Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
6 changes: 3 additions & 3 deletions experiments/zeta_fold_exploration/MANIFEST.json
Original file line number Diff line number Diff line change
Expand Up @@ -9,7 +9,7 @@
},
"generated_by": "make_manifest.py",
"note": "committed_files lists every file meant to be committed (this manifest excluded). raw_files are regenerable samples written to results/raw/ and not committed; their hashes are from the exploratory run; regeneration is expected to be byte-identical only with OMP_NUM_THREADS=1 OPENBLAS_NUM_THREADS=1 (README section 8).",
"committed_total_bytes": 554906,
"committed_total_bytes": 558582,
"committed_files": [
{
"path": "CLAIMS.md",
Expand All @@ -18,8 +18,8 @@
},
{
"path": "README.md",
"bytes": 106888,
"sha256": "acf855fe30ed4b799d55636952668ea238581ed46255660183693cf3c098f867"
"bytes": 110564,
"sha256": "c2c932769e70204fa060e827e7b5bcf1adae2940e248e966ede75ed98924943a"
},
{
"path": "dbn/check_collision.py",
Expand Down
57 changes: 53 additions & 4 deletions experiments/zeta_fold_exploration/README.md
Original file line number Diff line number Diff line change
Expand Up @@ -1154,6 +1154,19 @@ the saved `.out` file named in its docstring is that stdout. Approximate runtime
- `dbn/indep_ht.py 60 xi` (seconds), `dbn/indep_ht.py 60 c910` (~2 min),
`dbn/indep_ht.py 60 c3435` (~7 min),
`dbn/lobe_identity.py` (~8 min).
The `60 c910` and `60 c3435` commands above, and in the script's docstring, do not reproduce
the committed `indep_ht_c910.out` and `indep_ht_c3435.out`:
- **Current script.** With `dbn/indep_ht.py`, `60 c910` matches every line except the
iteration-4 step. `60 c3435` does not meet its stop rule within 12 iterations, although its
final `t*`, `z0`, A, B/A, C/A and κ lines match.
- **Where the committed files come from.** They are the stdout of the review-math run of the
earlier script, blob `c561d95b`. That run is retained, with the script, under
`recovery/session_01S16677s1SGeAJmN3QMz4ip/zeta_raw/review-math/` (commit `5d298b9e`).
- **Exact reproduction.** With Python 3.13.16, mpmath 1.4.1 (python backend) and
`OMP_NUM_THREADS=1 OPENBLAS_NUM_THREADS=1`, `python -B <c561d95b> 55 c910` and
`… 90 c3435` reproduce the two files exactly apart from `time`.
- **Limit.** The original command lines were not recorded, so this is a sufficient
invocation, not a log (§9).
- `theta/check_theta.py` and `stats/check_gue_transfer.py` (seconds),
`theta/project_theta_checks.py` (~7 min).
- `python3 -B experiments/universality/fold_controls.py`, run from the repository root
Expand Down Expand Up @@ -1236,6 +1249,39 @@ A nonauthor read from a non-Anthropic provider was requested, in four slices:
execution of the same programs and fixed inputs, not the replay's own streams, and it makes
no scientific acceptance.

**Residual trails reproduced exactly.** The replay's two differing files are
`results/dbn/indep_ht_c910.out` and `indep_ht_c3435.out`. At the documented dps 60,
`c910` matches every line except the iteration-4 step, and `c3435` differs from
iteration 3 on and does not meet its stop rule. The final `t*`, `z0`, A, B/A, C/A
and κ lines are identical.
- **Provenance.** The committed files are the same git blobs as the review-math outputs kept
in the packet author's recovery checkpoint, under
`recovery/session_01S16677s1SGeAJmN3QMz4ip/zeta_raw/review-math/` (commit `5d298b9e`).
Those were written by the earlier script `c561d95b`.
- **Mechanism.** That script sets the working precision before it computes the subinterval
constant `U = mpf('1.3')`. The current script (`736760f2`) computes `U` at 60 digits first.
At dps 60 the two scripts compute the same `U`. At another requested precision the
breakpoints differ, and the data note records that this shifts the quadrature noise in
the iteration-4 step.
- **Reproduction.** In the pinned environment, `c561d95b` reproduces both committed files
exactly, apart from `time`, as `55 c910` and `90 c3435`. These are the only matches in
the ranges tested: dps 50–70 for `c910` and dps 85–100 for `c3435`.
([data note](https://github.com/d6g8k5htny-coder/main/pull/458#issuecomment-6322382147);
outputs retained on draft [main#471](https://github.com/d6g8k5htny-coder/main/pull/471) at
`95a00ce4`.)
- **Separate fact.** With the current script, the documented `60 c3435` does not converge
([6304057120](https://github.com/d6g8k5htny-coder/main/pull/458#issuecomment-6304057120)).
- **Limits.** The original command lines were never recorded, so this is a sufficient
invocation, not a log of what was run. The reported values (`t*`, `z0`, A, B/A, C/A, κ)
were never affected. The diagnostic is Anthropic, the same provider as the packet author,
with organizational-independence credit 0.
- **Non-Anthropic re-execution.** A nonauthor
[xAI Grok 4.7 re-execution](https://github.com/d6g8k5htny-coder/main/pull/458#issuecomment-6322666857)
(Cursor agent `bc-8084f8ef`; source-exposed, organizational-independence credit 0) ran
`c561d95b` at `55 c910` and `90 c3435` in the pinned environment. Each output differs from
the committed file only in its `time` line. It also confirmed that the committed files are
the review-math blobs.

**Additional fresh-seed evidence REPORTED; record applicability PASS (scoped).**
Separately, [OpenAI GPT-6's R2a report](https://github.com/d6g8k5htny-coder/main/pull/451#issuecomment-6279371506)
records a separate summation implementation with seed **31415927**, 192 points against
Expand Down Expand Up @@ -1289,9 +1335,10 @@ A nonauthor read from a non-Anthropic provider was requested, in four slices:
`CLAIMS.md` as unverified background reading; its pages were not read, and nothing here
claims that Edwards was verified. The docstring of `stats/rs_zeta.py` still names Edwards. That
file is left byte-unchanged so that the replay and R2a records stay bound to its blob
`8b97a1f6`. The replay's two residual-trail discrepancies remain unresolved, its own 13
timing-only streams stay unretained (their classification now rests on the retained re-run
above), and R2a's own raw JSON is unpublished; no record closes all of slice C.
`8b97a1f6`. The replay's two residual-trail discrepancies now have the exact reproduction
above. Its own 13 timing-only streams stay unretained (their classification now rests on the
retained re-run above), and R2a's own raw JSON is unpublished; no record closes all of
slice C.
- **D.** A wording audit of the non-claims. **Read: AMEND** (the same review, §10.1). **Delta
readback: PASS (scoped)** (review 5487707101, with slice A).

Expand Down Expand Up @@ -1319,7 +1366,9 @@ review checked the 59-command multiset, hashes, counts, environment, discrepanci
binding at exact replay head `78976ae7`. It did not independently reapply timing masks to 13
timing-only replay byte streams that were not retained. The replay used the same programs and
fixed inputs: it is not a second implementation or a scientific acceptance. Both residual
trails remain unexplained. Scientific effect stays NONE.
trails were later reproduced exactly with the earlier review-math script (item C). The original
command lines were not recorded, so that is a sufficient invocation, not a log. Scientific
effect stays NONE.

The packet was integrated into main as `3e589ae0` on 12 October 2026 by a third OpenAI dot /
Codex lane (`/root/integrate_zeta_exploration_packet`), distinct from both reviewer lanes. The
Expand Down