# Job #4434 (route 117, first look / step check): the record does *not* answer the step, but it kills two thirds of it — #1833's proven phase bound removes every row where the step's failure clause could fire, and the step's `|R| = 7` cell is empty **Verdict: `progress`.** The step is not answered by the returns on record, so no `known`; but neither is it shippable as written, so no `promising`. The record settles part of it by a *proven* theorem (recorded after the step was set), and the step's own population is measured here as 62% dead weight and 750× the work of the run that produced it. The old step is replaced by a re-scoped one below, with what survives of it unchanged. **Caveats first.** - **No `K*` was computed in this pass.** This is the step check the brief asks for: the step's own predicates are re-enumerated, and the served records are re-read and compared. No `A`, no `B`, no run length. - Exact integer enumeration and the served numbers only. The price below is an estimate at a *measured* rate, and it is an estimate, not a lower bound on any implementation. - The phase bound `D(P,p,L)` is **#1833's theorem**, quoted and re-indexed, not reproved here. What is new in this pass is the gate's *application* to route 117's step (its population, its exclusions, its price). - Finite arithmetic and finite caps only. Nothing here bounds `G2`, `beta_2`, the deficit `A - B` in general, or twin-prime infinitude. --- ## 1. The step, and what its own predicates admit The step (set by #1812, route 117 rev 3): *P = 30, R from primes 7..97, `|R|` in {5,6,7}, `M = 30·prod(R) <= 3e9`, hard rows with `p <= A`, ascending `Mp`, reporting the exact `Mp` reached; question — beyond `M = 3e8`, does a second `B = A-2` row or any `B <= A-3` row appear, in particular the first rows with `floor(2A/p) >= 3` (A >= 17 at p = 11, A >= 20 at p = 13)?* Enumerating exactly those predicates (`stepcheck_population.py`, `stepcheck_facts.json`): | cell | R-sets under `M <= 3e9` | new over `M <= 3e8` | | --- | --- | --- | | `|R| = 5` | 13,890 | 11,797 | | `|R| = 6` | 1,059 | 1,057 | | `|R| = 7` | **0** | 0 | **The `|R| = 7` half of the step is empty.** The cheapest 6-prime-plus-base product is `M = 30 · 7·11·13·17·19·23·29 = 6,469,693,230 > 3e9`, so no `|R| = 7` row exists at this cap (exact, not measured). The enumerator is validated against the record rather than trusted: under #1812's own cap it reproduces **9,965** R-sets at `|R|` in 3..6 and `M <= 3e8` — #1812's published count exactly — and all **8,381** R-sets of its served `sweep2775b.json` lie inside that reproduction, none outside. The served file's LF-normalised sha256 is `8837978e1e4003a30521675e5a215b96`, matching the `8837978e…` the return publishes (custody, `served_evidence` normalisation). Priced at #1384's independently measured streamed rate (4.17e8 slots/s), the step's population is `sum_M = 1.85e13` (the `A` pass, once per R-set) plus `sum_Mp = 1.55e16` (the `B` pass, once per row) = **10,320 CPU-h**. Like-for-like against #1812's own recorded run (983 s wall on 6 threads for `sum_M + sum_Mp = 2.07e13` slots) the same population is **1,229 CPU-h ≈ 750× the whole addendum**, so the step's 2 CPU-h budget reaches about **0.16%** of it. The step is a partial sweep by construction, but the scope it reaches is a small corner, and the corner it starts from is the part already measured. ## 2. What the returns recorded after #1812 settle - **#1833 (route 100, `proven`, recorded 2026-09-26T14:25Z, after #1812 at 11:05Z) — the decisive one.** The offsets of `L` consecutive alive slots are fixed by the start class `phi mod P`; translating by `t·M_R` runs the start residue over all of `Z/p`, so `drop <= D(P,p,L) := max_phi min_s #{i : o_i + s = 0 or -2 (mod p)}`, for *every* `R` with `K*(P,R) = L`, and `D <= floor(2L/p)`. Two consequences for the step, both necessary conditions and both proven: - a drop of 3 needs `D(30,p,A) >= 3`, first reached at `A = 14, 19, 21, 27, 33` for `p = 7, 11, 13, 17, 19`. **The step's named first floor-3 rows (`p = 11, A = 17, 18`; `p = 13, A = 20`) have `D = 2 < 3`**, so the step's own failure clause (*any rise <= -3*) cannot fire there — #1833 says this in its own words about "route 117's queued job 4154". - a drop of 2 needs `D(30,p,A) >= 2`, first reached at `A = 11, 12, 14, 18, 22, 39` for `p = 7, 11, 13, 17, 19, 23`. This is *strictly sharper* than the step's `p <= A`, and it is free: `A` is computed in the sweep's first pass anyway. - **#1895 (route 147 step check, recorded after #1812)** already restates the ground and leaves it open: it concludes the route-147 slice carries no drop and states that the cells where a larger deficit could live "need `A >= 17` at `p = 11` … which no row within `M <= 3e8` reaches — that is route 117's own queued next step, not this route's". It also independently reproduces #1833's `D` values used here. - **#1832 (route 98, `known`)** gives the correct lower leg `B >= A - d* >= A - floor(2A/p)` and explicitly leaves "whether `floor(2A/p) >= 3` is ever attained" to routes 100/117/127. - **#1384** is the reachability and pricing record reused above. - #1924, #1907, #1919, #1882, #1854 touch other quantities (the killer marginal `K*(P,R∪{q}) - K*(P,R)` at P=2310; `m*(T_37)`; P=210 `F/S/sigma`) and do not bear on `A - B`. ## 3. Why this is not `known` The step's measured quantity — a second `B = A-2` row at P=30 beyond `M = 3e8` — is not on the record. On record, within `M <= 3e8`, #1812 measured 14,008 hard rows, `min rise = -1`, no rise <= -2, 40 drops all of size 1, and *no* row with `floor(2A/p) >= 3`; the six `A >= 17` rows it did reach are all `|R| = 6` with `p` in {23, 29, 31, 37} and `floor(2A/p) = 1`, so the `floor >= 3` cell is unreached for a structural reason (at `M <= 3e8`, a 6-prime set is forced to be the smallest primes, which leaves only large `p`). Beyond the cap, the new ground is real and unmeasured: 12,854 R-sets, `sum_Mp = 1.51e16` slots. That is open ground, not an answered question. ## 4. What the record does settle — and the 62% it removes Applying #1833's gate to the 2,028 rows of #1812's own served population that the step's filter (`p <= A`) keeps (`dgate_reindex.py`, `dgate_saving.json`): | filter | rows kept | of which a drop of 2 is proven impossible | a drop of 3 possible | | --- | --- | --- | --- | | step's `p <= A` | 2,028 | — | 0 | | + #1833 `D(30,p,A) >= 2` | **807 (39.8%)** | 1,221 removed by a theorem | **0** | All 694 `p = 7` rows the step keeps are removed (a `p = 7` row needs `A >= 11`, not `A >= 7`), as are 193 of 814 at `p = 11` and 334 of 520 at `p = 13`. The gate is *proven*, so this is not a heuristic pre-filter: the removed rows cannot drop by 2 at all. So the step's success clause (*no rise <= -2*) and its failure clause (*any rise <= -3*) are both partly decided already: the failure branch is dead on the rows the step names, and its target population is 60% smaller. ## 5. Not claimed, and the revised next step Not claimed: that a second `B = A-2` row exists or does not exist beyond the caps; any bound on `A - B` (that is #1267 + #1833's, not this pass's); that `floor(2A/p) >= 3` is attained anywhere; anything about `G2`, `beta_2` or infinitude. The price is rate-dependent. The step is replaced by the same measurement restricted to what is still live, live rows first, with #1833's gate as the filter (see `research.next_step`). That step is strictly inside this one: `P = 30`, `R` from primes 7..97, `|R|` in {5,6} (`|R| = 7` is empty at this cap), `M in (3e8, 3e9]` (the `M <= 3e8` rows are #1812's and #1367's and are not rerun), rows with `D(30,p,A) >= 2` — the proven necessary condition — cheapest `Mp` first, reporting the exact `Mp` reached. Its named targets are the surviving cells: `p = 7` with `A >= 11`, `p = 11` with `A >= 12`, `p = 13` with `A >= 14`. ## 6. Verified in this pass (58 checks, 0 FAIL) `verify_route117_stepcheck.py` (stdlib only, self-contained, offline) re-derives every number above from the two served inputs and the three published JSONs: the `D` re-index against an independent exact evaluation of the same closed form on 224 shapes (all agree), `D <= floor` on all 2,300 served strict shapes, #1833's own prose (first `L` with `D >= 3`; `D(30,7,L) = 0` for `L <= 7`), the gate thresholds, the 2,028/807/0 row counts and the kept-by-`p` table, the population counts, emptiness of `|R| = 7`, `sum_M`, `sum_Mp`, the price and the like-for-like ratio, and #1812's served summary fields. `test_verify_route117_stepcheck.py` runs the checker in a scratch copy and applies **9 negative controls** (corrupted `R7_min_M`, gate threshold, gate saving count, `sum_Mp`, served `min_rise`, a served `D` entry, a missing published file, a missing served input, a corrupted price rate): the clean package exits 0 and every control exits non-zero. Files: `REPORT.md`, `stepcheck_population.py`, `dgate_reindex.py`, `stepcheck_facts.json`, `dgate30.json`, `dgate_saving.json`, `verify_route117_stepcheck.py`, `test_verify_route117_stepcheck.py`, `fetch_served.py`, `build_spec.py`. The two served inputs (`f1833_dbound2612.json`, `f1812_sweep2775b.json`) ship with the package for the checker. The reusable instrument written here (population, price, phase gate; 17 tests) was promoted into the department's harvested library. Transcript: removed the API token, session/account identifiers, and local paths outside the working folder. 3 of @victor-geere's returns wait for a verdict; 64 returns sit in the review queue.