{ "outcome": "proposed", "evidence_md": "MEASURED (exact computation, one bounded exec, two independent executions agreeing on every number; ledger job2456-checks.py 11/11 PASS, 100.3 s / 100.9 s). Object: the tile gap word of the twin-admissible residues r mod x# (odd, gcd(r(r+2),x#)=1), qualify(g,q) <=> g mod q in {0,2,q-2}, R(q) = longest CYCLIC run of consecutive qualifying gaps, loose support = R(q)+1 (N-1840-01, return #971).\n\nControls at T23 (second, independent whole-period builder): D = 7 952 175, G2 = 204, R(29)=R(31)=R(37)=2, max R = 2 over 29..1009, R(q) = 0 for all 123 primes 206 < q <= 1009. All of N-1840-01's T23 numbers reproduced exactly.\n\nNEW at T29 (P29 = 6 469 693 230, D = 214 708 725, G2 = 258, 41 distinct gap values, gaps >= 240 exactly {240,258} - the three job #1645 controls hold): max R = 3, attained at q = 29 and 31; loose support bound max R + 1 = 4; the ladder primes (R >= 1) are the 21 primes 29..113; R >= 2 at 29,31,37,53,59,61; R >= 3 at 29,31; the 46 candidate primes q <= G2+2 = 260 were measured individually and the remaining 935 primes are exactly 0 by the structural argument (no gap <= 258 is 0, +2 or -2 mod q), confirmed on three measured probes 263, 509, 1009. The pre-registered falsifier of N-1840-01's next_step - some q with R >= 8, i.e. R+1 > 8 - does NOT fire at T29 (support bound 4), and does not fire at T23 either (3). The LMAX = 8 truncation stays idle one rung higher, and now for a measured reason: the qualifying primes are the small set 29..113.\n\nNEGATIVE (ledger job2456-linear.py, 1.0 s, all_pass False by design): with the seam gap removed from the word, R_linear(29)=R_linear(31)=R_linear(37)=2 at T23, so a run of r qualifying gaps predicts a path of r+1 nodes and L >= R_linear+1 = 3 at all three primes; the record's measured components are L(T23,29)=2, L(T23,31)=3, L(T23,37)=2 (#1840 / #159). The inequality is violated at q = 29 and q = 37 and holds at q = 31. So the loose run is NOT a component count under \"a qualifying gap is an edge between its two residues\" as read from the gap word. Scope: bounded negative; the served edge/node convention was not read, so the identification is open, not closed.\n\nFiles: runs/run_20260919_074908_Yz12pA/work/src2456/{job2456-checks.py, job2456-checks.log, job2456-checks.run1.log, job2456-checks.err, job2456-addendum.py, job2456-addendum.log, job2456-linear.py, job2456-linear.log, REPORT.md}. Reused unchanged: sieve_chunk of runs/run_20260917_131856_rcTy4g/work/job1645/src/job1645-t29gap.py (job #1645).", "proposal": { "title": "The loose-support ladder R(q) as a kill-graph instrument: does a maximal qualifying run sit inside one kill-graph component?", "contribution_md": "OBJECT. For each fold x and fold prime q, two finite objects already exist in the record with no proved relation: the loose run R(q) (longest cyclic run of consecutive qualifying gaps of the tile word; N-1840-01 / #971) and the kill-graph largest component L(T_x,q) (#1644 / #856 law, #161 diagonal, #1840 measurement). This route makes their relation the object: does a maximal run of consecutive qualifying gaps lie inside one component, and if so with which convention?\n\nSTEP THAT WOULD HAVE TO HOLD. A maximal run of r consecutive qualifying gaps is a path of r+1 nodes in any graph whose edges are those gaps, so L(T_x,q) >= R_linear(q)+1 (R_linear = the non-wrapping run, since a wrapped run is not a path in the linear word). Measured here at T23: R_linear = 2,2,2 at q = 29,31,37 -> predicted L >= 3,3,3 against recorded L = 2,3,2. The step is REFUTED at q = 29 and q = 37 as read from the gap word, so either the served kill graph's edge/node convention differs (it is defined in a served source, not in the gap word) or L is not the largest component in that sense. The route does not assume the identification; it reads the defining source first and then tests the corrected reading.\n\nEXACT DIFFERENCE FROM THE NEAREST PRIOR WORK. #971 established support = R+1, the structural collapse R = 0 for q > G2+2, and recorded max R = 2 at T23 with no T29 rung. #1840's successor question asked for exactly the T29 ladder, which this run answers (max R = 3, support bound 4, ladder primes 29..113, falsifier not fired) and extends with the T23 side (16 ladder primes, top 103, first zero 107) and with the negative above. #1645/#856 fix the kill-graph threshold for the T29 column (252 is not a gap) but say nothing about runs. Nothing in the record compares R with L.\n\nWHAT IT WOULD BUY. If a run sits inside one component, the ladder becomes a component predictor for every fold where no census exists - including the T29 column, where the record's diagonal has only the quoted L(T29,31) = 4 and no L(T29,29). If it does not, the two objects stay independent and the negative is on the record with a counted counterexample pair.\n\nCHEAPEST FIRST CHECK. Parse the served defining source (GET /projects/twin-primes/docs/research/a3-08-adjacent-pairs.js, one request, 0 CPU-h) for the exact edge and node convention, then recompute R_linear/R_cyclic (T23 already measured here, 1 s) and compare against the three recorded T23 components; only then spend a census at T29.", "prior_art_md": "Nearest prior work, all citable: (1) return #971 / job #1840, note N-1840-01-loose-support-is-r-plus-one.md - the support identity support = R+1, the structural collapse R(q) = 0 for q > G2(T_x)+2, the measured T23 ladder (max R = 2 over 170 primes 29..1009, R = 2 at q = 29,31,37), and the pre-registered successor question answered here. (2) job #1644 / return #856 - the gap-multiset support law L(T_x,p) >= 2 iff some tile gap is 0, +2 or -2 (mod p); this is the same predicate as qualify(g,q), which is why the two ladders' prime sets coincide while their sizes need not. (3) job #1645 (artifact job1645-t29gap.json) - the constant-memory segmented T29 gap word, D(T29) = 214 708 725, G2(T29) = 258, and the deciding fact 252 is not a gap of T29, which is also why the loose ladder is 0 at q = 127 at T29: among gaps <= 258 the only candidate in the three classes mod 127 is 252. (4) #159's recorded longest qualifying gap runs (2 at T23 by 31, 2 at T23 by 37, 3 at T31 by 37, 3 at T37 by 41) and its LMAX = 8 truncation, which the T29 ladder now also leaves idle. (5) #161's diagonal L(T_{p-},p) = 2,1,2,2,2,3,2,4 at folds 7..31 with the tile taken at the prime below the fold - the only component values available at T29 are the quoted L(T29,31) = 4 and the T23 measurements 2,3,2 used here as controls. Served sources to read before any census: docs/research/a3-08-adjacent-pairs.js (edge rule) and docs/research/OUTCOMES.md (closed routes).", "uncertainty_md": "NOT CLAIMED: no asymptotic statement, nothing about the target exponent, the margin, G2, or any level of distribution; the ladder is a finite statistic at two folds. The support identity support = R+1 and the collapse R(q) = 0 above G2+2 are prior art reused, not reproved. The negative is scoped: it refutes L >= R_linear+1 only under the reading \"a qualifying gap is an edge between its two residues\", which the gap word alone does not establish - the served kill-graph source was not read in this run (per the #1646 lesson, the defining artifact decides). The cyclic/linear distinction is a real degree of freedom: R_cyclic = 2 at all three T23 primes measured, R_linear = 2 as well, so the seam is not the explanation of the mismatch at q = 29 and 37. The T29 ladder's structural step (q > G2+2 => R = 0) is exact given the measured gap maximum 258, and all 46 candidate primes q <= 257 were measured, so the T29 ladder is complete for primes 29..1009 up to that structural step. The T31 rung is untested: D(T31) = 6 226 553 025 gaps cannot be materialised (12 GB as uint16), so the next rung needs a streaming qualifier per prime rather than a word build - an engineering step with a measured shape but no measurement yet. Attribution: the model that did this work is the session's declared deepseek/deepseek-v4-flash at effort unmeasured; usage stays pending." }, "next_step": { "question": "Does a maximal run of consecutive qualifying gaps of T_x sit inside one component of the served kill graph - i.e. does the served edge/node convention make L(T_x,q) = R_linear(q) + 1 hold at the three recorded T23 values and then at T29/q = 29,31?", "method": "Read the defining served source first: GET /projects/twin-primes/docs/research/a3-08-adjacent-pairs.js (one request, 0 CPU-h) and extract the exact edge rule, the node set and the isolated-node convention. Then rerun the local ledgers unchanged (job2456-linear.py, 1 s) to recompute R_linear and R_cyclic at T23 for q = 29,31,37 against the recorded L = 2,3,2; if and only if the corrected convention reproduces all three, rebuild the T29 gap word with the job #1645 segmented sieve (sieve_chunk, ~100 s, 429 MB as uint16) and run the same component census the record uses at q = 29 and q = 31, reporting L(T29,29) (record has no value; the run-based prediction is 3 or 4) and L(T29,31) against the quoted 4.", "success": "A convention read from the served source under which L = R_linear + 1 at all three T23 values, and a T29 census at q = 29 with L in {3,4} (and 4 at q = 31, matching the quoted diagonal entry) - which turns the measured ladder into a component predictor and gives the T29 column an exact component value the record does not have.", "failure": "No reading of the served source reproduces the three T23 values, or the T29 census gives L(T29,29) < 3 - then the run/component identification is closed with that counted counterexample and the ladder keeps its own meaning only (support bound 3 at T23, 4 at T29).", "budget_hours": 1.5, "compute": { "cpu_hours": 0.5, "ram_gb": 8, "disk_gb": 1 }, "required_tools": [ "numpy", "python3", "sah14-exec" ], "required_sources": [ "a3-08-adjacent-pairs.js", "outcomes.md", "return-971", "return-856", "job-1645-t29gap-artifact" ] } }