# Merge 02 — paired re-decision results ## Result On the supplied paired measurements, `cut-s2-on` is the only clearly positive one-entry change. Its gain is **+0.891348 ± 0.235613** on spread 2 (482 spread-2 seeds) and **+0.214815 ± 0.057375** over all 2,000 seeds. The other four changes are negative or too uncertain to support changing the incumbent setting. Thus the evidence-supported combined line from this merge is: ```python ON = dict(cancel=(2, 3, 4), cut=(1, 2, 4), postjump=(1,), premium=(), deadband=(3,)) ``` This is a recommendation from one-entry measurements, not a measurement of the combined file. In particular, it does not establish that independently measured flips compose without interaction. `d` below means `CANDIDATE.total_edge - T.total_edge`; positive is better. Each aggregate is the sum of the supplied unrounded block aggregates, with each reported seed having equal weight. `SE` is the sample standard error, `sqrt((sum_d2 - sum_d^2/n) / (n*(n-1)))`. `C/T` is the seed-count-weighted candidate mean runtime divided by T's; it is only a paired timing comparison within these records, whose timing scopes differ slightly by item. | Flip (changed spread) | Overall n | Overall sum_d / sum_d2 | Overall mean ± SE | Spread n | Spread sum_d / sum_d2 | Gain on changed spread ± SE | Worst single run (seed, spread, d) | C/T | Sources | |---|---:|---:|---:|---:|---:|---:|---|---:|---| | `cut-s2-on` (2) | 2,000 | 429.629583 / 13,253.333038 | +0.214815 ± 0.057375 | 482 | 429.629583 / 13,253.333038 | **+0.891348 ± 0.235613** | 1000194, 2, -38.609181 | 1.000339 | C2 | | `cut-s3-on` (3) | 2,000 | 85.195897 / 75,169.104831 | +0.042598 ± 0.137116 | 538 | 85.195897 / 75,169.104831 | +0.158357 ± 0.510038 | 1000147, 3, -37.940630 | 1.004538 | C3 | | `cut-s4-off` (4) | 2,000 | -124.533835 / 11,273.248905 | -0.062267 ± 0.053083 | 447 | -124.533835 / 11,273.248905 | **-0.278599 ± 0.237429** | 1001420, 4, -27.304220 | 0.996613 | C4 | | `postjump-s1-off` (1) | 1,723† | -12.335137 / 891.124579 | -0.007159 ± 0.017330 | 457† | -12.335137 / 891.124579 | -0.026992 ± 0.065380 | 1000087, 1, -8.336815 | 0.999173 | C1 | | `postjump-s2-on` (2) | 2,000 | -41.176189 / 334.325073 | -0.020588 ± 0.009133 | 482 | -41.176189 / 334.325073 | **-0.085428 ± 0.037774** | 1001915, 2, -11.089412 | 1.001045 | C5 | Because every flip touches only its named spread in the supplied records, the overall and changed-spread sums are identical; their means and SEs differ because the denominators differ. The `cut-s3-on` row includes a +249.584278 single-run result (seed 1000913, spread 3), which accounts for much of its large second moment and uncertainty. [C3] † `postjump-s1-off-4` is marked `complete: false` with only 223 reported seeds, so this pool is 277 seeds short of the planned 2,000. It is included at its actual reported seed count, not filled or extrapolated. Consequently the `postjump-s1-off` non-result is less complete than the other four rows. [C1] ## Integrity and coverage The supplied 20 item artifacts are all present, their byte SHA-256 values match the values supplied with this merge, their duplicate `artifacts/goal/items/.../result.json` copies match as well, and each record's `per_seed` count matches its reported `n` and lies within its stated seed range. Four complete blocks cover 1,000,000–1,001,999 for C2–C5. C1 has three complete 500-seed blocks plus the incomplete 223-seed fourth block. The T code was read from . Its `code` SHA-256 is `9fba1eb5659d5b0ac9b7273a968a520ef9a95551c697be22020993005af70192`, and it has exactly this one ON line: ```python ON = dict(cancel=(2, 3, 4), cut=(1, 4), postjump=(1,), premium=(), deadband=(3,)) ``` Replacing that sole line produced the following verified candidate hashes: | Flip | Candidate ON change | SHA-256 | |---|---|---| | `cut-s2-on` | `cut=(1, 2, 4)` | `e1dd5b9bb7d4434c95c92d6879c7edb3a05e33b442959c0dc2b536b033c28c6e` | | `cut-s3-on` | `cut=(1, 3, 4)` | `1a130fe5521b7cf103aab0c5c27778dc50ab4d5bcc3c937d5ca2b818cb1df7f8` | | `cut-s4-off` | `cut=(1,)` | `8fc15ca7148427ccb5f49ec5cf7ab2a8d34e0c594306969524792aaa0a448d34` | | `postjump-s1-off` | `postjump=()` | `2c6a11153bd7fe78483ce92f832c27fb9c727e210223fe22c9549c18c1306bac` | | `postjump-s2-on` | `postjump=(1, 2)` | `cddf479aea89aa7fbc24c954c4e48e83597187d722fa45fbd33025b3f8cbbe9a` | ## Source qualification The numerical rows above are calculations from the supplied local item artifacts (the source registry follows); no external page is used as support for a numerical gain. The task reports that one citation in each of these artifacts did not hold on later review: `cut-s2-on-1`, `cut-s2-on-3`, `cut-s4-off-1`, `cut-s4-off-2`, `postjump-s1-off-3`, and `postjump-s2-on-1` through `-3`. I therefore do **not** treat their cited external-source claims as independently verified provenance. The supplied JSON does not identify the failed page or quote, so that source cannot be rechecked or repaired here. To make the effect visible rather than silently discounting data, the changed-spread aggregates split by that qualification are below. They are descriptive subsets, not a reweighting rule: no defensible numerical weight was supplied for a failed citation. | Flip | Citation-flagged input subset: n, gain ± SE | Other supplied inputs: n, gain ± SE | Implication | |---|---:|---:|---| | `cut-s2-on` | 227, +0.380974 ± 0.274878 | 255, +1.345680 ± 0.370320 | Both subsets are positive; the recommendation does not depend on the flagged subset. | | `cut-s3-on` | none | 538, +0.158357 ± 0.510038 | No affected item; still too variable to recommend. | | `cut-s4-off` | 228, +0.032554 ± 0.321645 | 219, -0.602540 ± 0.349721 | The negative full-pool conclusion is driven by the non-flagged half, but remains imprecise. | | `postjump-s1-off` | 129, -0.108409 ± 0.088752 | 328, +0.005030 ± 0.084152 | Incomplete fourth block and mixed subsets support retaining T, not a change. | | `postjump-s2-on` | 349, -0.072295 ± 0.029982 | 133, -0.119889 ± 0.112302 | Both subsets are negative. | ## Source registry `C1`–`C5` refer to these supplied JSON measurement artifacts; the listed SHA-256 is the byte hash verified during this merge. Each contains the per-seed records, block aggregate, extrema, and timing used for its row. | ID | Supplied artifacts (SHA-256) | |---|---| | C1 | `artifacts/goal/items/postjump-s1-off-1/result.json` (`831d677df0166b422ba4df7e224232d4f6afcca47a3cef8d11cf881da6afba0c`); `artifacts/goal/items/postjump-s1-off-2/result.json` (`8c562066ae1fc3e26155e2b6d342257d2fa341a581e8642e40480de5ae4ae450`); `artifacts/goal/items/postjump-s1-off-3/result.json` (`2a6662122e913b54c19468dec9a454d8bbe9cc5397d31c6085547b2f5f9bbb58`); `artifacts/goal/items/postjump-s1-off-4/result.json` (`d38ce6d8815117edee6305eade1881e67381be562394fb27ae58330b777cf832`) | | C2 | `artifacts/goal/items/cut-s2-on-1/result.json` (`6fdf2ad95b7b94487cfa5c6794742d03cc20a3ab7c7e2e69fe9690d09e414182`); `artifacts/goal/items/cut-s2-on-2/result.json` (`a3f531efb932a279e3fe829897cde64379cab302ab50903ec264c3a113982051`); `artifacts/goal/items/cut-s2-on-3/result.json` (`e66eb345e16c1c9a69e3b699311f7a345137cd2ae4a742ca78581db81cce97f6`); `artifacts/goal/items/cut-s2-on-4/result.json` (`9c193968133805daeccc1d9e5ea3d4e41b76bfc89df34af4fb021b0f4720a200`) | | C3 | `artifacts/goal/items/cut-s3-on-1/result.json` (`a9a7cca7ace97b370b4d800769a349d5bbfc9c902814b4e69867ef75f2fd2eb3`); `artifacts/goal/items/cut-s3-on-2/result.json` (`96ca484c018228ba26a5bffaab67d76f4e8805d6bea97d32661ca25461c7d425`); `artifacts/goal/items/cut-s3-on-3/result.json` (`165bc7e5a7321c453d5086780466659ed5949e6ccac01a9f7fa502089d99ba45`); `artifacts/goal/items/cut-s3-on-4/result.json` (`e86226f5473657e732db25cb0cb45f7d0e9d5c1edf782aaa5e92afcc90213227`) | | C4 | `artifacts/goal/items/cut-s4-off-1/result.json` (`4ee317d6197259c74caca2e912c9d19bd56df9723b7f46965b0e5f5d3e84d093`); `artifacts/goal/items/cut-s4-off-2/result.json` (`bb7d5a02e37bbd19453ef097fb96cf809bc764d7e6d5dbf8f2f3e1730d4f5c19`); `artifacts/goal/items/cut-s4-off-3/result.json` (`82f5c501429976ae9406e8bf6eddbbfecbd5f2d18ccf64f132880e1d1eb5f09a`); `artifacts/goal/items/cut-s4-off-4/result.json` (`73e16c5ce2c3c1dee96b87e9faa7c7eeb843155c9839a6ad9cf06281049d0cd4`) | | C5 | `artifacts/goal/items/postjump-s2-on-1/result.json` (`6bc257071033c0757a0f67cb1850aa71d5c1acff90e65b4edec336704e95b216`); `artifacts/goal/items/postjump-s2-on-2/result.json` (`bfc7ce82f1ae7a4c323769a47dfd4daa93d37594b5ed720cb1fbc5093c1b1763`); `artifacts/goal/items/postjump-s2-on-3/result.json` (`503f58a53349573464569f35171a3f2e0be2928daf177ca765048321cc9fda63`); `artifacts/goal/items/postjump-s2-on-4/result.json` (`3826a7d1deac890b0f30b1f5bdba68c0cd6cae672f528ab5673173944e6a4c3a`) | The simulator provenance reported by the item records is .