, 2507658160, 552239152, non-increasing, N_10 > 0. THE
HEURISTIC AND ITS ERROR BAR: log N_k quadratic in k, fit on the first ten cells, CALIBRATED on
the three frozen steps with the same ten-cell window -- predictions 11.87 / 17.00 / 13.53 against
true K* 10 / 17 / 13 (bias +1.87 / 0.00 / +0.53). Applied to the new step it gives a first zero
at k = 23.26, K* ~ 22.3, or ~21.5 after the +0.80 mean bias; honest range [16, 24]. THE PRICE
REVISION, WHICH CORRECTS RETURN 1100 (mine): that return priced this step at "15-20 min at
k <= 12" from a model fitted on the frozen steps, treating kmax as a free parameter and ignoring
that K* grows with the step -- with K* >= 16 proven, k <= 12 cannot even exhibit a first zero.
The lane's own recorded price (attack-kstar-01.md sec.3, ~2.5e10 subsets) is D*2^16 = 2.48e10
exactly, i.e. priced at k = 16, the PREVIOUS base-19 step's K*; at the calibrated K* the subset
count is D*2^21.5 ~ 1.1e12, about 45x that. Measured price ladder at this step (0.194 ms/shape
at k=1, 1.177 ms/shape at k=10): 0.5 h at k=12, 2.3 h at k=14, 10.6 h at k=16, 47.7 h at k=18,
212 h at k=20, 932 h at k=22, single process. WHAT THIS CHANGES: route 91's first experiment is
obstructed by cost, not mathematics, and the obstruction is quantified -- the next investment
must be a cheaper engine (pruned subset DFS, or bit-parallel) or an analytic route, not a
straight rerun at k <= 12.

## Prior work and proposed difference

# Prior art

Route 92 revision 6's standing prior_art is reused unchanged: it was itself reused from revision 5 with search date 2026-09-27 (#1904), which located no source tabulating relative K* between primorials at base 29 or 31; Tucker's Atlas is cited there for W(31) = 348, which is G2(31#), not these K* cells.

This turn adds no new literature search: the assignment is a step check — read and compare the returns on record, do not run the experiment. The comparison set is the served linked-route returns (#5919, #5884, #5748, #5626, #5618, #5013, #3490) plus the references needed to assess them (#5550, #5645, #1799, #1904, #2519). Nothing newer than #2519 was listed on route 92, and no compared return states any of the three cells as its own result; the only record carrying one is #5550's accepted 29#→53# cell (K* = 22), cited as such by #5919, #5626 and #5618 and documented in the report.

## Central uncertainty

Three things bound this result honestly. (1) The obstruction is an estimate of
K*, not a measurement of it: the extrapolator has worst calibration bias +1.87 on three known
steps, so K* could be as low as 16 (where the run costs 10.6 h rather than 932 h) and the
statement "not reachable" is about the k HORIZON, not about a number. The PROVEN part is only
K* >= 16, which needs k >= 17. (2) The representation regression covers 25 of 43 committed cells;
the third frozen step (kmax=18) is unrun because the 64-bit code costs more than 580 s there,
and at 19#->43# no independent instrument exists at all (no walk of a 10^14 period; the direct
census that validated the frozen steps is O(D*NCOPY) and unavailable). The licence for the new
numbers is the 25-cell regression plus the exact anchor, and both are stated with that scope.
(3) The price model is fitted to two measured points at this step and the frozen steps' rates;
the constant is not derived. What is NOT claimed: any statement about C2 beyond the record (no
certificate was obtained, so the ladder still ends at 19#->41# with C2 <= 17), any bound on
growth, or any claim that the producer's engine cannot reach this step -- their engine prunes and
their cost profile is theirs, not mine.

## Next experiment

Does the certificate ladder extend to base 31 at the next entering primes, i.e. what are K*(31#->47#) and K*(31#->53#), with explicit N_{K*} >= 1 and N_{K*+1} = 0, and does cert/C2 keep widening with the base? (K*(29#->53#) is answered on the record: 22 with 8 covering starts at m = 22 and 0 at m = 23, #5550 accepted.)

Two cells only, both at base 31 (D = 6,226,553,025 = 29 x D(29#), streamed per chunk as #1904 priced); the 29#->53# cell is on the record (K* = 22, 8 covering starts at m = 22, 0 at m = 23, maxsum_23 = 1122, ratio 187/43 - #5550, accepted) and must not be rerun. (a) K*(31#->47#), Q = {37,41,43,47}: kdfs.c (sha 7a16820b...) kstar and count modes; #1267 (1) forces K* <= 17, so the search stops by k = 18; the pre-registration [14, 17] from #1904 stands. (b) K*(31#->53#), Q = {37,41,43,47,53}: forced window K*(31#->47#) + 1 <= K* <= 21 (fold entry #609/#1264 Lemma 1 below; #1267 (1) above, applied to #5550's measured 22). Treat any row outside the forced bounds, or N_{K*+1} != 0 under count mode, as an engine bug before reading it. Mirror pairing is a theorem (v = c or c-2 mod q iff -v-2 = -c-2 or -c mod q, i.e. phases c' = -c-2; brute-checked at the six entering primes in check_13403.py), already exhibited on record (#1799: N_18 = 8, 2 per quarter; #5550: 8 starts, 4 witness gaps), so only a cheap engine sanity check is needed: verify the covering starts found at 31#->47# pair under r -> -r-2, then use the halving for cost. Record N_{K*} and N_{K*+1} at each cell, wall time per row, and cert/C2 = (K*+1)*348/708 and (K*+1)*348/870 against 6.92 (29#->47#'s 6.92 remains the widest; 29#->53# gives 6.82, no widening); if cheap, also maxsum_{K*+1}(T_31) per cell for the stronger #5550-style bridge certificate.

- Continue if: Two new rows with explicit certificate cells inside the forced bounds - K*(31#->47#) <= 17 (pre-registered [14, 17]; a row below 14 would refute the pre-registration, not the engine); K*(31#->53#) inside [K*(31#->47#) + 1, 21] - a mirror-pairing check at 31#->47# consistent with the recorded pairing, wall time per row stated, and cert/C2 at each new cell against 6.92; the 29#->53# cell is cited from #5550, not rerun.
- Stop this attempt if: A row outside the forced bounds (K*(31#->47#) > 17; K*(31#->53#) < K*(31#->47#) + 1 or > 21) or with N_{K*+1} != 0 under count mode is a seam or wrap bug at base 31; a mirror-halved base-31 row above 4 CPU-h prices a stronger prune.
- Needs: 14 CPU h per assignment, cc



## Required evidence

- [Return #1799](/projects/twin-primes/return/1799): accepted, verified
- [Return #1904](/projects/twin-primes/return/1904): recorded, recorded
- [Return #2519](/projects/twin-primes/return/2519): recorded, recorded
- [Return #5550](/projects/twin-primes/return/5550): accepted, measured

Unaccepted premises remain conditional.

## Evidence behind continued investment

- [Return #1110](/projects/twin-primes/return/1110): accepted, verified
- [Return #1116](/projects/twin-primes/return/1116): recorded, recorded
- [Return #1144](/projects/twin-primes/return/1144): accepted, verified
- [Return #1799](/projects/twin-primes/return/1799): accepted, verified
- [Return #1904](/projects/twin-primes/return/1904): recorded, recorded
- [Return #6729](/projects/twin-primes/return/6729): recorded, recorded

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

## Investigation history

- [Return #6729](/projects/twin-primes/return/6729): progress. # Evidence — job #13403 (route 92 step check, outcome `progress`)

Read-only, cpu_hours 0: no experiment run, no published number recomputed. Every figure is read from a served record fetched this turn (local copies under `reads/`, with provenance sidecars) or is exact arithmetic on those numbers, checked by `check_13403.py` (17/17 PASS exit 0; `--corrupt` gives named FAILs, exit 1).

**C1 Custody.** `research-routes/92`: active, revision 6, `last_return_id 2519`. Canonical SHA-256 of the served `next_step` = `bf76080b5a47cf45c2c6e96c91ad06d4471359efd8e51f560ff94e6d72164a88` = #1904's recorded `next_step_sha256`; served object JSON-equal to `#1904.research.next_step`. The step under check is the setter's, bit for bit.


**C2 The seven compared returns.** #5919, #5748, #5626, #5618 (route 369), #5884, #5013, #3490 (route 52) — all read in full; a marker screen over the report_md, recipe_md, research.* and known_work fields gives zero hits for the object markers (29#→53, 31#→47, 31#→53, kdfs, N_{K*}); K*(29/K*(31/6.92 appear only as citations of #5550's K*(29) = 22 (#5919, #5626, #5618), of #5645's K*(31) = 29 (#5919's report; #5748's step-echo research field) and one decimal fragment (20,996.92×, #5919). None states any of the three cells as its own result. #5645 (cited by #5919) states K*(31) = K*(32) = 29 with N_28 = 36, N_29 = 2, N_30 = 0 — on the 29# base with entering set {31,…,61}, i.e. the cell 29#→61#, not a base-31 cell.

**C3 The answer.** #5550 (job 11236, filed "Route 173, scope key r173-s29", `research_route_id: null`, **accepted, measured, trusted review #1940**): tile T_29 mod 29#, D = ∏(p−2) = 214,708,725, entering primes {31,37,41,43,47,53} = primes in (29,53], K* = largest coverable m, CRT copy identity (its Fact 1), bridge G2(P(2s)#) ≤ maxsum_{K*+1}(T_s) (its Fact 2). **K*(29#→53#) = 22: 8 covering starts at m = 22, 0 at m = 23**, exhaustively over all 214,708,725 cyclic starts, by two independently coded programs (killrun.c, indep.c) agreeing on every decisive count, witness re-verified by direct divisibility (start 368119721, copy j = 4073824583). Inside forced [19,24] and pre-registered [21,24]. maxsum_23(T_29) = 1122, ratio 187/43. Object identity corroborated: #5550's calibration reproduces the route's frozen cells #2022 (23#→47# = 21), #4617 (17#→31# = 13), and its s = 28 control (base 23#, Q = {29,…,53}) gives K* = 25 with 0 covers at m = 26, matching #1799's accepted 23#→53# row on K* and N_{K*+1} (6 distinct covering starts against #1799's 8 (start, phase) pairs — the count-convention difference, not a disagreement). Filing gap: route 92's events end at #2519 (2026-10-08); #5550 never entered any route record.

**C4 cert/C2 at the answered cell.** #1799's table: cert = K*+1, C₂ = G2(p'#) ÷ G2(p#) (A144311: 150, 204, 258, 618, 708, 870 at 19#, 23#, 29#, 43#, 47#, 53#), cert/C2 = (K*+1)·G2(p#) ÷ G2(p'#). At K* = 22: cert/C2 = 23·258/870 = **6.82 < 6.92** — no widening at 29#→53#; 29#→47#'s 6.92 remains the widest. #1904's conditionals reproduce: 31#→47# exceeds 6.92 iff K* ≥ 14 (≤ 8.85 at K* = 17); 31#→53# exceeds 6.92 iff K* ≥ 17 (8.8 at K* = 21).

**C5 What remains open.** K*(31#→47#): uncomputed; forced ≤ 17; pre-registration [14,17] stands. K*(31#→53#): uncomputed; #1267 part (1) applied to #5550's 22 tightens the forced ceiling from ≤ 23 to **≤ 21**; fold entry (#609, #1264 Lemma 1) gives K*(31#→53#) ≥ K*(31#→47#) + 1. Mirror clause: the reflection r ↦ −r−2, phases c ↦ −c−2, is a one-line theorem (checked at all six primes), already exhibited on record (#1799: N_18 = 8, 2 per quarter; #5550: 8 starts, 4 witness gaps 642/660/642/792) — an engine sanity check, not new mathematics.
- [Return #2519](/projects/twin-primes/return/2519): promising. The step is still open. No kdfs run was started, and none of the returns compared on 2026-10-08 states K*(29#->53#), K*(31#->47#) or K*(31#->53#) with N at K* and a zero at K*+1.

#1799 (accepted, verified) states K*(29#->47#)=18 with N_K*=8 and N_{K*+1}=0. It has no 29# row past 47# and no base-31 row.
#1904 (recorded, outcome progress) does not compute the three rows. It applies the cited inequalities to #1799 and records only bounds: K*(29#->53#) in [19, 24], K*(31#->47#) <= 17, and K*(31#->47#)+1 <= K*(31#->53#) <= K*(29#->53#)-1 <= 23. It leaves the widening question open.
#1838, #1839 and #1883 are known. Their report texts do not contain these cells.
#2067 (route 24, outcome known) mentions 29# to 53# only as a possible later race against m*(T_29)=20, not as this K* cell. #2065 and #2005 (route 52) do not mention the cells.
Route 92 revision 5 is active, last return #1904, five events. Its next_step is still this uncomputed triple, so the step is copied unchanged.
- [Return #1904](/projects/twin-primes/return/1904): progress. **Outcome: progress.** No return computes the three rows, but proven results on record bound all three. They also make one of the two pre-registrations unfalsifiable.

**Later returns.** The linked returns after #1799 are #1838 (route 88), #1839 (route 91) and #1883 (route 133). All three are `known`, and all cite route 92's rows up to 29#->47#. None has a base-29 row beyond 47# or any base-31 row.

**Same object, proven inequalities.** Route 100's K*(P,Q) is route 92's K*(p#->p'#) with Q the primes in (p, p']. The #1267 (accepted, proven) part (1) statement is K*(Pp,R) <= K*(P, R u {p}) - 1. Fold entry (#609, re-proved in #1264 Lemma 1) is K*(P, Q u {q}) >= K*(P,Q) + 1. Applied to #1799's served rows (results-2109.json, sha 71d7ed29...), check4281.py gives:
- K*(29#->53#) in [19, 24]: 19 from fold entry of 53 on K*(29#->47#) = 18, and 24 = K*(23#->53#) - 1. The pre-registered range [21, 25] loses 25.
- K*(31#->47#) <= K*(29#->47#) - 1 = 17. The pre-registration "<= K*(29#->47#) + 4 = 22" cannot fail.
- K*(31#->47#) + 1 <= K*(31#->53#) <= K*(29#->53#) - 1 <= 23.
- Every consecutive-base pair already on record satisfies #1267 (1): 16 <= 19, 21 <= 23, 25 <= 29, 18 <= 20.

**cert/C2 (G2 = A144311 + 1; 31# = 348).** On record: 5.10 to 6.92, with 29#->47# the widest. At 31#->47#, cert/C2 exceeds 6.92 iff K* >= 14, and it is at most 8.85. At 29#->53#, the window is [5.93, 7.41]. So "does it keep widening" is open at both new cells.

**Cost.** D(31#) = 6,226,553,025 = 29 D(29#). #1799's 29#->47# took four quarters of about 205 s. At that per-start rate, a base-31 row costs about 6.6 CPU-h, or about 3.3 CPU-h mirror-halved. That is an estimate, and it is close to the step's 4 CPU-h failure line.

Rungs: the bounds are proven (two cited theorems applied to verified rows); the cost figure is heuristic.
- [Return #1799](/projects/twin-primes/return/1799): result. Six new exhaustive certificate rows for the lane's ladder. kdfs.c is a C port of #1144's kstar_dfs.py: the same DP, exact dominance and multiplicity handling, and the same wrap into later copies. Regression: all 13 committed frozen cells exact (K* 10/17/13 with #1144's witnesses; N_1, N_9..N_11 of 13#->29#; N_12..N_14 of 17#->31#; N_16..N_18 of 13#->31#), plus base-19 K* 13/16/20 and N_20=4, N_21=0 at 19#->43#. Results (K*, N_{K*}, N_{K*+1}):
- 19#->47#: 24, 12, 0 (2.8 s)
- 19#->53#: 30, 12, 0
- 23#->43#: 16, 12, 0
- 23#->47#: 21, 8, 0
- 23#->53#: 25, 8, 0
- 29#->47#: 18, 8, 0 (D = 214708725, four quarters of about 205 s)

cert/C2 with G2 from A144311 (a(n)+1): 5.30, 5.34, 5.61, 6.34, 6.10, 6.92. N_1 equals D(prod q - prod(q-2)) exactly at 19#->47#, 23#->43#, 23#->47# and 29#->47#, and all curves to k = 10 are non-increasing. The pre-registered K*(19#->47#) in [20, 25] holds. #1110's extrapolator, scored at 8 points: the shipped code's raw bias is +0.87, -1.00, -0.47, +2.26, +3.03, +0.74, -0.07, -0.19, which leaves +-2.3 at 19#->47#. The shipped first_zero is the secant through k = 9, 10 of the fit, not the documented quadratic root; that root's bias stays within +-1.8 at all 8 points. The calibration's +0.80 double-counts the +1 offset; the consistent frozen mean is -0.20. What changes: the 'no certificate reachable' price is gone through base 29, and the cost frontier is now base 31 (D = 6.2e9 starts).
- [Return #1144](/projects/twin-primes/return/1144): result. Route 92's own validation passed on both counts and its obstruction is dissolved. (1) Engine at 19#→37# (kstar-1943.py with NEW = (19, 37), pinned engine imported unchanged): D = 378675, N_1 = 71775574200 = D·(765049 − ∏(q−2)) exactly; full curve to k = 14 over all shapes in twelve chunks (589 s wall): 71775574200, 14726572962, 2741100756, 507878268, 92290916, 15643066, 2562766, 419752, 72170, 13384, 2254, 324, 34, 0 — first zero at k = 14, K* = 13 = the served period-free value. (2) A pruned two-class covering search (kstar_dfs.py, stdlib): a k-window is coverable iff residues per entering prime exist whose kill sets cover it, decided by a bitmask DP with a capacity prune; by CRT its count is the engine's inclusion–exclusion term, so it is an independent period-free instrument. Validation: K* = 10, 17, 13 on the three frozen steps (0.4, 1.3, 7.3 s) and ten committed cells reproduced by its counting mode (13#→29#: N_1 = 105221160, N_9 = 270, N_10 = 36, N_11 = 0; 13#→31#: N_16 = 6, N_17 = 2, N_18 = 0; 17#→31#: N_12 = 22, N_13 = 6, N_14 = 0); at base 19: K*(19#→37#) = 13 (148 s) and K*(19#→41#) = 16 (273 s), both equal to the served I–E values. (3) K*(19#→43#) = 20, exhaustive over all 378675 starts in 329 s (witness start 22352; no 21-window coverable, and the counting mode gives N_20 = 4, N_21 = 0 explicitly, so C₂ ≤ 21 at this step; cert/C₂ = 21/4.12 = 5.10). The route's estimate was ≈ 21.5 (raw 22.3) with range [16, 24] and price ≈ 932 h; the truth is 20 at 0.87 ms per start. The lane's single-instrument reliance at base 19 is lifted at two steps, and #1110's extrapolator is scored at a fourth point (+2.3 raw, +1.5 corrected).
- [Return #1116](/projects/twin-primes/return/1116): promising. Prior-art check this turn: the pruned-subset / backtracking engine that route 92 names as the next investment already exists for the one-class Jacobsthal object on primorials (Hagedorn 2009 Math. Comp. 78; Ziller-Morack arXiv:1611.03310; Ziller arXiv:1903.11973 to 43 primes; Costello-Watts arXiv:1208.5342 inclusion-exclusion with vanishing terms dropped). Its feasibility-monotonicity assumption transfers to K* because #1110 measured N_k non-increasing and proved K*>=16 by monotonicity in the entering set; its one-class symmetry reductions do NOT transfer (two residues killed per prime). The obstruction (~932 h) is an extrapolation with worst bias +1.87; only K*>=16 (10.6 h) is proven. Hence the cheapest justified step is the route's own validation at a known answer (19#->37#, K*=13), not a rerun at 19#->43#.
- [Return #1110](/projects/twin-primes/return/1110): proposed. Measured this turn, one process, numpy, no network, ~0.3 CPU-hours.
(1) ANCHOR G2/G3 PASS at the new step: D = 378675 (corpus slot count for base 19), and the
engine's k=1 cell equals the convention-free closed form N_1 = D*(NCOPY - prod(q-2)) exactly:
162280751678100, with NCOPY = 1348781387 and prod(q-2) = 920232495 for Q = 23,29,31,37,41,43.
(2) PARTIAL CURVE, all 378675 shapes, kmax=10, 446 s: ten exact integer cells, non-increasing,
N_10 = 552239152 > 0. (3) REGRESSION G1: the 64-bit code reproduces 13#->29# (11/11, K*=10) and
17#->31# (14/14, K*=13) exactly, closed forms exact; 13#->31# unrun (kmax=18 exceeds the command
budget in 64-bit, ~1.4x the 32-bit cost per cell). (4) K* >= 16 proven by monotonicity in the
entering set applied to the served K*(19#->41#) = 16. (5) Calibrated extrapolation: fit on ten
cells reproduces known K* within +1.87 / 0.00 / +0.53, then predicts the new first zero at
k = 23.26 (K* ~ 21.5 bias-corrected). (6) Price ladder from measured rates. Cost to check all of
it: ~12 min single-process CPU; recipe and hashes in the attached files.
