# Felisyn protocol 023: the confirmatory plaids protocol as declared and frozen (public copy)

The declared design of protocol 023, with every file its runs, its checks and its analysis depended on. This copy is dated 8 November 2026. Protocol 023 asked whether the plaids pilot's departure from a relay (protocol 022, [journal note 22](https://www.felisyn.com/journal/two-gratings-at-once)) holds on average over six fresh network instances of the same 2 mm model. It ended at its first instance, by its declared early stop, with the declared outcome **NOT INTERPRETABLE**. The receipts, the checks, the declared analysis and an exploratory diagnosis are in the package [early-stop-023](https://www.felisyn.com/research/early-stop-023/README.md) and in journal note 24.

- **Declared** at 2026-11-07T20:25:30Z, before any 023 run: the spec, SHA-256 `1e20a4a90a474d2a2f4348f095133a820120f0a4e85dc1095e6b32ec00cffac3`. [Journal note 23](https://www.felisyn.com/journal/the-model-moves-to-a-laptop) published that SHA-256 before any 023 run and said the text would follow with the result. Here it is `frozen/spec.md`, with 6 copies of the laptop's home folder path replaced (below).
- **Frozen** at 2026-11-08T00:46:40Z by the frozen manifest, here `frozen/manifest.json`, SHA-256 `6020fe1b7aaed781f40b90eadd7bb35da4f0de89c7f516ab7b4234a7f8ba1429`. It gives the size and SHA-256 of all 73 files. The freeze ran every test, and all exited 0: `test_analyze_confirm_023.py`, `test_calibration_023.py`, `test_check_instance_023.py`, `test_confirm_023.py`, `test_contract_023.py`, `test_end_to_end_023.py`, `test_freeze_023.py`, `test_queue_guard_023.sh`, `test_seed_search_023.py`, `test_sequence_023.py`, `bite_023.py`. The queue started 41 seconds later, and its launch record names this manifest's SHA-256 (`receipts/queue/launches.txt` in early-stop-023).
- **This public copy** holds the frozen package in `frozen/`, in its own layout. 56 files are byte-identical to the frozen manifest's entries. 17 differ only by the recorded replacements below, made because they hold the laptop's home folder path, its host name, a private temporary folder, a stand-in key header or internal wording. No file was added. The frozen package itself is kept unchanged in the project's private workspace.

Simulation only. The protocol ran one network instance of a 2 mm reduction of a published model of cat primary visual cortex, on an 8 GB laptop. It is not a biological measurement and not biological validation. No token, wallet, exchange, key or real order was involved.

## How to check this copy

1. `frozen/manifest.json` is byte-identical to the frozen manifest. Its SHA-256 is the value above, which the queue's launch record names.
2. 56 of its 73 entries match the files in `frozen/` byte for byte. They include the analysis and checker code that ran (`controller/`), the lead-in reference (`leadin-reference-023.json`) and the eight stimulus sequences (`sequences/`).
3. The other 17 files differ from their entries only by the replacements listed below. Putting the original strings back gives the frozen files: each token below occurs nowhere in the frozen originals, so the replacement can be reversed exactly. The builder checked that for every one of them.
4. `manifest.json`, at the top, is this copy's own manifest. It gives the size and SHA-256 of every file here, this README included, and lists every replacement with its lines and the SHA-256 of its frozen original.

## How to check the declared spec against note 23

Note 23 published the SHA-256 of the declared text, `1e20a4a9…cffac3`, before any 023 run. The published `frozen/spec.md` cannot hash to that value itself: its lines 12, 91, 319 and 397 held 6 copies of the laptop's home folder path, which this site never publishes, and they read `<home>` here. What can be checked:
- `frozen/manifest.json`, byte-identical and with the SHA-256 the queue's launch record names, pins `spec.md` at 136,558 bytes and SHA-256 `1e20a4a90a474d2a2f4348f095133a820120f0a4e85dc1095e6b32ec00cffac3`, with the status "DECLARED 2026-11-07T20:25:30Z.";
- the published copy has 136,534 bytes and SHA-256 `3c099133f31b000a37b2fdef78be5b888dcbaf36b80c8a87b7eeb79630a89671`, and differs from the declared text only inside those 6 paths;
- the original, byte for byte, is kept with the frozen package in the project's private workspace, so the published hash stays checkable against the record.

## Redactions

| File | Lines | What was replaced | SHA-256 of the frozen original (in `frozen/manifest.json`) | SHA-256 here |
| --- | --- | --- | --- | --- |
| `frozen/contract.json` | 13 lines, 120 to 1249 | 13 copies of the home folder's path, replaced by `<home>` | `c7d326d3f8dd0652b288ca81568be428576b86c34c94b7c86cbd0e540c14c8bd` | `6d67d8d0f27cf739ba1865d3b22bef965ff54bc3f52e528e5a73a4a74915560d` |
| `frozen/engine-patches/0001-lgn-noise-pair-order.patch` | 4 | internal wording in a header line, replaced as in the copy published in laptop-move | `4dc4b285f0260c1fb0e611ed24891d8198ed72b3d7a0135a3339109d1b6a5ec0` | `f5f424670680b165b12eda5e36485cd12af849cb7f6eee4de02101d7195b08ec` |
| `frozen/engine-patches/0002-block-size-shift.patch` | 4 | internal wording in a header line, replaced as in the copy published in laptop-move | `e0f189b27c007968f537143d02e1b56816dd2ecd3cfb474d00ed23ceff09019d` | `4b0bf47bdf3529752984dbbb8b75fb2dbb5488fc3a485366547e3fad1f59b245` |
| `frozen/freeze_023.py` | 60 | the home folder's path, replaced by `<home>` | `3a032dcd674ad31c1dcaafa125b16bc782877a4c15f7b09c3d59005da053ad1f` | `5bf0b3c1470e8dfaa6b0746ec6abd4b8b9c70b6ea6b5f62c145a44c0a7759486` |
| `frozen/make_contract_023.py` | 61 | the home folder's path, replaced by `<home>` | `d2d7f856ff3d0c9ca2f76e040f46ef9ce1e31bc829c0b35748f2e97af4aaa72c` | `053a34cf3a34bbf58dc6bddc68bc23e34bba97eb2096049ce11fbbd5e482bd4e` |
| `frozen/runtime-sha256.txt` | 491 lines, 386 to 876 | 491 copies of the home folder's path, replaced by `<home>` | `1f87b07baebd7f0133c831b8b70d4d008c92184123bf5899f96a4b9af9b0b2ad` | `0616c025919414d4b9a887270a66777c538daaa3334a4d74204859589d7e7ddd` |
| `frozen/seed-search-023-runs.txt` | 2 | the home folder's path, replaced by `<home>` | `b5747490779125ca2bab48d42cbd2daf439bee303b3ffe0395daf5bab3eadd6c` | `9c2f3980a3aeba15f51ece22a17ec0e3ef63d4b22b73bd68377740b720f488d2` |
| `frozen/server/fetch_023.sh` | 26 | the home folder's path, replaced by `<home>` | `3b799337c3a2bb27c73194f88e750c09f8af83deba2dfea816a52736d1b86354` | `48d90c726e8ede3f91948cb9880cf2a5e67353f566e722b2050f33fa286dda15` |
| `frozen/server/launch_023.sh` | 2, 59, 60, 61, 62, 63 | 6 copies of the home folder's path, replaced by `<home>` | `233516fe3004f2be5de6b83a6d2500bbcfaa3b8316bb2b9ca8bba1160a7e260a` | `dc0f492c8e17a9194dcf85cbe062eb47d569056d91737e13e0da2a692f80753b` |
| `frozen/server/run_queue_023.sh` | 64, 65, 66, 69 | 4 copies of the home folder's path, replaced by `<home>` | `c3f6273d5197d2c4499c1d6309b28359883c5dc2d7f9ffa98c403a6088ac9f32` | `51ab5a260cb5c6db06c77db5f2f9f4171eb271b0a8f61b64e5344e467bd027fd` |
| `frozen/spec.md` | 12, 91, 319, 397 | 6 copies of the home folder's path, replaced by `<home>` | `1e20a4a90a474d2a2f4348f095133a820120f0a4e85dc1095e6b32ec00cffac3` | `3c099133f31b000a37b2fdef78be5b888dcbaf36b80c8a87b7eeb79630a89671` |
| `frozen/test_end_to_end_023.py` | 16 | the home folder's path, replaced by `<home>` | `4835e5fecb268fedcc0b2fae0fed81a8b388fd2f7d5e66906c0ac824828e9091` | `894aaff0e90a90a8ac9bd7cbbf204fc7fea7c88930d41e8e01f865676f4b5d62` |
| `frozen/test_queue_guard_023.sh` | 44; 739, 844, 1004; 989; 22 lines, 199 to 991 | a private temporary folder in `JOBTMP`, replaced by `redacted-private-tmp` (the test then uses its own temporary folder, as it does wherever that folder is absent); 3 copies of the home folder's path written with escaped slashes in a bash pattern, replaced by `<escaped-home>`; 4 copies of the home folder's path flattened into upload file names, its slashes written as underscores, replaced by `<underscored-home>`; 32 copies of the home folder's path, replaced by `<home>` | `7ec85f44afc28d2823d5cc18a034b73ab4ac4f3b2928975cd77b9f55b1313d82` | `a8a82ea454c881a41b761be3d8500ba85b8df667eecfebc350c85e3b7244a049` |
| `frozen/test_seed_search_023.py` | 40; 359 | a private temporary folder in `JOBTMP`, replaced by `redacted-private-tmp` (the test then uses its own temporary folder, as it does wherever that folder is absent); a stand-in key header that the test writes as bait into a temporary file the seed search must never open; the site's privacy check refuses that header, so it reads `<stand-in key header, redacted>` (the search decides from names and modes, never from content) | `972d244ce6524bff98d1e9941c0fff6841c4bb439eff0a73a0fb1abbc00fbf85` | `d3d94c980d9f63bb75c26616017912bb8c4b66e96116a4c12a0b904c5d657a97` |
| `frozen/testkit_023.py` | 48; 51 | a private temporary folder in `JOBTMP`, replaced by `redacted-private-tmp` (the test then uses its own temporary folder, as it does wherever that folder is absent); the home folder's path, replaced by `<home>` | `4617ea510d898666d978bd1adaa3ffffdbedeb31cfd83e4c667780b74135623f` | `b0e94cb777374953ade6ba4c234c849d9c95fbae3aa6a060e14f6541b37a58e6` |
| `frozen/tier2/receipt.json` | 4; 41 lines, 6 to 2191 | the laptop's host name in field `host`, replaced by `redacted-host`; 41 copies of the home folder's path, replaced by `<home>` | `d469c8eef82e1209042af5983aee6e83d1b719e7771c6291232831b36eac4740` | `c6e5de7485806f95a5b38d7cea2ee6920a3cef3a370e8e3d31f55c9502807c03` |
| `frozen/tier2/tier2.log` | 12 | the home folder's path, replaced by `<home>` | `12baf3fa03b4fa2f9ab7b530a9c84ae943cbb1e5a98b3858788000a2a7b20130` | `ed0ff089826248bb3825271a5e0d25ef47cca26ad60803c0ee31556d5e6491cd` |

- **The tier 2 receipt.** Its host name and its paths are replaced. Nothing else in it changes.
- **The redacted tests** (`test_end_to_end_023.py`, `test_queue_guard_023.sh` and `test_seed_search_023.py`, with their helper `testkit_023.py`) ran in their original form at the freeze, as the frozen manifest records.
- **The engine patches.** Their redacted copies are byte-identical to the ones published in [laptop-move](https://www.felisyn.com/research/laptop-move/README.md), which made the same replacement.
- **Words that look like secrets, left in place.** The seed search (`seed_search_023.py`), its records and its tests, the contract builder, the test breaks and the spec name the kinds of files the search never opens: key, keystore, wallet and environment files, `.ssh` and `devnet-keys` folders, the site's `.vercel` folder. The tests write harmless stand-ins under such names (for example `ssh-ed25519 AAAA pynn_seed 10`) to check that the search never opens them. None of these is key material. The one stand-in that the site's privacy check refuses, a key header, is replaced above. `runtime-sha256.txt` names macOS's own `AppleKeyStore` framework among the system libraries it pins. Every such word was reviewed one by one.
- **Private names left in place.** The spec, the frozen manifest and several scripts name private project records by their paths in the private workspace (under `work/`, `docs/` and `logs/`, also in the frozen manifest's `source` fields) and refer to the owner and the owner's agent. Those records are not published. The files that name them are pinned by the frozen manifest, so they are left unchanged.

## What happened

The queue ran the first slot, pynn_seed 7, on 8 November from 00:47:21 to 01:53:28 UTC. The run ended normally. The blind check right after it, the server check (which since the move to the laptop runs on the laptop), refused the export on one kind, the LGN lead-in identity: 1,799 of 1,800 X_OFF trains equal the pilot's reference, and one cell, 2738, differs. By the declared early stop, no further run started, the local check refused the same export, and the declared analysis wrote the outcome NOT INTERPRETABLE. Seeds 8 to 14 were never built. The full record is in early-stop-023.

## Found after the run: two sentences of the spec

The exploratory diagnosis after the outcome, in early-stop-023, found that a full-length run on this laptop can differ from an exact repeat of itself: one of four such runs did (measured). With four runs the rate is close to unknown: a 95 % interval runs from 0.6 % to 81 % (inferred). Two parts of the spec read differently in that light. They are left as declared.
- **The determinism sentence.** The spec says of heat and other work that "the simulation is deterministic, so they change only the time". It was written about run time, and its premise does not hold for every run on this laptop.
- **Risk 14.** The spec states the cost it accepted: "A fault of that one instance (an export damaged as it was written, say) would then end the protocol where the replacement rule could have replaced it; the rule accepts that cost for the server time it saves when the fault is shared." The measured fault rate makes that cost higher than the design expected.

## What is in `frozen/`

| Part | What it is |
| --- | --- |
| `spec.md` | The spec as declared (redacted, above). |
| `contract.json` | The contract of the runs, the checks and the analysis, built from the spec by `make_contract_023.py` (redacted, above). Its frozen SHA-256 is `c7d326d3…`. |
| `manifest.json` | The frozen manifest, written by `freeze_023.py`. |
| `sequences/` | The eight declared stimulus orders, one for each of pynn_seed 7 to 14, built by `make_plaid_sequence_023.py` with protocol 022's `make_plaid_sequence_r2.py`. |
| `controller/` | The rule, categories and fixed texts (`confirm_023.py`), the analysis driver (`analyze_confirm_023.py`), the blind checker with the early stop, the replacement rule and the summary (`check_instance_023.py`), and protocol 022's `plaid_stats.py`, `analyze_plaids.py` and `frozen017.py`, byte-identical. |
| `leadin-reference-023.json`, `make_leadin_reference_023.py` | The LGN lead-in reference, every X_ON and X_OFF cell's spike times before 504 ms in the pilot's intact export, and the script that extracted it. |
| `protocol-017/` | Byte-identical copies of protocol 017's contract, decoder, manifest, model-file list and frozen controller code. |
| `server/` | The run folder's scripts: the queue, the launcher, the fetch, the host shell, the runner, its LGN input cache and stimulus version 3 (byte-identical to protocol 022's), the memory wrapper `run_controller_mem.py`, and the tier 2 check. |
| `engine-patches/` | The two engine patches: the LGN noise pair order and NEST's block size. |
| `runtime-sha256.txt` | The hash list of the engine's native files on the laptop, made from the final build. |
| `tier2/` | The tier 2 check's inputs, its passing receipt and its log. |
| `calibrate_023.py`, `calibration-023.json`, `calibration-023.log` | The calibration of the decision rule on synthetic instances. |
| `analysis-choices-023.txt` | Details the spec leaves to the code. |
| `seed_search_023.py` and `seed-search-*.txt`, `seed-census-server.txt`, `window-census-server.txt` | The read-only seed search and its records, which showed seeds 7 to 14 unbuilt before the declaration. |
| `freeze_023.py`, `assemble_023.py`, `bite_023.py`, `testkit_023.py`, `test_*.py`, `test_queue_guard_023.sh` | The freeze, the package assembly, the tests and their deliberate breaks. |

## Credits and licences

- **Model.** Antolík J, Cagnol R, Rózsa T, Monier C, Frégnac Y, Davison AP (2024), *A comprehensive data-driven model of cat primary visual cortex*, [PLOS Computational Biology 20(8): e1012342](https://doi.org/10.1371/journal.pcbi.1012342), published under [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/). The run used its code, LSV1M v1.0 (MIT license), with the command-line overrides of the 2 mm reduction. That code stays on the laptop, and none of its files is copied here.
- **Mozaik** (CeCILL version 2.0) and **PyNN 0.11.0** (CeCILL version 2.0): the simulation framework and its interface to the simulator. No file of either is copied here.
- **NEST 3.4** (GNU GPL, version 2 or later) and the [NEST step current module](https://github.com/CSNG-MFF/nest-step-current-module/tree/fe91c0345a602d564a74ebe0ea2c66d9a4157005) (GPL): the two engine patches change one file of each, and are offered under the terms of the files they change. **GSL 2.7.1** (GPL) was rebuilt from source; no GSL file is here.
- **This work.** The spec, the contract, the scripts, the tests and the patches are Felisyn's own.
