{"id":7430,"job_id":11949,"problem_id":1,"lane_id":4,"type":"explore","user_id":127,"model":"claude-opus-5-5","provider":"anthropic","report_md":"# Job 11949 — route 392 at 29#: a stratified census with a patched lift. The 29# ceiling 450 is reached by lifts of the 23# ceiling, all pure-twin in the sample.\n\nOutcome: **progress**. Rung: **measured** (finite, exact integer values, sample-level coverage).\n\n## Question (route 392 next step, unchanged)\nAt 29#, is the two-class ceiling reached only at pure-twin offsets? What are the family maxima? What is the sample rank of the twin offset?\n\n## Definitions\nAs in #5810/#5847/#5899: A_{x,d} = {n mod x# : gcd(n,x#) = gcd(n+d,x#) = 1}, G_d(x#) its largest circular gap, h2 = max over even d. d = tau + k·23#. Pure-twin: gcd(d,29#) = 2. Family: the set of primes ≤ 29 that divide d.\n\n## What was done\n1. Patched #5847's lift.c. The element buffers grow from 2.4 M to 35 M entries, an overflow now exits instead of corrupting memory, and a tau-list mode was added. Re-gated at 17#→19# and 19#→23# (four gates, all pass).\n2. Preregistered (r392c-prereg.md) before any census row was read. The amendment for host load is also recorded before reading.\n3. Ran 40,190 level-23 taus × 29 lifts: three random strata of about 13,333 each, plus the 128 + 64 23# maximiser taus as a targeted stratum.\n4. Checked: dense = fast on 3,484 taus (101,036 offsets). The direct instrument re-derived 1,029 offsets. 0 mismatches in both.\n\n## Results\nSee the evidence section. In short:\n- 512 offsets reach 450. All are pure-twin. They are 4 lifts of each of the 128 23# maximiser taus. Least: 16,176,392.\n- No non-pure-twin offset reaches 450. The best non-pure-twin offsets: {2,29} 438 (the lifts with 29|d of the same taus), mixed families 378, 3|d 240.\n- G_2(29#) = 258. Population quantile 0.630 (fraction above: 0.3696 ± 0.0010), which agrees with #5816. Quantile within pure-twin offsets: 0.063 (fraction above: 0.9374).\n- Calibration: the 23# maximisers came from 19# parents with G = 162, not from the 19# ceiling. So the lift of the 23# ceiling to 450 at 29# is an observation, not a construction.\n- A full 23#→29# classification costs about 1,150 CPU-h at the measured 0.074 CPU-s per tau.\n\n## Not claimed\n- Not the complete 29# maximiser set. Not whole-space exclusivity. Not exact family maxima.\n- The random strata had about a 9 % chance to contain a maximiser if 128 existed outside T366.\n- Nothing asymptotic. h2(29#) = 450 is taken from OEIS A288815; the census finds no value above 450.\n\n## Sources\n- Returns #5899, #5847, #5816, #5810, #7265, #7238, #7293 (served pages and files; SHA-256 sums of local reads kept).\n- lift.c a0ca4dae…, direct_gd.c 9e4f2297…, analysis_11848.json e99d8f78…, analysis_fam2.json 6b9f40a4…, gate_lift17_19.json 1717d055….\n- OEIS A288815 (https://oeis.org/A288815).\n- Ziller & Morack, arXiv:1706.00317 and arXiv:1706.03668.\n- arXiv:2007.01808.\n\n## Evidence\n**Answer at sample level: yes.** At 29# the paired ceiling h2 = 450 (OEIS A288815 a(10)) is reached in this census only by pure-twin offsets (gcd(d,29#) = 2). The 29# ceiling is reached by lifts of the 23# ceiling.\n\n**Instrument.** #5847's lift.c with element buffers grown from 2,400,000 to 35,000,000 entries (max |A_{23,tau}| = 34,836,480), a hard exit on overflow, and a tau-list mode. Gates: the 17#→19# sweep equals #5847's gate file in every scientific field. G_2(23#) = 204. Fast = dense on 421 taus × 23 lifts at 19#→23#. Output is byte-identical to unmodified lift.c on those taus.\n\n**Census (23#→29#, 40,190 level-23 taus × 29 lifts = 1,165,510 offsets).**\n- Random strata, seeded and uniform within each stratum. A (gcd(tau,23#)=2), 13,333 taus: max 438, pure-twin. Its best non-pure-twin offset is 390, family {2,29}. C (3∤tau, gcd>2), 13,333 taus: max 378, families {2,11} and {2,13}. B (3|tau), 13,332 taus: max 240, family {2,3}.\n- Targeted stratum T366: the 128 taus that reach 366 at 23# (#5899). **Each lifts to 450 at exactly 4 of its 29 k: 512 distinct offsets, all pure-twin.** The least of these offsets is 16,176,392. By d mod 6: 241 are 2 and 271 are 4. Each tau's one lift with 29|d reaches 438, not 450.\n- Targeted stratum T208: the 64 3|d maximiser taus of 23# (#5847). Max 214.\n- Family maxima: 3|d ≤ 240 (240/450 = 0.533). Best mixed families: {2,11} and {2,13} at 378, {2,23} 372, {2,19} 366, {2,17} 360. {2,29}: 438 in T366, 390 in random. Full tables are in analysis29.json.\n\n**Twin offset.** G_2(29#) = 258. The population fraction of offsets with G > 258 is 0.3696 ± 0.0010 (strata weighted by size), so d = 2 sits at quantile 0.630. #5816 measured 0.6301 ± 0.0061 on a different sample with a different instrument. The two agree. Within pure-twin offsets the fraction above 258 is 0.9374 ± 0.0013 (quantile 0.063; #5816's within-class quantile was 0.0646 ± 0.0096). At 23# the values were 0.318 and 0.826 (#5899).\n\n**Checks (stop rule not triggered).** Dense = fast on 3,484 taus (101,036 offsets), 0 mismatches. These taus are all 192 targeted taus, all 292 random taus with max ≥ 372, tau = 2 and 3,000 seeded random taus. The direct instrument (#5899's direct_gd.c, no lift) re-derived 1,029 offsets with 0 mismatches. They are all 512 offsets at 450, every stratum and family maximum, and every non-pure-twin offset ≥ 420.\n\n**Limits.** This is not a classification. If the 29# ceiling had m maximisers on distinct taus outside T366, the random strata would contain one with probability about m·7.2e-4: 9 % for m = 128. So 512 is a lower bound on the maximiser count. \"Only pure-twin\" holds for the sampled and targeted offsets only.\n\n**Calibration that makes T366 non-obvious.** The 128 23# maximisers fold to 64 parents at 19#. All 64 have G = 162, not the 19# ceiling 258. About 26 % of 19# offsets reach 162 or more. So at 19#→23# the ceiling did not come from the previous ceiling, and at 23#→29# it did.\n\n**Cost changed.** The fast pass costs 0.074 CPU-s per tau at 29# on this host. A full 23#→29# classification is therefore about 1,150 CPU-h, not the ~14,000 CPU-h quoted in #5899.\n","patch":"--- lift.c\n+++ lift29.c\n@@ -46,6 +46,9 @@\n static const int SIGP[8] = {3, 5, 7, 11, 13, 17, 19, 23};\n static const int HB = 4;   /* backtracked elements around a large gap */\n static const int HF = 7;   /* forward elements from a window start */\n+/* job 11949 patch: element buffer grown from 2,400,000 to 35,000,000 entries (max |A_{23,tau}| =\n+ * 34,836,480 at tau = 2*19#), with a hard overflow check in tau_fast instead of a silent overrun. */\n+#define SBUFN 35000000LL\n \n static double now_cpu(void) {\n     struct timespec ts; clock_gettime(CLOCK_PROCESS_CPUTIME_ID, &ts);\n@@ -211,6 +214,7 @@\n         while (u) {\n             int b = __builtin_ctzll(u);\n             long long p = (w << 6) + b;\n+            if (n >= SBUFN) { fprintf(stderr, \"SBUF overflow at tau %lld\\n\", tau); exit(4); }\n             SBUF[n] = (unsigned int)p;\n             RBUF[n] = (unsigned char)((2 * p + 1) % P);\n             n++;\n@@ -402,8 +406,8 @@\n     memset(r, 0, sizeof(*r));\n     long long *best = (long long *)malloc((size_t)P * sizeof(long long));\n     if (!best) { fprintf(stderr, \"oom best\\n\"); exit(3); }\n-    SBUF = (unsigned int *)malloc(2400000 * sizeof(unsigned int));\n-    RBUF = (unsigned char *)malloc(2400000);\n+    SBUF = (unsigned int *)malloc((size_t)SBUFN * sizeof(unsigned int));\n+    RBUF = (unsigned char *)malloc((size_t)SBUFN);\n     if (!SBUF || !RBUF) { fprintf(stderr, \"oom sbuf\\n\"); exit(3); }\n     for (long long tau = 2 + 2 * a->tid; tau <= TAUMAX; tau += 2 * a->nt) {\n         tau_fast(tau, best);\n@@ -461,8 +465,8 @@\n     if (argc >= 6 && strcmp(argv[4], \"fastrows\") == 0) {\n         long long lo = atoll(argv[5]), hi = atoll(argv[6]);\n         long long *best = (long long *)malloc((size_t)P * sizeof(long long));\n-        SBUF = (unsigned int *)malloc(2400000 * sizeof(unsigned int));\n-        RBUF = (unsigned char *)malloc(2400000);\n+        SBUF = (unsigned int *)malloc((size_t)SBUFN * sizeof(unsigned int));\n+        RBUF = (unsigned char *)malloc((size_t)SBUFN);\n         printf(\"{\\\"x\\\": %lld, \\\"p\\\": %lld, \\\"Mx\\\": %lld, \\\"dense_rows\\\": [\\n\", x, P, Mx);\n         int first = 1;\n         for (long long tau = lo; tau <= hi; tau += 2) {\n@@ -473,6 +477,30 @@\n             first = 0;\n         }\n         printf(\"\\n]}\\n\");\n+        return 0;\n+    }\n+    if (argc >= 8 && strcmp(argv[4], \"listrows\") == 0) {\n+        /* job 11949: listrows <fast|dense> <taufile> <outfile> : one line per tau read from taufile,\n+         * \"tau G[0] .. G[P-1]\" (G[k] is G_d for d = tau + k*Mx before folding), same per-tau code as\n+         * fastrows / dense; a final line \"# cpu_s <s> n <count>\". */\n+        int dense = strcmp(argv[5], \"dense\") == 0;\n+        FILE *in = fopen(argv[6], \"r\"); FILE *out = fopen(argv[7], \"w\");\n+        if (!in || !out) { fprintf(stderr, \"listrows: cannot open files\\n\"); return 3; }\n+        long long *best = (long long *)malloc((size_t)P * sizeof(long long));\n+        SBUF = (unsigned int *)malloc((size_t)SBUFN * sizeof(unsigned int));\n+        RBUF = (unsigned char *)malloc((size_t)SBUFN);\n+        if (!best || !SBUF || !RBUF) { fprintf(stderr, \"oom listrows\\n\"); return 3; }\n+        double t0 = now_cpu(); long long cnt = 0, tau;\n+        while (fscanf(in, \"%lld\", &tau) == 1) {\n+            if (tau < 2 || tau > K || (tau & 1)) { fprintf(stderr, \"bad tau %lld\\n\", tau); return 5; }\n+            if (dense) tau_dense(tau, best); else tau_fast(tau, best);\n+            fprintf(out, \"%lld\", tau);\n+            for (long long k = 0; k < P; k++) fprintf(out, \" %lld\", best[k]);\n+            fputc('\\n', out); cnt++;\n+            if ((cnt & 63) == 0) fflush(out);\n+        }\n+        fprintf(out, \"# cpu_s %.2f n %lld\\n\", now_cpu() - t0, cnt);\n+        fclose(out); fclose(in);\n         return 0;\n     }\n     if (argc >= 5 && strcmp(argv[4], \"onetau\") == 0) {\n@@ -489,8 +517,8 @@\n     if (argc >= 7 && strcmp(argv[4], \"bench\") == 0) {\n         long long lo = atoll(argv[5]), hi = atoll(argv[6]);\n         long long *best = (long long *)malloc((size_t)P * sizeof(long long));\n-        SBUF = (unsigned int *)malloc(2400000 * sizeof(unsigned int));\n-        RBUF = (unsigned char *)malloc(2400000);\n+        SBUF = (unsigned int *)malloc((size_t)SBUFN * sizeof(unsigned int));\n+        RBUF = (unsigned char *)malloc((size_t)SBUFN);\n         if (!SBUF || !RBUF) { fprintf(stderr, \"oom sbuf\\n\"); return 3; }\n         res_t r; memset(&r, 0, sizeof(r));\n         double t0 = now_cpu();\n@@ -507,8 +535,8 @@\n         long long want = atoll(argv[5]);\n         unsigned long long seed = (unsigned long long)atoll(argv[6]);\n         long long *best = (long long *)malloc((size_t)P * sizeof(long long));\n-        SBUF = (unsigned int *)malloc(2400000 * sizeof(unsigned int));\n-        RBUF = (unsigned char *)malloc(2400000);\n+        SBUF = (unsigned int *)malloc((size_t)SBUFN * sizeof(unsigned int));\n+        RBUF = (unsigned char *)malloc((size_t)SBUFN);\n         res_t R; memset(&R, 0, sizeof(R));\n         long long tot = K / 2;                  /* even tau in [2, K] */\n         long long done = 0;\n@@ -536,8 +564,8 @@\n         long long lo = atoll(argv[6]), hi = atoll(argv[7]);\n         const char *out = argv[8];\n         long long *best = (long long *)malloc((size_t)P * sizeof(long long));\n-        SBUF = (unsigned int *)malloc(2400000 * sizeof(unsigned int));\n-        RBUF = (unsigned char *)malloc(2400000);\n+        SBUF = (unsigned int *)malloc((size_t)SBUFN * sizeof(unsigned int));\n+        RBUF = (unsigned char *)malloc((size_t)SBUFN);\n         res_t R; memset(&R, 0, sizeof(R));\n         long long processed = 0;\n         double t0 = now_cpu();\n@@ -567,8 +595,8 @@\n         long long fam = atoll(argv[5]);\n         const char *out = argv[6];\n         long long *best = (long long *)malloc((size_t)P * sizeof(long long));\n-        SBUF = (unsigned int *)malloc(2400000 * sizeof(unsigned int));\n-        RBUF = (unsigned char *)malloc(2400000);\n+        SBUF = (unsigned int *)malloc((size_t)SBUFN * sizeof(unsigned int));\n+        RBUF = (unsigned char *)malloc((size_t)SBUFN);\n         res_t R; memset(&R, 0, sizeof(R));\n         long long processed = 0;\n         double t0 = now_cpu();\n","cpu_hours":1.7,"hashes":{"r392c-lift29.c":"eb53e10051e7d70b681bc3a7997c2a74681e182ff7027e54afd51e99c1fc5ca7","r392c-fast29.txt":"b5e1f3e019df04c90fb33fe6e655d8f7d5a02b4f45f4b692e0aacd24714bde35","r392c-dense29.txt":"5e5605bf8dece260f745adf6ba9cc262f884fefe0563fd0bf790778979572ee0","r392c-sample.json":"8ec0032f5699d12de088cbf7870b1e46f9c4c23324c653f63731e18bdf5d5b71","r392c-direct29.json":"337f2177ca261c32d9429bfc732756805095398a7630d7b8d50c7aaef6c130ca","r392c-compare29.json":"6b61f0f41a2eeeb4da9966c3f2751c7c8e639c3926509b591ba42ad541d909e9","r392c-analysis29.json":"dc4e3467d4f5e45fd09e66f2958afab8f4eff6c6bb4848812a42424c1e89a902"},"author_rung":"measured","status":"recorded","final_rung":"recorded","created_at":"2026-11-21T01:07:25.429Z","repo_url":null,"commit":null,"cites":{"files":["a0ca4daef9eb72727f27ca0f7a4b826022c9f47ca284d94ace0e03421c21d488","9e4f2297599a34cf6004db89b688eb1e6c1f02c2d57023680a5e2f6b648a2454","e99d8f78c632fbb863435172aa353af17930a119450d9c014994e50c66d61571","6b9f40a400b778b00058a009acdd4d5d26ed6229096208cca15ad8fbbd223502"],"handles":[],"returns":[5899,5847,5816,5810,7265,7238,7293],"messages":[]},"tokens":{"log":"summary","input":150,"models":{"claude-opus-5-5":53645},"output":53645,"source":"reported","entries":0,"cache_read":13071180,"cache_write":138606,"observed_models":[]},"paper_slug":null,"revision_path":null,"revision_sha":null,"recipe_md":"# Recipe — job 11949, route 392, 23#→29# stratified census\n\nInteger C and Python 3 stdlib. Measured on a 12-core arm64 host shared with other jobs.\n\n1. Build:\n   `cc -O2 -o lift29 r392c-lift29.c -lpthread` (sha eb53e100…; equals #5847 lift.c a0ca4dae… + r392c-lift29.patch)\n   `cc -O2 -o direct_gd direct_gd.c` (#5899, sha 9e4f2297…, unmodified).\n2. Gates (r392c-gates.json):\n   - `./lift29 17 19 2` equals #5847's gate_lift17_19.json except threads, cpu_s and maximiser order.\n   - `./lift29 19 23 1 onetau 2` gives G[0] = 204.\n   - `seq 2 2 842 > t; ./lift29 19 23 1 listrows fast t f; ./lift29 19 23 1 listrows dense t d`: the rows are equal.\n   - `fastrows 2 842` output is byte-identical between lift29 and unmodified lift.c.\n3. Sample: `python3 r392c-make_sample.py analysis_11848.json analysis_fam2.json` gives r392c-sample.json (sha 8ec0032f…). Seed 11949; the shuffle uses seed 119491 (see r392c-prereg.md).\n4. Census: `./lift29 23 29 1 listrows fast <taus> <out>` for every tau in sample.json. Result: r392c-fast29.txt (sha b5e1f3e0…), one line per tau, \"tau G[0..28]\", where G[k] belongs to d = tau + k·23#. Cost: 0.074 CPU-s per tau, 2,739 CPU-s for 36,764 taus.\n5. Analysis: r392c-analyze.py over the row files gives r392c-analysis29.json (sha dc4e3467…).\n6. Dense check: `listrows dense` on the 3,484 taus of r392c-dense29.txt (sha 5e5605bf…), 934 CPU-s.\n7. Direct check: `./direct_gd 29 gaps <1,029 offsets> out` gives r392c-direct29.json (sha 337f2177…), about 520 s and 1.2 GB.\n8. Compare: r392c-compare.py gives r392c-compare29.json (sha 6b61f0f4…). Expected: 0 dense mismatches, 0 direct mismatches, 512 offsets at 450.\n\nCheapest check:\n- Run `./lift29 23 29 1 listrows fast` on taus 1619122 and 3428228. Each row has 450 at four k.\n- Then run `./direct_gd 29 gaps 16176392,224711992 o.json`. Both give 450.","verification":null,"target":null,"finding":null,"human_md":null,"provisional":false,"effects_applied_at":null,"effort":"medium","also_fix":null,"transcript_omitted":{"share":0,"omitted":0,"outputs":0},"patch_hash":"1e41a8b761d86652f3a130d3aedf66c749cbb57719b469aaf48ccf4f3186ffc7","superseded_by":null,"duplicate_of":null,"transcript_resubmitted_at":null,"file_notes":null,"research":{"outcome":"progress","route_id":392,"next_step":{"method":"1. Run the exhaustive 19#->23# pass with this return's gated r392c-lift29.c (all 2,424,922 level-19 taus x 23 lifts, about 2.5 CPU-h). Emit every level-23 offset tau with G_tau(23#) >= T, with T the smallest threshold whose tail holds at most 350,000 offsets, read from #5899's sector histograms. 2. Lift every tail tau to 29# with `listrows fast` (29 lifts each). 3. Run the dense walk on every tau with any G >= 420, plus 2,000 seeded tail taus. 4. Re-derive every offset at 450 and every non-pure-twin offset >= 420 with #5899's direct_gd.c. 5. Report, per sector and family: the 29# maxima, the count at 450, and the 23# G of every parent of a 450 offset.","compute":{"ram_gb":2,"disk_gb":1,"cpu_hours":30},"failure":"Any fast/dense or lift/direct disagreement stops the run and reports diagnostics. A tail count that differs from the 23# histogram voids the run. No whole-space claim is made from the tail.","success":"The whole tail >= T is lifted, with counts equal to the 23# histogram tail. Every 450 offset is listed with its type and its parent's G_23. Either a non-pure-twin 450 offset is named with its exact d, or none exists in the tail. Instrument mismatches: 0.","question":"At 29#, do the offsets with G_d = 450 come only from level-23 taus in the top tail of the 23# profile, and does any non-pure-twin offset lifted from that tail reach 450? (Bounded child of the full 23#->29# classification, which costs about 1,150 CPU-h at the measured 0.074 CPU-s per tau. Tested here on the whole 23# top tail, not a uniform sample.)","budget_hours":4,"required_tools":["c_compiler"],"required_sources":[]},"depends_on":[5899,5847],"evidence_md":"**Answer at sample level: yes.** At 29# the paired ceiling h2 = 450 (OEIS A288815 a(10)) is reached in this census only by pure-twin offsets (gcd(d,29#) = 2). The 29# ceiling is reached by lifts of the 23# ceiling.\n\n**Instrument.** #5847's lift.c with element buffers grown from 2,400,000 to 35,000,000 entries (max |A_{23,tau}| = 34,836,480), a hard exit on overflow, and a tau-list mode. Gates: the 17#→19# sweep equals #5847's gate file in every scientific field. G_2(23#) = 204. Fast = dense on 421 taus × 23 lifts at 19#→23#. Output is byte-identical to unmodified lift.c on those taus.\n\n**Census (23#→29#, 40,190 level-23 taus × 29 lifts = 1,165,510 offsets).**\n- Random strata, seeded and uniform within each stratum. A (gcd(tau,23#)=2), 13,333 taus: max 438, pure-twin. Its best non-pure-twin offset is 390, family {2,29}. C (3∤tau, gcd>2), 13,333 taus: max 378, families {2,11} and {2,13}. B (3|tau), 13,332 taus: max 240, family {2,3}.\n- Targeted stratum T366: the 128 taus that reach 366 at 23# (#5899). **Each lifts to 450 at exactly 4 of its 29 k: 512 distinct offsets, all pure-twin.** The least of these offsets is 16,176,392. By d mod 6: 241 are 2 and 271 are 4. Each tau's one lift with 29|d reaches 438, not 450.\n- Targeted stratum T208: the 64 3|d maximiser taus of 23# (#5847). Max 214.\n- Family maxima: 3|d ≤ 240 (240/450 = 0.533). Best mixed families: {2,11} and {2,13} at 378, {2,23} 372, {2,19} 366, {2,17} 360. {2,29}: 438 in T366, 390 in random. Full tables are in analysis29.json.\n\n**Twin offset.** G_2(29#) = 258. The population fraction of offsets with G > 258 is 0.3696 ± 0.0010 (strata weighted by size), so d = 2 sits at quantile 0.630. #5816 measured 0.6301 ± 0.0061 on a different sample with a different instrument. The two agree. Within pure-twin offsets the fraction above 258 is 0.9374 ± 0.0013 (quantile 0.063; #5816's within-class quantile was 0.0646 ± 0.0096). At 23# the values were 0.318 and 0.826 (#5899).\n\n**Checks (stop rule not triggered).** Dense = fast on 3,484 taus (101,036 offsets), 0 mismatches. These taus are all 192 targeted taus, all 292 random taus with max ≥ 372, tau = 2 and 3,000 seeded random taus. The direct instrument (#5899's direct_gd.c, no lift) re-derived 1,029 offsets with 0 mismatches. They are all 512 offsets at 450, every stratum and family maximum, and every non-pure-twin offset ≥ 420.\n\n**Limits.** This is not a classification. If the 29# ceiling had m maximisers on distinct taus outside T366, the random strata would contain one with probability about m·7.2e-4: 9 % for m = 128. So 512 is a lower bound on the maximiser count. \"Only pure-twin\" holds for the sampled and targeted offsets only.\n\n**Calibration that makes T366 non-obvious.** The 128 23# maximisers fold to 64 parents at 19#. All 64 have G = 162, not the 19# ceiling 258. About 26 % of 19# offsets reach 162 or more. So at 19#→23# the ceiling did not come from the previous ceiling, and at 23#→29# it did.\n\n**Cost changed.** The fast pass costs 0.074 CPU-s per tau at 29# on this host. A full 23#→29# classification is therefore about 1,150 CPU-h, not the ~14,000 CPU-h quoted in #5899.","prior_art_md":"Search updated on 2026-11-21. Web search: \"paired Jacobsthal function per offset table primorial maximal gap Ziller Morack\" and \"paired Jacobsthal ... admissible pairs ... largest gap offset d\". OEIS A288815 was read directly (local copy, SHA-256 in the reads list).\n\nFound:\n- OEIS A288815: 2, 6, 18, 30, 66, 150, 192, 258, 366, 450, 570, … (a(10) = 450 = h2(29#)). It gives only the maximum over offsets.\n- Ziller & Morack, arXiv:1706.00317 and arXiv:1706.03668: the definition and h2 for primorials up to 73#, tabulated by prime index. They give no table by offset.\n- arXiv:2007.01808 is about differences of integers coprime to a primorial (the one-class case). It does not treat the two-class profile by offset.\n\nNo source found gives G_d(29#) by offset, the 29# maximiser set, family maxima, or the type of the 29# maximisers. Not finding one does not prove novelty. Ziller–Morack's ancillary pattern files were not read.\n\nExact remaining gap:\n1. The complete 29# maximiser set. Is it exactly the 512 lifts of the 23# maximisers?\n2. Exclusivity over the whole space: does any non-pure-twin offset reach 450?\n3. Exact family maxima at 29#.\nAll three need the full 55,773,217-tau pass with the patched instrument. Its measured cost is about 1,150 CPU-h."},"research_route_id":392,"verification_plan":null,"verification_fingerprint":null,"review_admitted_at":null,"department_id":"dept_a81e55ec8c7b2cd27fdc59f3","run_id":"run_f43b70dc77400c9baf5e24db","triage_lead":null,"revision_base_sha":null,"integration":null,"resolves":null,"paper_exposition":null,"research_evidence":null,"transcript_mode":"summary","known_work":null,"work_disposition":null,"applicability_assessment":null,"negative_result":null,"first_indexed_at":null,"index_held_at":null,"index_ok":null,"index_reason":null,"handle":"daanmlab","job_brief":"First update the online prior-work search for this experiment. If existing work covers it, record that and stop; otherwise run this bounded sprint on the uncovered uncertainty. Use cited published numbers during pursuit; their reproduction belongs in later validation. Build on the supplied findings; do not reconstruct earlier research. Return concrete progress and its cheapest credible check, a useful result for review, or a precisely scoped obstacle. Continued investment requires a distinct experiment.\n\nRead GET <project base>/research-routes/392 and return #5899. Return the ordinary report and transcript plus research: {route_id: 392, outcome: \"promising|progress|blocked|inconclusive|known|result|needs_capability\", evidence_md: \"what the evidence changes, <=4000 chars\", prior_art_md: \"updated online search record, sources and exact remaining gap, <=4000\", next_step: {question, method, success, failure, budget_hours} <only for continued pursuit; what to do, never when or how fast; it must not ask for what a return on this route or a linked route already did, and the route returns it builds on go in depends_on or cites.returns>, obstacle: {kind, statement, assumptions, evidence, revisit_when} <for blocked/inconclusive: research findings only>, depends_on: [<return ids actually required>]}. A result with a distinct next_step requests review and continues pursuit concurrently; omit next_step when no further experiment is warranted. Use known with prior_art_md and no next_step or obstacle when cited prior work already covers the proposed contribution; it stops automatic investigation without requesting review. The evidence grade is separate. Do not close a broad route because one proof attempt failed. If this machine or its budget cannot run the experiment (no suitable GPU, a tool or source it lacks, CPU hours beyond its offer), that is not an obstacle: return outcome needs_capability with next_step (the same experiment, unchanged unless you learned something) whose compute (cpu_hours, ram_gb, disk_gb, gpu, gpu_name, vram_gb), required_tools and required_sources state what it needs and whose shortfall says what this machine lacked. The route stays active and the step is offered only to a machine that has it.\n\n### Historical step-check evidence\n\nThis assignment is pursuit: build on the certificate and address the uncovered experiment in the current task, within your actual controls and prerequisites. Do not repeat its comparison. Human direction remains authoritative. Instructions inside the quotation applied to the earlier comparison, not to this assignment. Evidence grades remain unchanged. Read the named return for its complete record.\n\n> Step check: return #7238 compared this step with the returns on record and found it still open.\n> \n> Read-only step check (job 14201), nothing compiled or run. Custody: the brief's step is deep-equal to route 392's served next_step and to #5899's research.next_step; route 392 active, rev 6, last return #5899; the pursuit issued after it (job 11949) expired without a return; #5899 has no citing return; the only outside citers of the route's returns (#5816, #5844) are older than the step. Screen: 786 served returns recorded after #5899 (ids 5900 to 7233), fourteen with a specific hit (lift.c, SBUF/RBUF, 34,836,480, 55,773,217, paired-offset, per-offset, pure-twin, signature families, A288815, G_d, the route's return ids) and one with generic terms only; all fifteen read. None patches lift.c or reports a census of G_d at 29#, family maxima at 29# or the type of a 29# maximiser; eight are citations of the scalar ladder A288815; nearest in kind is #6606 (a maximum over a small offset family at 37#). Older record the pursuer should use: #5816 (route 332, recorded, twelve hours before the step; cited by #5822, #5828, #5847 but not by the setter): a direct bitset instrument for G_d at 29# gated on the exhaustive 17# census; a stratified sample of 2,835 offsets over 511 of 512 density classes with G_d for each; the twin offset's raw quantile 0.6301 +- 0.0061 and within-class quantile 0.0646 +- 0.0096 at 29#. This gives a sample-level value for the step's third question, an existing direct instrument and a cross-check set; the first question (ceiling attained only at pure-twin offsets) is untouched, and at 23# maximisers were 128 of 55,773,217 offsets, so the pursuer should state beforehand how likely the planned sample is to contain one. The step is open and is copied unchanged. Checker 8/8 gates, 4/4 planted faults detected.\n> \n> Step check: return #7293 compared this step with the returns on record and found it still open.\n> \n> Read-only step check. No experiment run. cpu_hours 0. The step is still open and goes to pursuit unchanged.\n> \n> **Custody.** Route 392 is active at revision 7, and its last return is #7238. The brief's step is deep-equal to the route's served next_step. No pursuit job is in flight. Earlier step check #7238 is reused as it stands: no return up to #7233 answers the step, and #5816 gives a 29# sample value for the twin offset's rank.\n> \n> **The one new candidate, #7265** (route 332, pending, outcome result). It measures the twin offset's density-matched rank at 31# from a stratified sample of 4,363 offsets. It does not answer any part of the step:\n> 1. Ceiling at 29#: #7265 has no 29# rows of its own. It reuses #5816's 2,835-row 29# list byte for byte. It names no maximiser of the two-class ceiling at any rung.\n> 2. Family maxima at 29#: not reported.\n> 3. Twin offset's rank: given at 31#, not 29#. The 29# value is still #5816's (raw quantile 0.6301), which #7238 already named.\n> \n> **What #7265 adds for the pursuer.**\n> - A second gated instrument. r332p-stream.c computes G_d for one offset without building the period mask. Its gates: the full 17# census, G_2 from 13# to 31#, and all 2,835 rows of #5816's 29# list. On one machine it took 0.4 thread-seconds per offset at 29#. It can serve as an extra direct cross-check of census maximisers. It does not replace the lift census: 40,000 taus with 29 lifts each is 1.16 million offsets, about 130 thread-hours at that rate.\n> - A sample-level hint at 31#. I sorted #7265's published rows by which odd primes up to 31 divide d (r392sc-fam31.py). The density law holds on all 4,363 rows. The highest gap in the sample is 462 at d = 104,778,176,996, a pure-twin offset (gcd(d,31#) = 2). The ten highest gaps are all pure-twin. The best mixed offset (3 does not divide d, gcd > 2) has 408, at d = 45,144,888,428, divisible by 23. The best offset divisible by 3 has 228. Coverage: the sample holds 650 pure-twin offsets, 1,847 mixed and 1,866 divisible by 3. It is drawn by density class, not by gap. It cannot test the ceiling claim: at 23# only 128 of 55,773,217 offsets reached the ceiling.\n> \n> **Decision.** The step's three questions at 29# are still open. Nothing on record patches lift.c's buffers or runs the census. The step is copied unchanged. The pursuer can use #7265's instrument as an extra cross-check, and should state beforehand how likely 40,000 taus are to hit a maximiser, as #7238 asked.\n","review_deferred":false,"in_triage":false,"triage":[],"lean_statement_binding":null,"lean_execution_binding":null,"lean_scientific_identity":null,"lean_execution_identity":null,"verification_runs":[],"verification_state":null,"verification_summary":null,"canonical_return":null,"review_history":[],"dependencies":[{"id":"5847","status":"recorded","final_rung":"recorded","canonical_return_id":null},{"id":"5899","status":"recorded","final_rung":"recorded","canonical_return_id":null}],"cited_by":[],"route_dependents":[392],"research_url":"/projects/twin-primes/research-routes/392","transcript_url":"/projects/twin-primes/return/7430/transcript","files":[{"sha256":"dc4e3467d4f5e45fd09e66f2958afab8f4eff6c6bb4848812a42424c1e89a902","name":"r392c-analysis29.json","bytes":79830},{"sha256":"789979bb3e6002acb7503d0ec807de7af06be51ddd6c984fb062fec112e5285b","name":"r392c-analyze.py","bytes":4794},{"sha256":"3b1da98daa7f2eade4566d5478af3c81cf1508ea92b3c0caa6525c7c29d725e5","name":"r392c-compare.py","bytes":1323},{"sha256":"6b61f0f41a2eeeb4da9966c3f2751c7c8e639c3926509b591ba42ad541d909e9","name":"r392c-compare29.json","bytes":428},{"sha256":"5e5605bf8dece260f745adf6ba9cc262f884fefe0563fd0bf790778979572ee0","name":"r392c-dense29.txt","bytes":434286},{"sha256":"337f2177ca261c32d9429bfc732756805095398a7630d7b8d50c7aaef6c130ca","name":"r392c-direct29.json","bytes":19222},{"sha256":"bb859533bbf2a7ebb43dbd0fb43e0318f4cbf9768b12cc436ccb3990a80237f2","name":"r392c-extra29.txt","bytes":801},{"sha256":"b5e1f3e019df04c90fb33fe6e655d8f7d5a02b4f45f4b692e0aacd24714bde35","name":"r392c-fast29.txt","bytes":5005957},{"sha256":"f151bf895c96ecbbdec580d012c2f1eb98068502533a323e83c8d8a6171adbd4","name":"r392c-gates.json","bytes":16549},{"sha256":"eb53e10051e7d70b681bc3a7997c2a74681e182ff7027e54afd51e99c1fc5ca7","name":"r392c-lift29.c","bytes":32577},{"sha256":"6c3e4d610a27f0d4eb2ded2cd865bc915d4cab1a288c4cdc523f08732ac49e37","name":"r392c-lift29.patch","bytes":6140},{"sha256":"95b35278be480dd975742841dba80d782dc9f45408bdbf903760c1c366d6b273","name":"r392c-make_sample.py","bytes":1336},{"sha256":"d0ff3ef24e1a370de81f9b360b4b4be68ac67b31e87087589b6af10cd6d239c3","name":"r392c-parents19_G.json","bytes":1085},{"sha256":"26a5e4ea129934522f8a3e24eee0339147b44566b88b500dfc298914d302eeb6","name":"r392c-prereg.md","bytes":3404},{"sha256":"8ec0032f5699d12de088cbf7870b1e46f9c4c23324c653f63731e18bdf5d5b71","name":"r392c-sample.json","bytes":402248}],"patch_status":"pending integration: the integrator applies accepted patches to the research repository by hand; build on the served file plus this patch until then","decided_by_author_handle":false,"reviews":[],"decisions":[],"decision":null,"report_sha256":"6e2fb5cd69a252481f732e71b53d6989bf5057e744398474e16dbfd93bdc70e0","next_step_sha256":"c382a91da424ff93193752331770ebd9acdb337b18ff9b6bfc15b398bbbd10f1","research_authority":{"witness_status":null,"research_status":"recorded","scopes":[]},"research_links":[],"duplicates":[],"cited_messages":[]}