d6g8k5htny-coder · Oct 20, 2026 · Oct 20, 2026
diff --git a/reviews/zeta_fold_slice_c_r2_replay_20261020/README.md b/reviews/zeta_fold_slice_c_r2_replay_20261020/README.md
@@ -0,0 +1,62 @@
+# R2 replay venue: different-provider replay of the zeta fold slice C package
+
+**Status: venue only.** No replay has run in this record yet. The packet README's
+§9 slice C line stays NOT_PERFORMED. Scientific effect NONE.
+
+## Why this exists
+
+- **R2 of main#451** is the full replay of the 59 README §8 commands of
+  `experiments/zeta_fold_exploration` by a provider other than Anthropic.
+  - It is split into chunks in the [ledger](https://github.com/d6g8k5htny-coder/main/pull/451#issuecomment-6281117591).
+  - The coordinator's open offer is [main#451 comment 6281445636](https://github.com/d6g8k5htny-coder/main/pull/451#issuecomment-6281445636).
+- **A Cursor cloud agent starts from a pull request's head branch.** main#451's
+  branch was deleted after its merge, so a request there failed at startup
+  ([6281525014](https://github.com/d6g8k5htny-coder/main/pull/451#issuecomment-6281525014)).
+  This draft gives one verified non-Anthropic execution worker a place to run
+  the replay and record it.
+- **Owner approval.** Dylan Roy chose "Allow a venue PR" in Claude Code session
+  `session_017Mi3hxjaxV45x6zo6o1ee3` on 20 October 2026 (about 02:25Z), with
+  these conditions, verbatim:
+
+  > Use one verified non-Anthropic execution worker for R2, check existing
+  > claims first, and preserve all 59 commands, environment details, output
+  > hashes and discrepancies. Keep README §9 unchanged until the complete replay
+  > is reviewed. No merge or scientific acceptance is authorized by the venue
+  > approval.
+
+## Source
+
+- **Packet:** `experiments/zeta_fold_exploration`, tree
+  `bca06db86581b2ff878fa321c4040c7b8f7bd55d`. It is the same at `0a7e5145` (the
+  ledger's commit), at `73a3cfcd` (the main#451 merge) and at `5cab7663` (this
+  branch's base).
+- **Expected outputs:**
+  - the 66 committed files under the packet's `results/`;
+  - the 9 raw-sample hashes in the packet's `MANIFEST.json`.
+- **Commands, output mapping and timing masks:**
+  `reviews/zeta_fold_slice_c_rerun_20261019/`, specifically `scripts/commands.sh`,
+  `scripts/driver.py` and `comparison/timing_mask_rules.txt`.
+
+## What the finished record holds
+
+1. **Performer:** provider, model and session.
+2. **Environment:** Python, numpy and mpmath versions, BLAS, CPU count and
+   thread pins, plus every deviation from the pinned environment.
+3. **Jobs:** all 59, each with its exact command, exit code and wall time.
+4. **Committed output files:** for each, the SHA-256 of the replay output and
+   its class:
+   - IDENTICAL;
+   - TIMING-ONLY, meaning equal after the masks;
+   - DIFFERENT, with the diff.
+5. **Raw samples:** SHA-256 and bytes, compared with `MANIFEST.json`.
+6. **Residual trails:** for `indep_ht_c910.out` and `indep_ht_c3435.out`,
+   compared with both the committed copies and the same-provider copies.
+7. **Discrepancies:** every one, explained or not.
+
+## Limits
+
+- Partial chunks are progress records, not R2.
+- A run by an Anthropic performer earns no R2 credit.
+- This venue authorizes no merge and no scientific acceptance.
+- README §9 changes only after the complete replay is reviewed, in a separately
+  reviewed change.
diff --git a/reviews/zeta_fold_slice_c_r2_replay_20261020/RESULT.md b/reviews/zeta_fold_slice_c_r2_replay_20261020/RESULT.md
@@ -0,0 +1,22 @@
+# R2 replay record
+
+Pickup `ZF-SLICE-C-R2-REPLAY-20261020` on main#458, comment 6282014067.
+Performer: Cursor cloud agent `bc-8084f8ef-3895-401f-833b-33fbb172458c`, provider xAI, model Grok 4.7.
+Chunks C1–C7, all 59 jobs. Scientific effect NONE. This file does not update the packet README §9 and does not accept the replay.
+
+The directory `README.md` is the venue stub and was not edited.
+
+| record | what it holds |
+|---|---|
+| `environment.md` | Python, numpy, mpmath, BLAS, host, thread pins, deviations |
+| `jobs.md` | 59 commands, exit codes, wall times |
+| `jobs.jsonl` | the driver ledger, including stdout SHA-256 |
+| `outputs.md` | 66 committed `results/` files, replay SHA-256 and class |
+| `outputs.json` | the same rows as JSON |
+| `raw_samples.md` | 9 raw samples against `MANIFEST.json` |
+| `residuals.md` | quoted lines for the two residual trails |
+| `diffs/` | unified diffs of the two DIFFERENT files |
+| `discrepancies.md` | every difference found, including the class-set swap |
+| `summary.json` | totals |
+
+Totals: 59 jobs, 0 nonzero exit codes, summed wall 2492.9 s. Outputs: 50 IDENTICAL, 14 TIMING-ONLY, 2 DIFFERENT. Raw samples: 9/9 match. `track_low10_dt_sensitivity.out` rebuilt and IDENTICAL.
diff --git a/reviews/zeta_fold_slice_c_r2_replay_20261020/diffs/results__dbn__indep_ht_c3435.out.diff b/reviews/zeta_fold_slice_c_r2_replay_20261020/diffs/results__dbn__indep_ht_c3435.out.diff
@@ -0,0 +1,21 @@
+--- committed/results/dbn/indep_ht_c3435.out
++++ replay/results/dbn/indep_ht_c3435.out
+@@ -1,8 +1,15 @@
+ 0 -0.384561128939277724013128792521 223.275780175619648649859557793 0.0002587
+ 1 -0.384561116627981708329326162492 223.275780185615520178211864857 2.2307e-8
+ 2 -0.384561116627981414864807288345 223.275780185615519913542719805 5.5813e-16
+-3 -0.384561116627981414864807288345 223.275780185615519913542719805 8.2634e-32
+-4 -0.384561116627981414864807288345 223.275780185615519913542719805 1.2375e-63
++3 -0.384561116627981414864807288345 223.275780185615519913542719805 8.2336e-32
++4 -0.384561116627981414864807288345 223.275780185615519913542719805 1.7025e-33
++5 -0.384561116627981414864807288345 223.275780185615519913542719805 2.4785e-34
++6 -0.384561116627981414864807288345 223.275780185615519913542719805 1.2692e-33
++7 -0.384561116627981414864807288345 223.275780185615519913542719805 1.2365e-33
++8 -0.384561116627981414864807288345 223.275780185615519913542719805 7.0004e-34
++9 -0.384561116627981414864807288345 223.275780185615519913542719805 1.0982e-33
++10 -0.384561116627981414864807288345 223.275780185615519913542719805 5.585e-34
++11 -0.384561116627981414864807288345 223.275780185615519913542719805 6.2944e-34
+ t* -0.384561116627981414864807288345 z0 223.275780185615519913542719805
+ A -1.7635316019e-36 B/A -0.486921742083 C/A 0.0700905480743 kappa_r -0.1938233735 kappa_l 0.749687436
+-time 413.21186900138855
++time 410.47313690185547
diff --git a/reviews/zeta_fold_slice_c_r2_replay_20261020/diffs/results__dbn__indep_ht_c910.out.diff b/reviews/zeta_fold_slice_c_r2_replay_20261020/diffs/results__dbn__indep_ht_c910.out.diff
@@ -0,0 +1,12 @@
+--- committed/results/dbn/indep_ht_c910.out
++++ replay/results/dbn/indep_ht_c910.out
+@@ -2,7 +2,7 @@
+ 1 -1.87790907297305930415832167505 99.3284693055762452728350223774 1.1114e-6
+ 2 -1.87790907297253526476683961842 99.3284693055757509480227687829 1.0184e-12
+ 3 -1.87790907297253526476683971813 99.328469305575750948022768886 2.0282e-25
+-4 -1.87790907297253526476683971813 99.328469305575750948022768886 4.128e-50
++4 -1.87790907297253526476683971813 99.328469305575750948022768886 8.3279e-51
+ t* -1.87790907297253526476683971813 z0 99.328469305575750948022768886
+ A 2.76274650385e-16 B/A -0.419057782426 C/A 0.0629860725066 kappa_r -0.09927456 kappa_l 0.448990254
+-time 112.64180207252502
++time 142.09393763542175
diff --git a/reviews/zeta_fold_slice_c_r2_replay_20261020/discrepancies.md b/reviews/zeta_fold_slice_c_r2_replay_20261020/discrepancies.md
@@ -0,0 +1,80 @@
+# Discrepancies
+
+Nothing was edited, fixed, or rerun after the first pass. Both driver lanes exited 0. All 59 jobs exited 0.
+
+## Counts against the committed packet
+
+- 66 tracked files under `results/`: 50 IDENTICAL, 14 TIMING-ONLY, 2 DIFFERENT.
+- `results/dbn/track_low10_dt_sensitivity.out`, rebuilt from the five `track_dt*` stdout logs by the header rule, is IDENTICAL to the committed file.
+- 9 of 9 raw samples match `MANIFEST.json` in SHA-256 and byte length.
+- The five untracked `results/dbn/track_log_dt*.txt` files were left in the execution clone and are not in this record. Their SHA-256 values:
+
+| file | sha256 |
+|---|---|
+| `track_log_dt0.01.txt` | `38ef62f02aad4c45bcfde7fa23a140353fb6f3dd1e85fc5b2f9a0ee05d84d612` |
+| `track_log_dt0.02.txt` | `756ed7b7bd759c83613a8a24986a15562d955b73abb6bdc4fb923b0c706b6266` |
+| `track_log_dt0.025.txt` | `e2973730468b0c0eacaf97ee70ba83be89b95812afb523fa118b7d3075297f84` |
+| `track_log_dt0.04.txt` | `edbd12eb54668f2d22b26fe27877292a8f6027509174d3597d1b2c9880521bcf` |
+| `track_log_dt0.05.txt` | `a4d38f87fdd71148132f7e60357f234db45ef6bd838405f67086956e9b785341` |
+
+## The two DIFFERENT files
+
+`results/dbn/indep_ht_c910.out` and `results/dbn/indep_ht_c3435.out`. Unified diffs are in `diffs/`. Line quotes are in `residuals.md`.
+
+Converged lines are the same as the committed file in both cases: `t*`, `z0`, `A`, `B/A`, `C/A`, `kappa_r`, `kappa_l`.
+
+Post-convergence residual steps differ from the committed file and match the same-provider copy at `reviews/zeta_fold_slice_c_rerun_20261019/reproduced/results/dbn/`, except the `time` line.
+
+`indep_ht_c910.out`, iteration 4, the only residual-step line that differs:
+
+```
+committed:      4 -1.87790907297253526476683971813 99.328469305575750948022768886 4.128e-50
+this replay:    4 -1.87790907297253526476683971813 99.328469305575750948022768886 8.3279e-51
+same-provider:  4 -1.87790907297253526476683971813 99.328469305575750948022768886 8.3279e-51
+```
+
+`indep_ht_c3435.out`, from iteration 3. The committed file stops at iteration 4. This replay and the same-provider copy both continue through iteration 11:
+
+```
+committed:      3 ... 8.2634e-32
+committed:      4 ... 1.2375e-63
+this replay:    3 ... 8.2336e-32
+this replay:    4 ... 1.7025e-33
+this replay:    5 ... 2.4785e-34
+this replay:    6 ... 1.2692e-33
+this replay:    7 ... 1.2365e-33
+this replay:    8 ... 7.0004e-34
+this replay:    9 ... 1.0982e-33
+this replay:   10 ... 5.585e-34
+this replay:   11 ... 6.2944e-34
+```
+
+The same-provider copy has those same eight replay lines (iterations 3 through 11). `t*` and `z0` on every one of those lines are the committed values. The `time` lines differ across all three copies and are timing, not the residual trail.
+
+This replay does not explain the trail. It reproduces the same-provider trail rather than the committed trail.
+
+## Class membership versus the same-provider rerun
+
+The same-provider `comparison/COMPARE.md` has 14 TIMING-ONLY files including `results/dbn/indep_collision.out` and not including `results/stats/rs_spot_check.out`. This replay swaps those two, and the count stays 14.
+
+`indep_collision.out` is byte-identical to the committed file here (440 bytes). The same-provider copy differs only in two timing tokens, which the mask would call TIMING-ONLY:
+
+```
+committed / this replay:   (1 s)    and   (2 s)
+same-provider copy:        (2 s)    and   (3 s)
+```
+
+The other lines of that file match.
+
+`rs_spot_check.out` is TIMING-ONLY here. One token differs:
+
+```
+committed:    window [999700,1e+06]: ours 573  nzeros 573  (0.1s)
+this replay:  window [999700,1e+06]: ours 573  nzeros 573  (0.0s)
+```
+
+The counts on that line are 573 and 573 in both copies. The mask `\([0-9.]+ ?s\)` treats `(0.1s)` and `(0.0s)` as the same timing token.
+
+## Environment deviations
+
+Recorded in `environment.md`: Clang-built uv CPython 3.13.16 instead of a distro `/usr/bin/python3.13` venv; kernel `6.12.94+`; CPU at 2400 MHz. Numpy, mpmath, the mpmath backend, and the OpenBLAS version string match the pinned run.
diff --git a/reviews/zeta_fold_slice_c_r2_replay_20261020/environment.md b/reviews/zeta_fold_slice_c_r2_replay_20261020/environment.md
@@ -0,0 +1,41 @@
+# Environment
+
+Performer: Cursor cloud agent `bc-8084f8ef-3895-401f-833b-33fbb172458c`.
+Provider: xAI. Model: Grok 4.7 (run model `grok-4.7-high-fast`). Not an Anthropic model.
+Pickup: https://github.com/d6g8k5htny-coder/main/pull/458#issuecomment-6282014067
+
+## Source
+
+- Repository commit: `5cab7663820898aaee2466cbc8662eb04984cde0` (this PR's base).
+- `experiments/zeta_fold_exploration` tree: `bca06db86581b2ff878fa321c4040c7b8f7bd55d`.
+- Separate clone: `/tmp/r2`, detached at that commit. The packet scripts overwrote result files only in that clone.
+- Runner: `reviews/zeta_fold_slice_c_rerun_20261019/scripts/driver.py` from that commit, both lanes at once (`ZF_LANE=dbn` and `ZF_LANE=stats`). `ZF_PKT=/tmp/r2/experiments/zeta_fold_exploration`, `ZF_PY=/tmp/r2-venv/bin/python`, `ZF_LOG=/tmp/r2-logs`.
+
+## Interpreter
+
+- Python 3.13.16, `main, Oct 14 2026, 01:04:01) [Clang 23.1.3 ]`.
+- Venv `/tmp/r2-venv`, created from the uv standalone `cpython-3.13.16-linux-x86_64-gnu`.
+- `numpy==2.5.3`.
+- `mpmath==1.4.1`, backend `python`. `gmpy2` is not installed. Pillow is not installed.
+- NumPy BLAS and LAPACK: `scipy-openblas` `0.3.34.106.0`.
+- OpenBLAS configuration string: `OpenBLAS 0.3.34.106.0  USE64BITINT DYNAMIC_ARCH NO_AFFINITY Haswell MAX_THREADS=64`.
+
+## Host
+
+- Linux `6.12.94+` x86_64, glibc 2.39 (Ubuntu GLIBC 2.39-0ubuntu8.9).
+- 4 vCPU, `Intel(R) Xeon(R) Processor`, `cpu MHz` 2400.000.
+- 15 GiB RAM.
+
+## Thread pins
+
+Set by `driver.py` for every job, and present in the process environment:
+
+`OMP_NUM_THREADS=1 OPENBLAS_NUM_THREADS=1 MKL_NUM_THREADS=1 PYTHONDONTWRITEBYTECODE=1`
+
+## Deviations from the pinned run in `reviews/zeta_fold_slice_c_rerun_20261019/environment.json`
+
+1. This image has no `/usr/bin/python3.13`. The venv is the uv standalone CPython 3.13.16 built with Clang 23.1.3, not a venv of a distro `/usr/bin/python3.13`. The patch version is 3.13.16.
+2. Kernel `6.12.94+`. The pinned host kernel is `6.18.44`.
+3. The CPU model string here is `Intel(R) Xeon(R) Processor` at 2400 MHz. The pinned record says `Intel(R) Xeon(R) Processor @ 2.10GHz`.
+
+Package pins, the mpmath backend, and the OpenBLAS version string match the pinned run. The jobs were run with these deviations recorded.