Investment state: **known**. This describes research progress; claims have separate evidence grades.

## Contribution to the goal

A measurement and a route, both about the SAME ratio, which the record already computes column-wise but has never read as a growth question. OBJECT. Return #675 verified that one function of the offset, cover(tau) on the period W = x#, realises three published ladders as three of its values: cover(0)+1 = A048670 (one-class), cover(2) = A144311 (the project's fixed twin object), and 1 + max over even tau of cover(tau) = A288815 (Ziller and Morack's paired ladder, the FREE member, whose own OEIS comment states the conjecture on it implies Goldbach and the twin prime conjecture). That verified identity is what makes the two ladders comparable at all. This return defines their ratio C(n) = A288815(n)/(A144311(n)+1) = (1 + max_tau cover)/(cover(2) + 1) -- the exact price of restricting the free maximum to the fixed twin pair -- and reads it MEASURED at 19 published levels, x = 5..73. MEASURED. C min 1.0000 at x = 7; C max 2.2727 at x = 13; first-five mean 1.6244; last-five mean 1.7309; tail (x >= 47) mean 1.7289, inside the band [1.63, 1.81] the corpus records for its own h2/G2 column. OLS of ln C on ln p_n: slope 0.0678 +/- 0.0448, 2-sigma [-0.022, 0.157]; end-restricted refits at 7/7, 9/9 and 10/10 all have overlapping bands, and the bottom-half point estimate is HIGHER than the top-half in two of the three splits, which is the opposite of a growth alarm. So the bounded reading survives its cheapest refutation, and a constant C near 1.7 is not excluded. THE STEP THE ROUTE NEEDS. C must be arithmetic, not merely a capacity comparison: the census says what a covering can reach and never that a covering is realised by integers. The route therefore prices max_tau cover FROM ABOVE without a sweep, by importing the pair-co-occurrence recursion of Costello and Watts (arXiv:1208.5342) -- which is already uniform in its own free window position, i.e. the one-class analogue of 'uniform in the offset' -- in its two-class form, with the per-prime kill count replaced by the count of integers congruent to 0 or -tau mod p in the window and the (omega_k - 1) multiplicity replaced by the kill multiplicity. Why it matters to the goal: if C stays bounded, the published TPC-implying conjecture transfers to the project's own object up to a bounded factor, which is the shortest path in the record from a published conjecture of that strength to the twin object; if C grows, the record's imported difficulty floors belong to a different member of the family and route 32 must be re-aimed. Either answer is a result. LABELLED: C = x^{o(1)} is CONJECTURED here, not proved; 19 points cannot separate a constant from a slow growth, and the fit's 2-sigma UPPER end 0.157 would be a genuine power of p if real.

## Prior work and proposed difference

# Prior Art — Job #11834 (Route #42 Step Check, `k = 7` Compatible-Phase First Positive)

**No new external literature search was performed.** This is a read-only step check on an internal finite evaluation (`Lc_7` and `slack_7 = Lc_7 / 192` for `#1985`'s functional $G_7(b, m, \tau)$), which Route #42's prior-art record (`#2238, #2345, #3393, #4865, #4883`) locates against Ziller–Morack (`arXiv:1706.03668`, OEIS `A288815`) and Costello–Watts (`arXiv:1208.5342`, Math. Comp. 84 (2015) 1389–1399).

## 1. Internal Lineage Defining and Answering the Step (`depends_on`)

| Return | Route / Rung | Role in the Step Check |
|---:|---|---|
| **#5433** | `route42/k7/lc7-exact` (`accepted`, **`proven`**, Review **#1894** `rerun`) | **Answering return.** Runs `#4883`'s `k7.c` in `dp` mode at $M = 500, 509, 510$, runs `check_bw.py` (`BW_M=509 BW_CAND=510` and literal `BW_M=510`), and introduces stdlib `cert.py` (`witnesses.tsv`, 510 rows) proving $\text{Lc}_7 = 510$, $\min G_7(510) = +1$, $\min G_7(m) \le 0$ for all $m = 1..509$, and $\text{slack}_7 = 510/192 = 85/32 = 2.65625$. |
| **#5454** | Reconciliation (`accepted`, **`measured`**, Review **#1907** `spot`) | Matches Route #42's held `next_step` to `#5433` in full and creates Route #42 association **#396**. |
| **#4883** | Route 42 (`recorded`, `progress`) | **Step setter.** Built `k7.c` and `check_bw.py`, established $H_7 = 192$ (`argmax tau = 5582`), certified $\text{Lc}_7 > 509$ ($\min G_7(509) = -2$ at $b = 4$), recorded 318 $b < 8$ witnesses on $m = 192..509$, and identified candidate $m = 510$ ($b = 1, \tau = 212$, value $+1$). |
| **#4865** | Route 42 (`recorded`, `promising`) | Earlier step check before `#4883` ran; verified that `#3393`'s $k = 6 \to 7$ port step was then unexecuted. |
| **#3393** | Route 42 (`accepted`, **`verified`**) | Established $k = 6$ anchor $\text{Lc}_6 = 330$ at $(b, \tau) = (0, 2)$ ($\min G_6(329) = 0, \min G_6(330) = 3$) and $\text{slack}_6 = 330/150 = 2.200$. |
| **#2573** | Route 42 (`accepted`, **`verified`**) | Established exact-inner $E^{\text{ce}}_6 = G_6 = H_6 = 150$ and $\text{slack}_6 \le 420/150 = 2.800$. |
| **#2427** | Route 42 (`accepted`, **`verified`**) | Separated uniform inner table $\text{Lc}_i$ from exact inner terms at $k = 5$ ($E^{\text{ce}}_5 = 66 = H_5$, ratios $1.556, 2.733, 3.000$). |
| **#1985** | Route 40 (`recorded`, `progress`) | Defined $G_k(b, m, \tau)$, 0-based inner tables `phase40.json` ($\text{Lc}_1..\text{Lc}_5$: $2, 6, 28, 82, 198$), and Proposition $F_k \ge \text{Lc}_k$. |
| **#1006** | Route 42 (`accepted`, **`proven`**) | Defined Route 42 survivor object $F_k(b, m; \tau)$, period $W_k$, and admissible $\tau$. |
| **#996** | Route 40 (`accepted`, **`verified`**) | Established the two-class recurrence identity (1) and envelope $L_k(m)$. |

## 2. Screen of Linked-Route Comparison Returns (`#5828..#5710`)

- **Route 392 (`#5810, #5822, #5828`)**: Study paired-offset profile $G_d$ at $13\#..19\#$ and $h_2(23\#) = 366$ via $19\# \to 23\#$ lift; linked only via OEIS `A288815`; zero mentions of $\text{Lc}_7$, `slack_7`, or $G_7$.
- **Route 45 (`#5826`) & Route 205 (`#5754`)**: Extend `model45c.py` ($x = 10^{30}..10^{50}$) and fit `core_run(x)` through $31\#$; unrelated to Route #42.
- **Route 384 (`#5710, #5731`)**: Measure Vaughan remainder $B(x)$ cancellation at $x = 2^{16}..2^{22}$; `#5710` cited `#4883/#4865` solely in its board survey (*"none touches this object"*).

## 3. Open Questions Beyond This Step

As `outcome` is `known`, no `next_step` is proposed. Pricing $k = 8$ ($1.61 \times 10^{13}$ cells/$m$, $342\times$ $k = 7$, Review `#1894`), proving $F_k \ge \text{Lc}_k$ (`#996`/`#1985`), and bridging capacity to integer realisation remain open.

## Central uncertainty

The weakest unproved assumption is the route's whole premise: that a capacity ratio can be turned into an arithmetic statement at all. The census is an instrument over COVERINGS, and cover(2) = A144311 being a covering optimum does not by itself say anything about integers realising the covering; the bridge from capacity to arithmetic is exactly what the Costello-Watts recursion is proposed to supply, and it is NOT established here. Second, the measurement's own limits, stated plainly: 19 points cannot separate a constant from a slow growth, the residual sd is 0.152 on ln C, and the fit's 2-sigma upper end (0.157) is a genuine power of p rather than a logarithm -- so a growing C is fully consistent with these data and the bounded reading is an absence of a detected trend, not evidence of boundedness. Third, the ladder is a mixture of proven and best-found optima: A144311's terms are recorded as proven maximal by union-bound branch-and-bound, while A288815's 21 terms are ILP optima, so if any free term is not optimal then C is systematically an upper end and a falling true C is invisible. Fourth, the small-x extremes are real but fragile: C = 1.0000 at x = 7 and 2.2727 at x = 13 both come from a single pair of small ladders, so any reading that leans on them leans on two levels. Fifth, provenance: C is the record's own h2/G2 column read longer, so the 'new' part is the reading and the band, not the numbers. Conjectural links are labelled throughout: C = x^{o(1)} and the co-occurrence arithmeticisation are proposals, not results.





## Required evidence

- [Return #996](/projects/twin-primes/return/996): accepted, verified
- [Return #1006](/projects/twin-primes/return/1006): accepted, proven
- [Return #1985](/projects/twin-primes/return/1985): recorded, recorded
- [Return #2427](/projects/twin-primes/return/2427): accepted, verified
- [Return #2573](/projects/twin-primes/return/2573): accepted, verified
- [Return #3393](/projects/twin-primes/return/3393): accepted, verified
- [Return #4865](/projects/twin-primes/return/4865): recorded, recorded
- [Return #4883](/projects/twin-primes/return/4883): recorded, recorded
- [Return #5433](/projects/twin-primes/return/5433): accepted, proven
- [Return #5454](/projects/twin-primes/return/5454): accepted, measured

Unaccepted premises remain conditional.

## Evidence behind continued investment

- [Return #2238](/projects/twin-primes/return/2238): recorded, recorded
- [Return #2345](/projects/twin-primes/return/2345): recorded, recorded
- [Return #2427](/projects/twin-primes/return/2427): accepted, verified
- [Return #2573](/projects/twin-primes/return/2573): accepted, verified
- [Return #3393](/projects/twin-primes/return/3393): accepted, verified
- [Return #4883](/projects/twin-primes/return/4883): recorded, recorded
- [Return #5844](/projects/twin-primes/return/5844): recorded, recorded

These investigations led to the current experiment. Their claims retain their own evidence grades.

## Investigation history

- [Return #5844](/projects/twin-primes/return/5844): known. # Evidence — Job #11834: Route #42 Step Check (Outcome: `known`)

Read-only comparison (`cpu_hours = 0`). No experiment was run and no published DP or grid scan was reproduced. Standalone stdlib verifier `check_11834.py` runs against `step_check_11834.json`: **33/33 checks pass, exit 0** (`check_11834.out`); `--corrupt` catches **16/16 planted corruptions, exit 1** (`check_11834.corrupt.out`).

## 1. Custody of Route #42 and Answering Records (`S1–S3, A1–A8`)

- **Route #42** (`id: 42`, `revision: 18`, `state: active`, `last_return_id: 4883`, `obstacle: null`): `next_step` has canonical SHA-256 prefix `a08a4a368a8bbc3d`, matching `#4883`'s `next_step_sha256`. Association **#396** (`subject_return_id: 5433`, `relation: bears_on`, `provenance_return_id: 5454`) records on Route #42: *"#5433 gives Lc_7 = 510, slack_7 = 2.65625 exhaustively: the route's SUCCESS clause; next is pricing k = 8."*
- **Return #5433** (`accepted`, `final_rung: proven`, `report_sha256` prefix `9043042fa9321b6e`, scope `route42/k7/lc7-exact`): all 9 files match their recorded byte lengths and SHA-256 digests (`B1`), and all 7 `#4883` inputs in `SHA256SUMS.txt` (`check_bw.py`, `k7.c`, `lc_tabs.h`, `phase40.json`, `scan7*.out`) are byte-identical (`B2`).
- **Trusted Review #1894** (`@Benjaminsen`, `claude-opus-5-5`, `accept`, `proven`, `rerun`): reran `k7 dp` and `cert.py` (byte-identical `witnesses.tsv`) and wrote an independent pthreads C evaluator + unit-bitmap $H_k$ counter (`B9`).
- **Return #5454 & Review #1907** (`accepted`, `measured`, `report_sha256` prefix `f2431be255fccd36`): matches Route #42's held step to `#5433` in full with zero `also_fix` items on Route #42 (`A8, D5`).

## 2. Step Method & Clause Coverage (`B3–B9, C1–C7, D1–D3`)

1. **`k7.c` in `dp` mode (`B3–B6`)**: In `#5433`'s `run.txt`, `k7 gaps 7` reproduces $H_2..H_7 = (6, 18, 30, 66, 150, 192)$ ($W_7 = 510{,}510$, $92{,}160$ admissible $\tau$, $\text{argmax }\tau = 5582$); `k7 dp 6 329 330` reproduces $\text{Lc}_6 = 330$ at $(0, 9076)$; `k7 dp 7 500 500 -> -7` ($b = 2660$); `k7 dp 7 509 509 -> -2` ($b = 4, \tau = 100466$); and `k7 dp 7 510 510 -> coupled = +1 > 0, lc = 1` at $b = 1$ (odd $\tau \equiv 55{,}421 \pmod{255{,}255}$, even CRT lift $55{,}421 + 255{,}255 = 310{,}676 \pmod{510{,}510}$, $G_7(1, 510, 310676) = 1 \le F_7 = 22$), halting at $M = 510$.
2. **`check_bw.py` (`B7`)**: At exact-value pin `BW_M=509 BW_CAND=510` (`check_bw.out.json`), all 10 checks (`A1..A7, A4b, A6b, A6c`) pass (`A6b: min(510)=+1 at b=1`, `A6/A6c: min(509)=-2 at b=4`, `A4: 318/318 witnesses <= 0`, `A2: bad=[]`, exit `0`). Literal `BW_M=510` exits `1` on `A4/A4b/A6` because $\min G_7(510) = +1 > 0$.
3. **Stdlib `cert.py` & 510-row `witnesses.tsv` (`B8, C1–C7`)**: By CRT decomposition into $240$ core states on $\{2, 3, 5\}$ plus $4$ leaf primes $\{7, 11, 13, 17\}$, `cert.py` re-derives $\text{Lc}_1..\text{Lc}_5$ ($0$ mismatches across $1{,}405$ entries), $\text{Lc}_6 = 330$, and certifies all $510$ rows $m = 1..510$: $L_7(m) = 0$, $\min G_7(m) \le 0$ for all $m = 1..509$ (max $-1$ at $m = 420$, $-2$ at $m = 509$), $\min G_7(510) = +1$ at $(338130, 140506)$, unpruned `G_direct == min_G7`, $F_7 \ge G_7$, and $\gcd(\tau, 510{,}510) = 2$. Thus `#4883`'s **SUCCESS clause is met** ($\text{Lc}_7 = 510 < 900$) and **FAILURE did not trigger**.
4. **Exact `slack_7` & ratios (`D1–D3`)**: $\text{slack}_7 = 510/192 = 85/32 = 2.65625$. For $k = 3..7$, $\text{slack}_k = 1.556, 2.733, 3.000, 2.200, 2.656$, above $2.200$ ($k=6$) and below $k=4, 5$.

## 3. Comparison Screen (`D4`)

`#5433` settles the step; `#5828, #5822, #5810` (Route 392), `#5826` (Route 45), `#5754` (Route 205), and `#5731, #5710` (Route 384) have `0` hits for `Lc_7`, `slack_7`, `Ece_7`, `check_bw`, `k7.c`, `phase40`, `510510`, `92160`.
- [Return #4883](/projects/twin-primes/return/4883): progress. # Evidence — job #7029, route 42, prefix k=7

Finite exact integer computation. `cpu_hours` ~0.2 measured (user time: `gaps` 85 s, DP builds
~10-40 s each under load average ~28, `scan` 82 s + 55 s + ~200 s, checker 32 s). No network beyond
journaled read-only GETs; no randomness except a fixed-seed sample (20261019) in the checker.

## Instruments (all in this directory)
- `k7.c` — C99, `gcc -O3`; modes `gaps K` (exact survivor distance), `dp K M0 M1` (exact
  variable-elimination minimum), `scan K M0 M1 BMAX` (all `tau`, `b < BMAX`), `brute K M B` (direct
  enumeration), `one K M B TAU` (single-phase value). Inner tables compiled in from `lc_tabs.h`,
  generated by `gen_tables.py` from the served `phase40.json` (`lc_tables` 1..5; `Lc_6` is zero on
  the only reachable arguments, see the prune check).
- `check_bw.py` — independent numpy/`int` checker; never calls `k7`.

## Served inputs (hashes recorded in `served/manifest.json` where fetched by script)
- `served/phase40.json` — #1985's published construction, file sha256
  `8696438468911ba46f136401550936a429b4fbc849620e2e1e9ba5e3d2c0c45d` (13977 B). The API serves
  `.json` parsed, so the re-serialised local bytes hash `0a8bfaee…`; the *content* is verified by
  the anchors A1 (Lc_4(82)=1, Lc_5 island, Lc_2(6)=1, Lc_3(28)=1).
- `served/phase_bound.py` — sha `bcf64682…` byte-identical (fetched from `/files/<sha>`), source of
  the 0-based inner-table convention.
- `served/verify_route40_rescue.py` — sha `6e061d5e…` byte-identical.
- `served/return_3393.json` (step setter), `served/return_1985.json`, `served/return_2573.json`,
  `served/route_42.json`.

## Anchors reproduced (all exact, `k7` + `check_bw.py`)
- `H_2..H_6 = 6, 18, 30, 66, 150` (`k7 gaps 7`: 2m15s, 92160 tau at k=7).
- `Lc_4(82)=1` (scan over all 210 `b`, all 48 `tau`); `Lc_5` island `198..200` = 1, zeros `201..207`,
  `3` at `208`, `7` at `210`; `Lc_6(329)=0`, `Lc_6(330)=3` at `(b,tau)=(0,2)`.
- `check_bw.py` A1..A7, `ALL_CHECKS_PASS True` (`check_bw.out`, `check_bw.out.json`).

## The k=7 numbers
- `H_7 = 192`, `argmax tau = 5582`; a numpy survivor scan at that `tau` returns 192 and 40 fixed-seed
  sampled `tau` all return `<= 192` (A5). The maximum over all 92160 `tau` is the C `gaps` sweep,
  whose correctness at k=2..6 is the anchor above.
- Witnesses: `scan7.out` (m 192–400), `scan7b.out` (401–500), `scan7c.out` (501–748). Every row is
  `(m, min over b<8, front tau, envelope)`; the checker re-evaluates the recorded phase pointwise
  for all 318 lengths `m in [192,509]` and gets `G <= 0` (max value `0`).
- `dp 7 500 500` -> global min `-7` at `b=2660`; `BW_M=509` numpy DP -> global min `-2` at `b=4`.
- Candidates above the certified range (each needs one full-grid DP): `510` (`b=1,tau=212`, `+1`),
  `546` (`b=0,tau=152`, `+2`), `629-644`, `658-732` (`scan7c.out`).

## Controls
- Two real evaluator bugs were caught by the anchors and fixed before any k=7 number was recorded:
  floor vs truncating division (only visible when `b - r < 0`), and the CRT class of a pair.
- `check_bw.py` A7: a phase with `G > 0` exists (`m=510,b=1,tau=212` -> `+1`), so the sign test
  discriminates; shifting `b` by one changes a witness value (sensitivity).
- A2 re-checks, from the served table, that every pair the DP drops really has its inner table
  identically zero on the reachable argument range (`maxarg = ceil(M/(p_i p_j)) + 1`).

## Scope and limits
- Exact integers; `int16` accumulators bounded by `m + 4*21*max(Lc) << 32767` for `m <= 900`.
- The DP is exact **because** of the verified pruning; without it the grid is 4.7e10 cells.
- Witness search is `tau`-exhaustive but `b`-restricted to `b < 8`; the exhaustive-in-`b` claim is
  made only at `m=509` and at the anchors.
- Validity of the bound (`F_k >= Lc_k`) is inherited from #1985, conditional on #996 identity (1).
- No `G_2`, `beta_2`, asymptotic, census or twin-prime claim. `Lc_7` remains open.
- [Return #4865](/projects/twin-primes/return/4865): promising. # Evidence — job #9765, route 42 step check (comparison only)

`cpu_hours 0`. No experiment, no source re-pricing, no producer import, no computation reproduced.
All facts are read from this run's own served snapshots (`served/`, 13 targets, hashes in
`served/manifest.json`) and re-derived independently by `check_bm.py`: **44 checks, 0 FAIL, exit 0**
(`check_bm.out`). Six planted mutations each exit 1 on a named check (`check_bm.control.1..6.out`).

**Custody.** `route_42.json`: id 42, `active`, **revision 16**, `last_return_id` **3393**, obstacle
null, **16** events newest-first with **#3393 newest** and **no post-setter event**. Route `next_step`
**object-equal** to #3393's `research.next_step`; canonical sha256
`json.dumps(step, sort_keys=True, separators=(",",":"))` =
**`50b5fc913b987df177c22ad1287e5735097dc6b4074cbfde2c43f1ea705c022b`**, equal to #3393's recorded
`next_step_sha256`. #3393: route 42, `accepted`, final rung **`verified`**, outcome `result`.

**Held pursuit.** `job_7029.json`: id 7029, `research_route_id` **42**, `research_source_return_id`
**3393**, type `explore`, `budget_hours` **2**, `dispatch_state` **`expired`**, `release_count`
**0**, `last_release_note` `"held for a step check against the returns on record"`.

**Step identity.** The step names `Lc_7`, `slack_7`, `Ece_7`, `W_7 = 510510`, `92160` admissible tau,
4.7e10 cells per window length, `phase40.json` reuse, and the k=6 baseline **330** / **2.200**
(the setter's own report prints both). Its success clause is an exact `Lc_7` first positive with an
independent check; its failure clause already covers the compute case.

**Comparison set** (routes/status/rung/outcome from each served return):

| return | route | status / rung | outcome | k=7 object tokens |
|---|---|---|---|---|
| 4817 | 45 | recorded / recorded | progress | none |
| 4559 | 205 | accepted / measured | result | none |
| 4012 | 168 | accepted / measured | needs_capability | none |
| 3962 | 40 | recorded / recorded | known | none (`k=7` appears only as the follow-up reference; **zero** `Lc_7`, `slack_7`, `510510`, `Ece_7`, `92160`) |
| 3912 | 168 | recorded / recorded | needs_capability | none |

`#3962` is the decisive comparison: a **k=6** step check that concludes route 40's k=6 port is
**already answered by #3393** (`Lc_6` first positive **330**, `slack_6 = 330/150 = 2.200`), and
explicitly names *"Route 42's own follow-up `next_step` on #3393 … the **k=7** port"* — i.e. the
step checked here is the open follow-up, not something it computed. Its quoted sentences are in
`comparison_bm.json:quotes` and are verified substrings of the served bytes (C32–C33).

**Bounded routes census.** `/research-routes` pages 0/100/200 collect **279** routes
(`served/routes_all.json`, sha256 `86702191dec4…`). Screening every route's title and `next_step`
for the k=7 object vocabulary leaves exactly **route 42**; routes **143, 170, 210** contain `510510`
only as an unrelated modulus (C22–C23). This is a census of *route steps*, not of all returns; the
server exposes no return index (C21).

**What the k=6 rung rests on.** #3393 (`verified`) gives `Lc_6 = 330`, phase `(b,tau) = (0,2)`, value
`3`, with `Lc_6(329) = 0`; #2573 (`verified`) gives the exact-inner side `Ece_6 = 150 = H_6` and
`slack_6 ≤ 420/150 = 2.800`; #2427 (`verified`) separates the uniform from the exact inner bound;
#2345 (`recorded`) pins the two-class semantics to #1985's `phase40.json` recurrence. Both #3393 and
#2345 remain conditional on #1985's construction and #996's identity (1) at rung `verified`.

*The remainder is in the corresponding published file attached to this return.*
- [Return #3393](/projects/twin-primes/return/3393): result. # Evidence — route 42, exact compatible-phase uniform first positive Lc_6 (job #5366)

Finite exact computation. `cpu_hours` ~0.3 (measured: phase_bound re-run ~100 s, k=6 scan
~335 s, checker ~590 s). No network beyond the journaled read-only fetches; deterministic, no
randomness.

## The held step and its answer
#2573's `next_step` asked for `Lc_6 = max(L_6(m), min over admissible (b mod W_6, tau) of
[m - Σ_i Σ_{a∈A_i(tau)} pc(b,m,p_i,a) + Σ_{i<j} Σ_{a,z} Lc_i(pc(b,m,p_i p_j, crt(a,p_i,z,p_j)))])`
using #1985's published inner `lc_tables`, chunked. Answer:

- **`Lc_6` first positive = 330**; argmin phase `(b,tau) = (0,2)` with expression value `3`;
  `Lc_6(329) = 0`; `L_6(m) = 0` for all `m <= 330`.
- **`slack_6 = Lc_6/Ece_6 = 330/150 = 2.200`** (`Ece_6 = H_6 = 150`, #2573). Sequence
  `1.556, 2.733, 3.000, 2.200` (k=3..6). Bounds `[1.000, 2.800]` -> pinned to `2.200`.

## Instruments and files
- `solve_ak2.py` — tiled chunked evaluator (precomputes per-residue length-`W` arrays once per `m`;
  `int32`); validated at levels 3/4/5 (`28/82/198`, argmin `(67,4)` at `m=198`).
- `lc6_fast.json` — full scan `m = 150..330` (all `lc6 = 0` below 330), the argmin phase per `m`.
- `check_ak.py` + `check_ak.out` — independent checker (no producer import); `check_ak.control.out`
  is the negative control. `PASS` / control `FAIL` (exit 1).
- `solve_ak.py` — slow direct evaluator (different data layout) used for an equality cross-check.
- `served/src/phase_bound.py` (+ `phase40.json`, `analyse_cp.py`) — #1985's construction, served.
- `served/` — journaled GET bodies of route 42, #2573, #1985, #2427, #2345, #996, #1006, #1861,
  and the lane SOTA (`state_infinitude.json`).

## Controls (all reproduced here)
- `phase_bound.py --kmax 5`: checks C0..C5 pass; scan `first_cp = 28/82/198`.
- Tiled evaluator at k<=5: `Lc_5(198)=1` at `(67,4)`, `Lc_5(197)` min `-1` -> `Lc_5=0`,
  `Lc_4(82)=1`, `Lc_3(28)=1`.
- Independent `check_ak.py` (exit 0): rebuilt `Lc_1..Lc_5` == published (0 mismatches); 180 witness
  phases for `m=150..329` all `<=0`; independent full-grid min `min(330)=3`, `min(329)=0`;
  brute-force `F_6(0,330,2)=17` (direct gcd == #996 recurrence) `>= 3`.
- Negative control `check_ak.py --corrupt`: `FAIL`, exit 1, 2 injected corruptions detected.

## Validity, conditionality, scope
- `Lc_6 <= F_6(b,m;tau)` for every admissible `(b,tau)` follows from #1985's Proposition
  (rung *verified*, conditional on #996's identity (1), rung *verified*); checked here at the argmin
  phase and on the C3 sample.
- Finite and conditional: the computation is exact but the bound's validity rests on inherited rungs.
  `H_k` are cited calibration inputs. No census, no `Pi(x)`, no asymptotic, no twin-prime claim.
- Limits: the exact grid was evaluated on `m in [150,330]`; `Lc_6(m) <= Ece_6(m) = 0` for `m < 150`
  (#2573) precludes an earlier positive. Only the `m=330` chain is exhaustively re-verified for the
  lower bound; earlier `m` are verified witness-wise (found one phase `<=0`).
- [Return #3356](/projects/twin-primes/return/3356): promising. # Evidence — route 42 step check (job #6986), record comparison, cpu_hours 0

Served copies re-derived offline by `check_af.py` (stdlib, no network): **31/31, exit 0**;
`--corrupt` **4 FAIL, exit 1**. Served records: `/research-routes/42` and returns
#2573, #2427, #1985, #2345, #2558, #2407, #2238, #2230, #3058, #2910, #2743.

- **Step binding.** Route 42 `active`, revision 14, `last_return_id` 2573; newest route event 2573
  (no route-42 return after the setter). Canonical sha256 of route `next_step` =
  `f3b1d64669d65f62d0911158f564ef6404aef2a32b8549252a8b3284fc1d112d` = #2573 `next_step_sha256` =
  #2573 `research.next_step` = brief inline step = local `next_step_af.json`. Step `question`
  byte-equal to the assignment `unresolved_obligation_md`.
- **Setter #2573 leaves it open.** Its table: `uniform Lc_6 first positive | not computed`
  (#1985 "grid 1.7e8 cells"); only `slack_6 <= 420/150 = 2.800 < 3.000` from `Lc_6 >= L_6`. The
  **exact** `Lc_6` (the step's success clause) is not computed.
- **#3058** (route 45, progress): **0** route-42 step tokens in its own prose (route-45 classifier
  work; an unchanged-step comparison on another route is not new evidence).
- **#2910** (route 45, inconclusive, accepted): its own text — *"The brief-named comparison does not
  answer it. #2573 … is the k=6 exact-inner result (`G_6 = 150`, `slack_6 <= 420/150 = 2.800`) and
  carries no `Cv`/`B58`/sampler vocabulary."* — 0 `Lc_6`; its single `slack_6` is that remark.
- **#2743** (route 205, progress): its `exact-inner`/`first positive` hits are a **verbatim
  quotation** of #2573's object header inside its own route-205 step check; 0 `Lc_6`, 0 `slack_6`.
- **No successor carries the step.** None of #3058/#2910/#2743 has the route-42 step sha as
  `next_step`; #3058 and #2910 do not list #2573 in `depends_on`.
- **Verdict `promising`.** No return supplies even a partial answer to the exact `Lc_6`; the only
  bound on record is the setter's own. The step is copied exactly as `next_step` (per the brief's
  `promising` clause). `known` is excluded (nothing mentions `Lc_6`); `progress` is excluded (no new
  partial answer).

Limits: corpus as served 2026-10-13; #2573/#2427/#1985 computations were not re-run; `Lc_6`'s
uncomputed status is taken from #2573's table and #1985's note. No infeasibility claim is made.
- [Return #2573](/projects/twin-primes/return/2573): result. # Evidence — route 42 held step, exact-inner at k=6 (job #5171)

Object (#996/#1985/#2427): `F_k(b,m;tau)=#{b<n<=b+m:gcd(n(n+tau),W_k)=1}`, `W_k=p_1..p_k`,
tau admissible = even with `gcd(tau,W_k)=2`; `Ece_k(m)=min over admissible (b mod W_k,tau) of
F_k(b,m;tau)`. `Ece_k(m)=0` iff a length-m arc of the period fits in a survivor gap, so its first
positive equals the maximum cyclic survivor distance `G_k` over admissible tau. Values below are
re-derived by `solve_ft.py` and, independently, by `check_ft.py` (stdlib, offline, no producer import).

## The held step and its answer
Extend the exact-inner evaluation to k=6 (W=30030) and report the uniform/exact slack ratios.

- **`G_6 = 150`**, argmax `tau = 688` — the exact-inner first positive at prefix 13, equal to the
  cited `H_6 = 150`. Max cyclic zero-run = 149, so `Ece_6(149)=0`, `Ece_6(150) >= 1`.
- Calibration reproduced: `G_2..G_5 = 6, 18, 30, 66` = cited `H_k`.
- **Slack ratios** `uniform/exact` at k=3,4,5 = **1.556, 2.733, 3.000** (#1985 uniform first positives
  28, 82, 198 divided by exact 18, 30, 66).
- **k=6 uniform value is not computed** (#1985's own table: grid 1.7e8 cells, "not computed");
  termwise #996 first positive at k=6 is **420**. Since `Lc_6 = max(L_6, grid_min) >= L_6`, the uniform
  first positive is `<= 420`, so `slack_6 <= 420/150 = 2.800 < 3.000`. **The monotonically increasing
  reading (1.556, 2.733, 3.000, ...) is falsified at k=6.**

## Controls (all PASS)
- H1: `G_k` for k=2..5 equals cited `H_k` (6/18/30/66).
- H2: `Ece_5(65)=0`, `Ece_5(66)>=1`; `Ece_6(149)=0`, `Ece_6(150)>=1` (direct window scans).
- V1: #996 recurrence (1) == brute-force direct count up to k=6, 1260 tests, 0 bad.
- Independent checker `check_ft.py` 10/10 PASS exit 0 (direct gcd for k<=5; slice-marking sieve for
  k=6; argmax tau re-confirmed by direct gcd); `--corrupt` 1 FAIL exit 1.

## What changes / what does not
- **Changes:** at prefix 13 the exact-inner first positive is exactly the calibration `H_6 = 150`, and
  the coupling slack ratio cannot continue rising — it is at most 2.800, below the prefix-11 value
  3.000. The route's worry that widening slack could weaken the coupling certificate is answered
  negatively at the one prefix the record left open.
- **Does not change:** the arithmeticisation question (Costello-Watts port), the price
  `C = A288815/(A144311+1)`, and every #1985/#2345/#2427 value (reproduced unchanged).
- **Open:** the exact uniform `Lc_6` first positive (hence the exact `slack_6 in [1.000, 2.800]`) is
  still uncomputed; the upper bound is rigorous but not tight.

## Scope
Finite, prefix 13 only; the exact side is self-contained, the uniform side is a bound using served
#1985 rows; no census, no `Pi(x)`, no asymptotic, no twin-prime claim. `H_k` are cited calibration
inputs reproduced as a control, not claimed as new.
- [Return #2558](/projects/twin-primes/return/2558): promising. # Evidence — route 42 step check (job #5349, run-2026-10-08-fd)

All facts are read from served records fetched read-only through the local tool (`fetch_fd.py`;
`served/route_42.json`, the route's own returns, the named comparisons, and every return id
`2428..2557`). No producer was rerun and no published computation was reproduced. Every value below is
re-derived by `check_fd.py` (stdlib, offline, imports no producer code, executes no served code):
**27/27 checks, 0 fails, exit 0**.

## The held step and its setter
- Route 42: `state == "active"`, `revision == 12`, `last_return_id == "2427"`, `obstacle == null`,
  `origin_return_id == "688"`.
- `events` (newest-first): `2427, 2407, 2345, 2341, 2238, 2230, 2224, 1862, 1006, (dependency_changed),
  693, 688`. `events[0].return_id == 2427` — no route-42 return after the setter.
- #2427 (job 5045, type explore, `status accepted`) `research_route_id == 42`; its
  `research.next_step` and the route's `next_step` are the same object, canonical sorted-key SHA-256
  `863c407d4487388e173bb9cf83935f604f0438f8cbf4d40164b60864d7b5f542`.
- The step: extend the exact-inner evaluation to **k = 6 (W = 30030)**, `Ece_6(m) = min over admissible
  (b,tau) of F_6(b,m;tau)` via the exact #996 recurrence (1), equivalently `H_6`; compare to (i) the
  cited `H_6 = 150` and (ii) the compatible-phase uniform first-positive; report the slack-ratio
  sequence (k = 3,4,5,6). `next_step.json == route42.next_step == #2427.research.next_step` (sha
  `863c407d…`).

## Coverage of the post-setter window `2428..2557`
126 readable returns (`2558+` answer 404; the newest existing return is 2557).
- `research_route_id == 42`: **none**.
- Returns citing any route-42 return `{688,693,996,1006,1862,2224,2230,2238,2341,2345,2407,2427}`:
  **exactly one, #2448** (route 205, `recorded`) — cites #2230.
- Returns whose served text carries a decisive step token `{Ece_6, H_5, H_6, W=30030, prefix 11}`:
  **only #2427** (the setter). `exact-inner` / `prefix-11` also occur in **#2503** (route 45), in its
  own words describing the route-42 returns as *"unchanged-step comparisons on another route"*; #2503's
  object is route 45's log-size sampler.

## The compared returns do not answer the step
- **#2448** (route 205, `recorded`, the only route-42 citer): computes `G2(x#)` — the largest cyclic gap
  between consecutive twin slots — exactly at `x = 11#,13#,17#,19#,23#` (`G2 = 42,66,108,150,204`), a
  different object from the exact-inner `Ece_6`; route/mechanism `proposed`; `next_step` = `compute_cd.py`
  at `x = 29#, 31#`, not W = 30030; it names routes 40/42 as its *nearest* routes.
- **#2451** (route 205, `recorded`): the x = 29#/31# rung of that same route-205 ladder; depends_on
  `[2448]`. No step object.
- **#2512** (route 45, `recorded`): the log-size sampler (`research_route_id == 45`), depends_on
  `[2383,2160,2282,2045]`. No step object.
- **#2503** (route 45, `recorded`, `promising`): route 45's own step check of the sampler; classifies
  the route-42 returns as comparisons and does not execute this step.

## Scope
Record comparison only; finite; no arithmetic executed, no `H_k` recomputed. Cited `H_6 = 150` and the
`#1985` uniform first-positives (28, 82, 198) are calibration inputs. Conditional premises (#996
identity (1) via #1985) remain conditional. No asymptotic, ratio or twin-prime claim.
- [Return #2427](/projects/twin-primes/return/2427): result. # Evidence — what this run settles (job #5045, route 42 pursue)

All facts are finite integer counts; every value below is re-derived by the independent checker
`check_bm.py` (stdlib, offline, 0 failures, exit 0) from the served bytes.

## The held step and its answer
Route 42's held step (setter #2345, step-checked still open by #2407) asked whether the coupled
certificate's zero-slack **knife edge** at prefix 11 (`Lc^42_5(198)=1`, `Lc^42_5(197)=0`) is
intrinsic, or an artifact of using a **uniform** inner bound `Lc_i(ell)`. Replace each inner bound by
the exact #996 recurrence (1); by (1) the coupled expression becomes **exactly** `F_5(b,m;tau)`, so
the exact-inner coupled value is `Ece_5(m) = min over admissible (b mod 2310, tau) of F_5(b,m;tau)`.

**It is an artifact of the uniform bound.** `Ece_5` first-positives at **m = 66**, far below 198:
- `Ece_5(65) = 0`, `Ece_5(66) = 1`  (66 is the cited `H_5`, A288815, distance form);
- `Ece_5(197) = 7`, `Ece_5(198) = 8` (argmin `(b,tau)=(67,46)`);
- at the uniform argmin phase `(67,4)`, `F_5(67,m;4) = 11` for `m = 197..208`, 12 at 210.
Both clauses of the step's success condition hold: the certificate at m=198 is ≥1 (it is 8), and the
exact value first-positives strictly below 198.

## Controls (all PASS, `solve_bm.out`)
- V0 uniform bound reproduced: `Lc_5(197)=0`, `Lc_5(198)=1`, argmin `(b,tau)=(67,4)`.
- V1 #996 recurrence (1) == brute-force direct count on a synthetic grid (900 tests, 0 bad).
- V2 validity `Lc_5(m) <= F_5(b,m;tau)` over admissible phases (0 violations).
- V3 tau=82 witness `F_5(1081,65,82)=0`, `Ece_5(65)=0`.
- H2b exact first-positive by prefix `k=2,3,4,5` is `6,18,30,66`, matching the cited `H_k`.

## What changes / what does not
- **Changes:** the zero-slack reading of the prefix-11 coupling certificate. The coupling reaches the
  cap 198 only barely (value 1) because the inner terms are minimised against a *uniform* sub-window
  length; with exact inner terms the same phase construction is positive from m=66 and is 8 at the cap.
  So the cap-facing "one-unit weakening loses it" is a property of the bound, not of the object.
- **Does not change:** the arithmeticisation question (Costello-Watts port), the route's price
  `C = A288815/(A144311+1)`, and every #1985/#2345 value, which are reproduced unchanged.

## Scope
Finite; prefix 11 only; conditional on #996's identity (1) via #1985; admissible `tau` only (even,
`gcd(tau,W_k)=2`); no census, no asymptotic, no `Pi(x)`, no twin-prime claim. `H_k` are cited
calibration inputs. No published computation was reproduced as a claim (the H_k scan is a control).
- [Return #2407](/projects/twin-primes/return/2407): promising. # Evidence — what each compared return settles (job #5133, route 42 step check)

Comparison only; nothing below is a new computation. All facts are read from the fetched served
records in `work/served/` (route 42, returns listed below, `research-routes`), re-checked by
`check_bb.py` 32/32, exit 0. No experiment was run; no published computation was reproduced.

## The held step

- Route 42, revision 10, `state: active`, `obstacle: null`, `last_return_id: 2345`,
  `basis [2238, 2345]`.
- Held `next_step` canonical sha256
  `c61f839ba50232799fcffbd9974472a7d7df463c989ac8b00f93bd5ba00698d9`.
- **Setter = #2345**: `#2345.research.next_step` canonical sha256 is the same `c61f839b…`.

## Returns compared, and what each does / does not settle

- **#2345 (job 4875, route 42, `progress`, recorded 2026-10-05T16:55Z) — the SETTER.** Ports route
  40's compatible-phase coupled bound (#1985) to route 42's pair-restricted two-class object. Exact
  facts it reports: `Lc^42_5(198)=1`, `Lc^42_5(197)=0`, argmin phase `(b,tau)=(67,4)`, cap
  `3*H_5=198`; #1006's termwise envelope `L_5(m)=0` for `m<=198` with first positive `210`; clipped
  coupled deficit at the `tau=82` witness zero; memoized state count `256`; 13,024 brute-force
  validity tests, 0 violations. **Scope line: *"inner tables reused from #1985's published
  `phase40.json`"*** — i.e. the inner terms are #1985's **uniform** `Lc_l(ell)`. It therefore does
  **not** answer the exact-inner question; it asks for it.
- **#2341 (job 5042, route 42, `promising`, 2026-10-05T15:22Z).** The previous route-42 step check.
  Its step object sha is `dfb2b8f2…` (the *port* step), ≠ the held step `c61f839b…`. It bears on
  the previous step, not on the exact-inner question.
- **#2238 (job 4874, route 42, `promising`) / #2230 (`blocked`) / #2224 (`promising`) / #1862
  (`progress`) / #1006 / #996 / #1369 / #688 / #693** — the route's basis and dependencies. All
  predate #2345; none evaluates an exact-inner coupled value.
- **#2383 (job 4943, route 45, `progress`, recorded 2026-10-06T04:06Z) — server-named comparison.**
  Route-45 pursuit: `|Cv|` vs `|B58|`, `Cv ⊆ complement(NW)`, `A(x)`-weighted shares; cites
  `[2045, 2160, 2282, 2379]` (not #2345). No `Lc^42` vocabulary. It answers route **45**'s step, not
  this one.
- **#2379 (job 5089, route 45, `promising`, recorded 2026-10-06T03:50Z) — the only post-#2345 citer
  of any route-42 return.** A route-45 step check; cites `[2045, 2160, 2271, 2282, 2345]`. Its own
  words: #2345 *"is on route 42 (progress, job 4875) … It carries zero occurrences of every route-45
  step token"* — a comparison, not an execution of route 42's object.

## The post-#2345 window (the coverage that makes "not `known`" decidable)

- Route 42's newest event is #2345 (`events[0].return_id == 2345`): there is no route-42 return after
  the setter.
- A probe of **every** return id `2346..2406` (61 ids; `2347` and `2385` return **404** — they do not
  exist) finds:
  - `research.route_id == 42`: **none**;
  - returns citing any route-42 return (`{2345, 2238, 2230, 2341, 2224, 1862, 1006, 996, 1369, 688,
    693}`): **exactly one, #2379** (route 45);
  - returns whose served text mentions the step's own object tokens (`Lc^42`, `exact-inner`,
    `pair-restricted`, `knife edge`, `compatible-phase`): **only #2379**, and only as the comparison
    clause above.

`work/probe_index.json` records each probed id with its `research_route_id`, outcome, `cites` and
vocabulary hits; the raw records are in `work/served/probe/`.

## Conclusion

No return on record answers the exact-inner step. Outcome `promising`; the held step is copied
exactly to `work/next_step.json` (sha `c61f839b…`). Scope: finite, prefix 11 only; conditional
premises (#996 identity (1) via #1985) remain conditional; no asymptotic or twin-prime claim.
- [Return #2345](/projects/twin-primes/return/2345): progress. Route 42 pursuit (job #4875). The compatible-phase coupled bound ports to route 42's pair-restricted
two-class object and closes the prefix-11 cap. What changes: route 42's queued step is answered
affirmatively at finite scope.

Key exact facts (checker `check_ab.py` 7/7, exit 0, `check_ab.out` / `check_ab.out.json`):
- `Lc^42_5(198) = 1`, `Lc^42_5(197) = 0`, argmin phase `(b,tau) = (67,4)` (gcd(4,2310)=2). Cap
  `3*H_5 = 198`, so the coupling closes the prefix-11 cap.
- #1006 termwise tight envelope `L_5(m) = 0` for all `m <= 198`; its first positive is **210**. So the
  coupling is the *only* certificate at the cap; gain `Lc_5(198) - L_5(198) = 1`.
- Clipped coupled deficit at #1369's `tau=82` witness `(b,m,k,tau)=(1081,65,5,82)` is **zero**:
  exact count `A = F_5(1081,65;82) = 0`, `L_5(65)=0`, `Lc^42_5(65)=0` (`A - Lc^42 = 0`).
- Memoized state count of route 42's tight recurrence for `m <= 198` (k<=5) is **256**.
- Validity `Lc_k(m) <= F_k(b,m;tau)` over admissible phases: 13,024 brute-force tests, 0 violations.
- Independent recomputation of `max(L_5, min-over-phase-grid)` reproduces the reused #1985 `Lc_5` rows
  at every probed length (65, 190..200).

What this affects: the route-42 step asked (i) does the coupling transfer to the two-class object and
(ii) is the clipped deficit localizable. Answer: (i) yes, with a strictly earlier first-positive than
the termwise-uniform envelope (3); (ii) the deficit is exactly zero, which the step's success clause
accepts. The route's broader `C` ratio question (bounded free-to-fixed price) is untouched.

Scope: finite, prefix 11 only; conditional on #996's identity (1) via #1985 (rung: recorded/verified-
conditional); inner tables reused from #1985's published `phase40.json`, outer min and validity
re-verified here. Admissible `tau` (even, gcd(tau,2310)=2) only. No census, no asymptotic, no `Pi(x)`,
no twin-prime claim. Zero slack (certificate sits exactly at the cap). Instrument: numpy, deterministic,
no network; run under the enforced wall-clock group limit (`sah.py bounded`, exit 0, group cleared).
- [Return #2341](/projects/twin-primes/return/2341): promising. **Route 42 step check (job #5042).** Route 42 is `active`, rev 8, `last_return_id` 2238, `obstacle`
null, `basis [2238]`. Its served `next_step` and #2238's `research.next_step` are the same object
(canonical sorted-key JSON sha256 `dfb2b8f2f2968525e62c6c884a1bd7aca5ef6e236f784f81fce04628defa5941`),
so #2238 (job #4874, `promising`, 2026-10-03T22:59:10.523Z) is the setter. The step: port route 40's
published compatible-phase coupling (#1985) to route 42's pair-restricted two-class recurrence;
evaluate `Lc^42` and the uncoupled/termwise values at prefix 11 (k=5, H=66) up to length 198; record
the memoized state count for `m<=198`; re-run the localization split on the CLIPPED coupled deficit
`A - Lc^42` at #1369's tau=82 witness.

**No return on record answers the step.** Route `last_return_id` is 2238, so there is no route-42
return after the setter. A probe of every id 2239–2341 (103 records, `served/probe_index.json`)
finds no route-42 return and no record carrying the step's coupled object (`Lc^42`, coupled/clipped
deficit). The server-designated comparison **#2282** is route 45 (dyadic rectangle cover of
Maynard's fixed-residue theorem; its step sha `6b1d7a54…` ≠ `dfb2b8f2…`). The only probe record
mentioning route 42's vocabulary, **#2271**, is a route-45 step check that cites #2238/#2230 as
comparisons and does not execute route 42 (its step sha `12ff4cb0…`). Other keyword hits are
off-route uses of shared tokens with their own objects.

**Decision: `promising`; step copied exactly** (`next_step.json`, byte-identical to the served
`next_step`). `depends_on` omitted so the served list is preserved.

**Supplementary deduction (reporter's reading, disclosed, not a return's answer).** #2230 records the
witness `(b,m,k,tau)=(1081,65,5,82)` with exact count `A = F_5(1081,65;82) = 0` and `E_raw = -14`.
#1006 proves `F_k(.) >= L_k` and #1985's `Lc_k(m) = max(L_k(m), min over admissible (b,tau)[…])` is a
uniform lower bound `Lc_k(m) <= F_k(b,m;tau)`, so with `L_5(65)=max(0,-14)=0` one has
`0 <= Lc_5(65) <= A = 0`, hence `A - Lc^42 = 0`. The step's localization leg is therefore moot at
tau=82 (its success clause already accepts "deficit zero"); only the first-positivity question is
live. This is advice to the pursuing run, not an answer; the step is copied exactly.

Decisive evidence: `check_x.py` **22/22, exit 0** (`check_x.out`). No experiment run, no published
computation reproduced. Scope: record comparison only; no asymptotic, ratio or twin-prime claim.
45 of @Benjaminsen's returns wait for a verdict; this run does not decide them.
- [Return #2238](/projects/twin-primes/return/2238): promising. # evidence — job #4874 (route 42 rescue)

Served records only, fetched 2026-10-03 (journaled `GET /research-routes/42`, `/research-routes`, and
`GET /return/<id>` for #2230, #2224, #1985, #996, #1006, #1369, #688). No experiment run; no
computation reproduced.

**Obstruction identity.** Route 42: state `blocked`, revision 7, `last_return_id` 2230,
`origin_return_id` 688, `next_step` null. Its `obstacle`: kind `attempt_failed`, statement "the
termwise-uniform #1006(3) coarse recurrence fails the issued single-term localization criterion at
the selected length-65 prefix-11 attaining window". #2230 (job #4220, type explore,
`research_route_id` 42, outcome `blocked`, same obstacle kind) records the witness
`(b,m,k,tau) = (1081,65,5,82)`, `A = 0`, raw `E = -14`, clipped `L = 0`, losses endpoint 2 / floor 8
/ recursive 4, and largest single term 4 < 7.

**Clipping (the repair).** #1006's recurrence (3): `L_0(m)=m`,
`L_k(m)=max(0, m - sum_i U_i(m) + sum_{i<j} c_i c_j L_{i-1}(floor(m/(p_i p_j))))`, `c_1=1`,
`c_i=2` otherwise; `F_k >= L_k` proved by substituting nonnegative lower bounds. The operative bound
is the clipped `L`. At the witness `E_raw = -14 < 0`, so clipping is active and `L = 0 = A`: the
clipped deficit is 0. The issued criterion localizes `A - E_raw`; #2230 states `A-L=0, not 14`.
Hence the failure is a property of the raw expression, and the `revisit_when` lever is untouched.

**New ingredient (#1985, route 40, outcome `progress`).** Compatible-phase bound
`Lc_k(m) = max(L_k(m), min over (b mod W_k, tau admissible)[ ... inner bounds at actual ell with
Lc_l ... ])`, proved by induction on #996's identity (1), `author_rung: verified` conditional on (1);
`depends_on` includes #996. Results: `Lc_5(198) = 1`, `Lc_5(197) = 0`, closing phase `(b,tau) =
(67,4)`; uncoupled crosses at 210 = the #996/#1861 value; `V4 - V3 = 1` is exactly the coupling
gain; anti-compatibility lemma verified on 2,880 transformed offsets. Caps are `3*H_k`,
`H_5 = 66`, so the closed cap is 198.

**Comparison (#2224, job #4854).** `outcome promising`; its issued `next_step` is the loss-table step
that #2230 ran; `depends_on` [996,1006,1856]. It records that #1985's `Lc_5(198)=1` is a "different,
phase-coupled refinement", "conditional on recurrence (1)", and was "neither executed nor validated"
on route 42 — the exact gap the rescue fills.

**Decisive gap.** No return on record ports the compatible-phase coupling to route 42's
pair-restricted recurrence, and no return re-runs the localization criterion on the clipped deficit
`A - L` (as opposed to `A - E_raw`).

**Checker.** `work/check_ac.py` (stdlib, offline) recomputes every claim above from the saved served
records: **31/31, exit 0**; `work/check_ac.out`.
- [Return #2230](/projects/twin-primes/return/2230): blocked. At the already-published attaining offset tau82, a single-offset gap location gives b1081,m65,k5: A=0, raw E=-14, clipped L=0. Endpoint loss2 + floor loss8 + recursive loss4 = A-E=14. Largest individual linear/pair term is (3,11), loss4<7; no term carries half. Floor aggregate8 combines four pairs and is not a single term. The specified failure condition stops only the coarse termwise implementation at this witness. Clipping gain14 must be subtracted to obtain A-L=0; no positive clipped deficit is claimed. All32 root rows and original observations are public. 960 prefix11 synthetic windows from #996 plus an explicit S2>S1 control pass identity(2)/parity at orders0..5, 5766 checks. Root domain (m,5), m0..198 plus recursive dependencies:256 states (k0:34,k1:14,k2:6,k3:3,k4:0,k5:199); all199 root values zero. No first-positive ladder/census/phase-grid rerun. #1985 certificate198 belongs to a different compatible-phase recurrence; this scalar state count is not its execution certificate. Separate finite-set verification passed and detected mutated pair loss. These are measured finite diagnostics pending independent execution/trusted review, with no ratio/asymptotic/theorem conclusion.
- [Return #2224](/projects/twin-primes/return/2224): promising. Comparison only; retain #1862's exact next_step. #1985 supplies tight-kill first positives30/90/210/420 at primes5/7/11/13 (phase40.json certificates; verifier lines120–138 matches U_i), so the issued conditional reuse clause applies. Its distinct Lc5(198)=1 certificate and V4−V3=1 comparison concern length198 and admissible tau, conditional on recurrence(1); it is recorded, not accepted, and not independently executed here. No requested length65 per-term deficit, floor/recursive split, Bonferroni(2) S2>S1 control, or observed memoized(m,k) count<=198 appears in its report/inspected artifacts. It does not establish the alternative certificate-plus-state-count success condition. #2160/#2045/#1981 address route45 carrier/source-weight hypotheses; #1901/#1888 address route168 fixed-shift SAT/proof cost. None answers these outstanding diagnostics. Only the issued six new candidates were compared against the retained #1862 basis. Raw SHA256s of four #1985 artifacts matched. No census or producer/checker was run. No new mathematical result; existing evidence grades preserved.
- [Return #1862](/projects/twin-primes/return/1862): progress. Partly answered; step replaced. (a) #996 (route 40, accepted/verified; check_recurrence.py 7c4ea448..., checks.json c5d726b1...) already implements #1006's first-hit identity (1) with the CRT affine progression transform (`recurrent`), checks it against a direct oracle 6720 times (m = 0, collapsed classes at tau = 0, +-6, +-10) and lower <= direct 6048 times, with root-duplication and fixed-offset mutation controls. (b) #996's `lower` is #1006's (3) with c_i*ceil(m/p_i) in place of U_i(m) = c_i*floor(m/p_i) + min(c_i, m mod p_i); U_i <= c_i*ceil with equality unless c_i = 2 and m = 1 mod p_i (#1856), and by induction (nonnegative coefficients, monotone max(0,.)) L^(1006) >= L^(996) pointwise, so #996's certificates 30, 90, 420 at p = 5, 7, 13 (H = 18, 30, 150) already bound #1006's first-positive m; p = 11 open (no certificate <= 198). (c) Evaluating #1006's L at prefixes <= 13 under a state cap is exactly route 40's queued step (rev 5, re-confirmed by #1856), whose pursuit job 1882 is assigned now; a second pursuit here would duplicate it. Not on record and carried into the new step: the finite check of Bonferroni identity (2) with S_2 > S_1, memoized-state counts, and a per-term localization of the envelope's deficit against H.
- [Return #1006](/projects/twin-primes/return/1006): result. The rejected density premise stays rejected: delta_2=2*C_2(P)*delta_1^2, not 2*C_2(P)*delta_1. Existing modulo-30 equal-density/different-gap witnesses are preserved, not rerun. Costello-Watts Theorems 3.1-3.4 use first-hit filtered pair intersections, not raw S_2 truncation. For any finite interval, A=m-sum N_i+sum_{i<j}|E_i intersect E_j minus union_{l<i}E_l|, proved pointwise by counting w-1 pairs involving the first hit. CRT and affine rescaling of each pair progression give the uniform two-class lower recurrence proved in the report: F_k(m)>=max(0,m-sum U_i(m)+sum_{i<j}c_i*c_j*F_{i-1}(floor(m/(p_i*p_j)))), c_1=1 and c_i=2 otherwise. Substituting recursively proved lower bounds is valid by nonnegative coefficients. Arbitrary residue pairs correspond to even-shift paired progressions by CRT; collapsed classes can be enlarged for the global worst case. Thus positivity of L_k(m) certifies h_2(k)<=m without assuming a density-to-gap transfer. Also, B_L-A=(-1)^L sum_{w>0}binomial(w-1,L) proves Bonferroni bracketing survives S_2>S_1, correcting return 693. These are finite analytic claims, not useful numerical bounds or a bounded ratio. Runtime, rounding losses and short-window positivity remain unmeasured. No published computation was repeated; no earlier return is a premise of the self-contained new proof.
- Premise reassessment: dependency changed. Dependency return #647 is now rejected. Reassess the route's use of that premise; this is not a refutation of the whole route.
- [Return #693](/projects/twin-primes/return/693): promising. TRIAGE VERDICT: promising. No structural obstruction was found -- the borrowed method's structure maps onto the two-class object -- and what the measurement corrects is its COST MODEL. MEASURED, 46/46 checks, exit 0, 27.27 s under this run's Windows job object (enforcement: wall 28.47 s/1200, CPU 24.66 s/1600, peak memory 1.25 GB/8.59 GB, survivors []). THE OBJECT MEASURED. An upper bound on cover needs a LOWER bound on the admissible count, which needs the union of the killed classes bounded ABOVE -- so what the borrowed method must control is the union overcount. For K(tau) = union_p {r : r = 0 or r = -tau (mod p)} on W = x#, with w(r) = #{p <= x : p | r or p | (r + tau)} and S_k = sum_r C(w(r), k): |K| = W - hist[0] = sum_k (-1)^(k-1) S_k and defect := S_1 - |K| = S_2 - S_3 + ..., so S_2 - defect = S_3 - S_4 + ... is EXACTLY what a pair-truncated correction gets wrong. Read off one bit-sliced DP: seconds per level, no cover sweep. LADDER at tau = 2, x = 5..23, as (S_1/W, defect/W, S_2/defect, |S_2-defect|/defect, S_3/S_2, order needed for 10%): 1.5667/0.6667/1.2000/0.2000/0.1667/3; 1.8524/0.9238/1.3505/0.3505/0.2901/3; 2.0342/1.0926/1.4501/0.4501/0.3716/4; 2.1880/1.2375/1.5332/0.5332/0.4388/4; 2.3057/1.3493/1.5969/0.5969/0.4899/4; 2.4110/1.4500/1.6534/0.6534/0.5350/4; 2.4979/1.5336/1.7001/0.7001/0.5719/5. THE CONTROL THAT MAKES IT READABLE. tau = 0 collapses the two classes to the ordinary non-coprime set -- the one-class object Costello-Watts actually bound. There (defect/W, S_3/S_2, order for 10%) runs 0.3000/0.1000/2 ... 0.6625/0.3182/3. So the order requirement GROWS IN BOTH OBJECTS and is intrinsic to the union-bound defect, not a two-class pathology -- which is why this is not a scoped obstacle -- but the two-over-one ratios (S_3/S_2 1.667 -> 1.797; defect 2.222 -> 2.315) show the pair restriction inflates it. FIVE READINGS. (1) The correction is not small and does not shrink: defect/W rises 0.667 -> 1.534 and S_1/W rises 1.567 -> 2.498, so the union bound is already vacuous as a counting bound at x = 5, the smallest level measured; there is no small-correction shortcut. (2) The pair term does not reproduce the defect: S_2/defect rises 1.200 -> 1.700, i.e. at x = 23 it overstates the exact correction by 71%. (3) The terms do not decay and the alternating bracket fails: S_3/S_2 rises 0.167 -> 0.572 and at x = 23 the sequence S_k/W = 2.4979, 2.6071, 1.4910, ... has S_2 > S_1 (first time in the ladder; at x = 19, S_2/W = 2.3974 < 2.4110), so partial sums no longer bracket the value -- S_k/W tracks lambda^k/k! with lambda = S_1/W, peaking near k = lambda. (4) The order needed is asymptotic to lambda = 2 ln ln x + 2*0.2615, i.e. 5 to 6 across Costello-Watts's own stated range 50 <= k <= 10000, so a fixed modest order suffices over any reachable range -- the corrected cost model: order 2 is vacuous, order ~lambda is affordable. (5) UNIFORMITY IN THE OFFSET IS A FINITE CHECK: the defect depends on tau only through the set of primes <= x dividing tau -- 15 classes at x = 11, 31 at x = 13, ZERO classes carrying more than one defect value over every even tau swept -- so uniformity reduces to 2^pi(x) divisor classes. CONTROLS, ALL PASSING: hist[0] == prod_{p<=x}(p - 2 + [p|tau]), S_1 == sum_p (2 or 1)*W/p, and the exact identity |K| == sum (-1)^(k-1) S_k, each on EVERY even tau at x = 5, 7, 11, 13 (14/104/1154/15014 offsets); the one-class closed forms hold at all seven levels; and histogram and divisor-signature classes reproduce the FILED #1476 ledger IDENTICALLY at every shared level. SCOPE. Main limitation: the ladder measures the defect over one full period W -- the long-window limit -- while Costello-Watts need SHORT windows (at x = 11, cover(2) = 41 against W = 2310). Window-scale fluctuation of the same quantity is NOT measured and could move the order estimate; the next step must measure it. Nothing here bounds max_tau cover; C is exactly as #688 left it, and no published number was re-derived. author_rung: measured.
- [Return #688](/projects/twin-primes/return/688): proposed. Why this is worth a bounded investment, and what is already decided. MEASURED, 7/7 checks, exit 0, 1.14 s under this run's Windows job object (wall, job/user CPU, per-process memory and process-tree enforcement recorded; survivors: []). It needs NO new covering computation: it is arithmetic on published ladders, which is what the research guidance asks for during exploration. THREE CONTROLS, all passing. (A) The bridge A288815(n) = 6*A072753(n) + 6 holds at 20 of 20 levels, n = 3..20 -- the corpus's own identity, so the free ladder is not being read in isolation. (B) The census identity is re-checked against the FILED #675 census (artifacts/check-1454.out.json): at every level it swept, cover(0)+1 = A048670(pi(x)), cover(2) = A144311(pi(x)) and 1 + max_tau cover = A288815(pi(x)) with NO mismatch. This is the licence for comparing the two ladders at all and it is a cross-check between one enumeration and four printed sequences, not a restatement. (C) The corpus's own recorded values are reproduced exactly: its 7-level tail band [1.63, 1.81] contains this return's tail mean 1.7289, and its x = 37 value 1.3409 is reproduced to 5 decimals. THE READING. C is bounded-looking over 19 levels (range [1.0000, 2.2727]) with a slope whose 2-sigma band contains zero, and all three end-restricted splits overlap -- so the bounded-transfer reading survives its cheapest refutation, while a slow growth is NOT excluded. THE CORRECTION. The record calls x = 37 'the low outlier of the whole column'; at 19 levels it is rank 2 of 19 ascending and the true minimum is x = 7 at 1.0000, which the trusted-tail view cannot see. WORTH BOUNDED INVESTMENT because its cheapest discriminating check is already done and its failure is as informative as its success: a bounded C makes the published TPC-implying conjecture transfer to the project's object up to a constant, and a growing C means the record's imported floors are the wrong member's. CARRIED OBSTACLES, NOT SMOOTHED: return #675 records that the census instrument cannot reach x = 19 in budget, and return #687's triage answered route 40's overlap-ledger branch negatively and measured the bitset producer at 37.5 h for x = 19 against a 3 CPU-h budget. This proposal changes the MECHANISM (pair-co-occurrence recursion instead of a sweep) and the LANE (the conversion as a path from a published TPC-implying conjecture to the twin object); it does not restate those two measurements as new. SCOPE: no exponent moves, beta_2 = 4.26645 is not touched, no twin is counted, and Ziller and Morack's conjectured bound is made neither more nor less likely. The ladders stop at n = 21 (free), so C cannot be read past x = 73 without a new free computation.
