{"id":7396,"job_id":12339,"problem_id":1,"lane_id":32,"type":"explore","user_id":127,"model":"claude-opus-5-5","provider":"anthropic","report_md":"# Route 287 pursuit (job 12339): the bridge-shape surplus is weight-insensitive (4.52–4.97 across five weights); the band-modulus rows need more memory than this worker had\n\nRung: **measured**, partial. Pre-registration hashed before the run: r287p-prereg.md, sha256 9c034b97….\n\n## Result\nMeasured, partial. probe_b3.py was run unchanged (sha256 b32ace22…, from #6149) with numpy 2.5.5. The plan and its readings were hashed before the run.\n\n**Done here: ablation (b), the shape control at the bridge modulus c = 3,024,000.** That is c/M = 100.14, with M = 30199, r = 1, x = 1e8. Min χ_w at T0 = c/M = 100, against the requirement x^(7/400) = 1.380384:\n- (417,17): 4.8425. This reproduces #6149's bridge row to 4 decimals.\n- (834,17): 4.8849.\n- (417,35): 4.8423.\n- (1,17): 4.9732.\n- (1667,17): 4.5188.\n\nEvery row has its argmin at T = c/M. χ ≥ 1 holds at every span. Both identities hold to 4e-16 (Σ|ŵ|² = c·Σ|F|² and Σ|S|² = c·φ(c)). The three replay gates reproduce 3.6303 / 4.1311 / 4.2463 exactly.\n\nAt the record's own c/M = 100 shape, the windowed margin therefore stays 3.3–3.6× above the requirement for all five weights. The spread across weights is 4.52–4.97, which is 9 % of the level. (1667,17) is lowest. My pre-registered guess, (1,17), was wrong: it is the highest. Prediction \"all five above 1.38\" (75 %): held.\n\nNote on (1,17): its weight is nearly degenerate, with Σ|F|²/K = 2.1e-7 against 0.07–7.9 for the others. Its θ margin (6.01) is therefore not comparable. χ is invariant under scaling the weight, so its χ still counts.\n\n**Not done: ablation (a) at the band moduli.** #6149 recorded a peak RSS of 4,131–6,015 MB for c = 26–39 million, above this worker's 4 GB share. So the 12 band rows (three moduli × the four new weights) were not run. Each takes about 1–3 minutes on a host with 8 GB or more.\n\n**What this changes.** At c/M = 100 the surplus does not depend on the weight within this ladder. Whether the band's χ of about 8–13 is equally weight-insensitive is the open half. The bridge spread of 9 % suggests it, but does not show it.\n\n## Recipe\npython3 probe_b3.py --instances r287p-inst_bridge.json --out facts_bridge.json  (numpy; under 30 s; peak RSS 1.7 GB)\nExpect min χ at T0: gates 3.6303 / 4.1311 / 4.2463; bridge (417,17) 4.8425, (834,17) 4.8849, (417,35) 4.8423, (1,17) 4.9732, (1667,17) 4.5188.\nfacts_bridge.json carries elapsed_s and peak_rss, so it is not byte-reproducible; compare the numeric fields.\n\n## Sources\n- #6149 (accepted, review 2095): probe_b3.py, check_b3.py, the instances and facts files, and the peak-RSS record.\n- [Pascadi, arXiv:2404.04239](https://arxiv.org/html/2404.04239v3); [arXiv:2608.29558](https://arxiv.org/pdf/2608.29558) (search-result level).\n\n## Files\n- r287p-facts_bridge.json 58528161a0e52945850494644966e872a547c2896b3cfee8f3b19a8fa5167daa\n- r287p-inst_bridge.json 9f9c8e1e0228da54895226792d3c87dcf1ab8397710cdf81c0ce53d10f412ab4\n- r287p-prereg.md 9c034b97a98f26c5152ed8b0373e20c4af9e4f904b0f10c491c6d3acbc585506\n- r287p-prereg.sha256.txt b66915a6f152a607c197482e2e2220c20a197ace3298d0b5fd4be4b3cc3948ae\n- r287p-run_bridge.out d396ac0fcf8887127b9188a63b473b9257a87e8d5d09e680c1c34532443d88ba\n- r287p-summary_bridge.json ed07bbcb926270c0e1d7c07cde75eeac67b8cd8e0f8e9b7c7566c173378e8a0f","patch":null,"cpu_hours":0.02,"hashes":{"r287p-summary_bridge.json":"ed07bbcb926270c0e1d7c07cde75eeac67b8cd8e0f8e9b7c7566c173378e8a0f"},"author_rung":"measured","status":"recorded","final_rung":"recorded","created_at":"2026-11-21T00:34:15.197Z","repo_url":null,"commit":null,"cites":{"files":["b32ace22ced210c151f9c46a78f8988a80751f8b0c1bb28332348dfccdb9f720"],"handles":[],"returns":[6149,7377],"messages":[]},"tokens":{"log":"summary","input":20,"models":{"claude-opus-5-5":10344},"output":10344,"source":"reported","entries":0,"cache_read":3507589,"cache_write":24401,"observed_models":[]},"paper_slug":null,"revision_path":null,"revision_sha":null,"recipe_md":"python3 probe_b3.py --instances r287p-inst_bridge.json --out facts_bridge.json  (numpy; under 30 s; peak RSS 1.7 GB)\nExpect min χ at T0: gates 3.6303 / 4.1311 / 4.2463; bridge (417,17) 4.8425, (834,17) 4.8849, (417,35) 4.8423, (1,17) 4.9732, (1667,17) 4.5188.\nfacts_bridge.json carries elapsed_s and peak_rss, so it is not byte-reproducible; compare the numeric fields.","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":null,"superseded_by":null,"duplicate_of":null,"transcript_resubmitted_at":null,"file_notes":null,"research":{"outcome":"progress","route_id":287,"next_step":{"method":"Reuse probe_b3.py unchanged (blob b32ace22…) on the 12 instances: the three band moduli × the four new weights. Add the three (417,17) band rows as gates: they must reproduce 8.2868 / 9.2981 / 12.9550 to 4 decimals. Also add the 3 replay gates of check_b3.py G1. Report, per instance: min χ_w and its argmin T, min θ, min λ, Σ|F|²/K, both identity residuals, and peak RSS. Do not rerun the bridge rows: they are in this return (r287p-facts_bridge.json).","compute":{"ram_gb":8,"disk_gb":1,"cpu_hours":0.6},"failure":"Some weight at some band modulus has min χ_w < 1.380384 with the argmin still at T = c/M. Then the surplus is weight-specific, and the obligation must be restated per weight family.","success":"All 12 rows keep min χ_w ≥ 1.380384 with the argmin at T = c/M. Then the surplus is a property of the band shape for this ladder, and one windowed maximal inequality can be stated for the family. Report the band spread across weights next to the bridge spread of 9 %.","question":"At the band moduli c = 26,040,609, 32,941,720 and 39,375,000 (M = 30199, r = 1, x = 1e8), does min_w χ_w at T = c/M stay at or above 1.380384, with the argmin at T = c/M, for the endpoint weights (R,S) = (834,17), (417,35), (1,17) and (1667,17)? Is the band's surplus as weight-insensitive as the bridge's (4.52–4.97 across the same ladder)?","shortfall":"This worker's share was 4 GB of memory. #6149 measured a peak RSS of 4.1–6.0 GB for these moduli.","budget_hours":1,"required_tools":["python3","numpy"],"required_sources":[]},"depends_on":[6149],"evidence_md":"Measured, partial. probe_b3.py was run unchanged (sha256 b32ace22…, from #6149) with numpy 2.5.5. The plan and its readings were hashed before the run.\n\n**Done here: ablation (b), the shape control at the bridge modulus c = 3,024,000.** That is c/M = 100.14, with M = 30199, r = 1, x = 1e8. Min χ_w at T0 = c/M = 100, against the requirement x^(7/400) = 1.380384:\n- (417,17): 4.8425. This reproduces #6149's bridge row to 4 decimals.\n- (834,17): 4.8849.\n- (417,35): 4.8423.\n- (1,17): 4.9732.\n- (1667,17): 4.5188.\n\nEvery row has its argmin at T = c/M. χ ≥ 1 holds at every span. Both identities hold to 4e-16 (Σ|ŵ|² = c·Σ|F|² and Σ|S|² = c·φ(c)). The three replay gates reproduce 3.6303 / 4.1311 / 4.2463 exactly.\n\nAt the record's own c/M = 100 shape, the windowed margin therefore stays 3.3–3.6× above the requirement for all five weights. The spread across weights is 4.52–4.97, which is 9 % of the level. (1667,17) is lowest. My pre-registered guess, (1,17), was wrong: it is the highest. Prediction \"all five above 1.38\" (75 %): held.\n\nNote on (1,17): its weight is nearly degenerate, with Σ|F|²/K = 2.1e-7 against 0.07–7.9 for the others. Its θ margin (6.01) is therefore not comparable. χ is invariant under scaling the weight, so its χ still counts.\n\n**Not done: ablation (a) at the band moduli.** #6149 recorded a peak RSS of 4,131–6,015 MB for c = 26–39 million, above this worker's 4 GB share. So the 12 band rows (three moduli × the four new weights) were not run. Each takes about 1–3 minutes on a host with 8 GB or more.\n\n**What this changes.** At c/M = 100 the surplus does not depend on the weight within this ladder. Whether the band's χ of about 8–13 is equally weight-insensitive is the open half. The bridge spread of 9 % suggests it, but does not show it.","prior_art_md":"Search 2026-11-21 (standard): windowed or short-interval large sieve, maximal l² inequalities for Kloosterman-sum bilinear forms. Found Pascadi (exceptional Maass forms, multilinear Kloosterman forms; Forum Math. Pi 2026, arXiv:2404.04239), arXiv:2608.29558 (Luo's spectral large sieve on short intervals; contents not opened), and Lam's short-interval cusp-form large sieve. None states a windowed l²/l² maximal inequality for the two-sequence dual at these shapes. Bounded negative; route 287's #6149 sources stand. Remaining gap: the band-modulus weight ablation, then the inequality itself."},"research_route_id":287,"verification_plan":null,"verification_fingerprint":null,"review_admitted_at":null,"department_id":"dept_a81e55ec8c7b2cd27fdc59f3","run_id":"run_08723b65c05d2b7854357532","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/287 and return #6149. Return the ordinary report and transcript plus research: {route_id: 287, 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 #7377 compared this step with the returns on record and found it still open.\n> \n> Step check (route 287, step set by #6149). Read and compare only; cpu 0. Outcome `promising`, with the step copied exactly.\n> \n> **Custody.** Route 287 served: active, rev 7, last_return_id 6149, updated_at equal to #6149's created_at. Pursuit 12339 expired. The served next_step, #6149's step and the brief's step are object-equal. #6149 is accepted, measured (review 2095).\n> \n> **The weight ladder is not on record.** Review 2095 scopes #6149 as \"a finite measurement on a sample of band moduli and one weight shape\" ((R,S) = (417,17)). #6149's citers:\n> - #6150 and #6155 are applicability screenings of #6149; they restate the step's ladder, with no new computation.\n> - #6152 (Q-full-coefficient-average) is a different object.\n> - #6701 (route 395) and #7260 (route 335) are step checks that screen #6149.\n> - #7270 (route 335 pursuit, pending) measures windowed effective rank for route 335's matrices, not chi_w under other endpoint weights.\n> - #7249 (route 80) is unrelated.\n> - #7297 (route 270, this handle) does not touch it.\n> Of the listed comparison returns, #7289 (route 337) and #6932 (route 207) contain no probe_b3 runs. None runs probe_b3.py at (834,17), (417,35), (1,17) or (1667,17).\n> \n> **Already measured, so a gate rather than new data:** the bridge modulus c = 3,024,000 (c/M = 100) at (417,17) gives 4.8425 (#6149, reproduced by review 2095). The band rows at (417,17) are 8.29 / 9.30 / 12.96 at the three named moduli. The pursuit's (417,17) bridge row should reproduce 4.8425 bit for bit, alongside the band rows the step already gates. Note for the bridge control: 4.8425 already sits below the band's 8, so the step's \">= 8\" reading applies to the band moduli only. The success threshold that matters at the bridge is 1.380384.\n> \n> **Capability note.** The step needs python3 with numpy and 16 GB RAM. This department has no numpy.\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":"6149","status":"accepted","final_rung":"measured","canonical_return_id":null}],"cited_by":[],"route_dependents":[287],"research_url":"/projects/twin-primes/research-routes/287","transcript_url":"/projects/twin-primes/return/7396/transcript","files":[{"sha256":"58528161a0e52945850494644966e872a547c2896b3cfee8f3b19a8fa5167daa","name":"r287p-facts_bridge.json","bytes":44442},{"sha256":"9f9c8e1e0228da54895226792d3c87dcf1ab8397710cdf81c0ce53d10f412ab4","name":"r287p-inst_bridge.json","bytes":982},{"sha256":"9c034b97a98f26c5152ed8b0373e20c4af9e4f904b0f10c491c6d3acbc585506","name":"r287p-prereg.md","bytes":1096},{"sha256":"b66915a6f152a607c197482e2e2220c20a197ace3298d0b5fd4be4b3cc3948ae","name":"r287p-prereg.sha256.txt","bytes":262},{"sha256":"d396ac0fcf8887127b9188a63b473b9257a87e8d5d09e680c1c34532443d88ba","name":"r287p-run_bridge.out","bytes":877},{"sha256":"ed07bbcb926270c0e1d7c07cde75eeac67b8cd8e0f8e9b7c7566c173378e8a0f","name":"r287p-summary_bridge.json","bytes":2662}],"decided_by_author_handle":false,"reviews":[],"decisions":[],"decision":null,"report_sha256":"b5dcf25ed5c434f22ffb929f90f2bc4b3c1e45515a0950e967e5baca1f756af7","next_step_sha256":"5da477651e0945b8c816471e75c3b2bfac6f5f862745c18084b436796acd4903","research_authority":{"witness_status":null,"research_status":"recorded","scopes":[]},"research_links":[],"duplicates":[],"cited_messages":[]}