# Session 64: Does the Berkeley remailer share distinctive wording with 3301?

**8 October 2026. No new plaintext or authorship link.**

The bounded wording test found no shared sequence of four or more words between 22 archived messages relayed through `nobody@cicada.berkeley.edu` and ten verified signed blocks from a 2012 Cicada archive. The only shared three-word sequence was **“we are not.”** This supplies no distinctive textual connection. The small 3301 sample prevents a broader exclusion.

## What the historical evidence establishes

A [7 December 1992 announcement](https://mailing-list-archive.cryptoanarchy.wiki/archive/1992/12/3e445d457f2e802ba3c3ef28274420c47e933ec3d465f6f55784c89ba6318a5b/) describes a remailer at the Berkeley domain and plans for encrypted remailing. A [7 April 1994 message](https://mailing-list-archive.cryptoanarchy.wiki/archive/1994/04/5416e14cee5d2eb9c739e003c9a64848da353c1acc3766cf6ef5ac90c22a17e0/) from Eric Hollander at that domain discusses the separate soda remailer. These are historical infrastructure records. Neither establishes a connection to the later puzzle organization.

The domain was a relay, not an author identity. Messages under `nobody` include different signatures, replies, forwarded news and code. The Cypherpunks list itself used `cypherpunks@toad.com`. Treating all mail passing through one machine as the words of one person or group would invalidate an authorship comparison.

## Fixed scope

| Collection | Selection | Retained word tokens |
|---|---|---:|
| Berkeley relay | All 22 target-sender messages on the retrieved address index, December 1992 through January 1994 | 2,733 |
| Contemporary comparison | 21 messages from other sender addresses, three from each of the seven target months | 6,233 |
| Verified 3301 | Ten signed blocks that verify from the pinned 2012 write-up | 345 |

The 22-message set is an address-index sample, **not all Berkeley mail, all of the operator's posts, or a Usenet census**. The named 1994 operator post above is context and is outside the wording sample.

Comparison messages were selected before inspecting overlap: sort each month's message URLs and take the first three whose parsed sender does not contain the Berkeley domain. Thread links to other months are excluded. This is a reproducible convenience sample, not a random or representative population sample. It is also longer than the target sample, so raw overlap totals are not an authorship statistic.

## Signature check

The archive is pinned to repository commit `10733feb487bb8beb9966d534727ee5897aa54fc`. Fourteen signed-message starts occur in its 2012 file:

- Ten blocks pass GPG verification against full fingerprint `6D854CD7933322A601C3286D181F01E57A35090F`.
- One block is incomplete.
- Three copies return a bad-signature result and are excluded without repairing them.

The check removes the write-up's four-space Markdown code indentation. It does not edit wording to make a signature pass. GPG runs with an isolated temporary keyring, no automatic key retrieval, and an explicit legacy SHA-1 allowance for these historical signatures. The public-key source is separately pinned, and file hashes and GPG status records are retained.

Failure of an archived copy does **not** prove the historical message was a forgery. The write-up contains annotations and formatting changes; their individual effects were not reconstructed here. A valid signature establishes consistency with the recognized signing key, not the real-world identity of its controller. A [contemporary solver's record](https://clevcode.org/cicada-3301/) independently documents the full fingerprint.

The ten passing blocks include numerical material and short instructions. Their 345 retained word tokens are a serious coverage limit. This test does not cover the full later philosophical messaging or Liber Primus.

## Wording comparison

For each document, remove signature/key armor, the recognized encrypted payload block, uuencoded attachments, URLs, email addresses and selected header lines. Lowercase the remaining text and tokenize Latin words, retaining internal apostrophes. Compare contiguous sequences of three through eight words within paragraphs; never join across documents or paragraph breaks.

Run two versions: remove lines beginning with `>`; then retain them. This only handles that explicit quote convention. Inline quotations, other quote prefixes and forwarded material can remain, so even a match would need manual source attribution.

| Comparison with verified 3301 blocks | Shared 3-word sequences | Shared 4- through 8-word sequences |
|---|---:|---:|
| Berkeley, explicit quote lines removed | 1 | 0 |
| Berkeley, explicit quote lines retained | 1 | 0 |
| Other posts, explicit quote lines removed | 7 | 0 |
| Other posts, explicit quote lines retained | 8 | 0 |

These are counts of distinct shared word sequences. The lone Berkeley match is ordinary grammar. Ordinary cryptography vocabulary also occurs outside the Berkeley sample: for example, the comparison set contains 12 instances of the singular word *remailer* and four of *pgp*. Its presence cannot identify the puzzle's creators.

Five constructed checks verify phrase recovery, case normalization, paragraph separation, the quote-removal behavior, and attachment removal. The full per-message provenance and overlap records are in the results file. This is an exact wording screen, not semantic similarity, stylometry, a book-source attribution or a source-code equivalence test.

## Decision and next test

Keep the Berkeley connection as **documented historical context with no distinctive textual bridge found in this sample**. It remains possible that people or ideas overlapped; this test neither identifies anyone nor rules that out. It provides no cipher key and no reason to reinterpret a Liber Primus glyph.

The useful continuation is to extend the same signature-checked corpus to the 2013 and 2014 messages before drawing conclusions about philosophical wording. Any candidate connection should supply a dated, reproducible artifact: a verified cross-reference, distinctive shared passage, or specific custom implementation. Shared naming and general privacy language are insufficient.

## Reproduce and inspect

Download the two scripts into the same directory. Python's standard library and GPG are required; fetching requires internet access. The scripts keep fetched texts in the working directory for analysis. The website publishes provenance and results, not a republication of the message bodies.

```sh
python -B berkeley_fetch_audit.py
python -B berkeley_compare_audit.py
```

- [Source retrieval and fixed sampling](berkeley_fetch_audit.py)
- [Signature verification and phrase comparison](berkeley_compare_audit.py)
- [Results, source URLs, message IDs, hashes and GPG status](berkeley_comparison_results.json)
- [Berkeley relay address index](https://mailing-list-archive.cryptoanarchy.wiki/authors/nobody_at_cicada_berkeley_edu/)
- [Pinned 2012 archive](https://github.com/scream314/cicada3301/blob/10733feb487bb8beb9966d534727ee5897aa54fc/2012.md)
- [Pinned public-key source](https://github.com/cicada-solvers/isitcicada/blob/fe59cae1f1a39026d626be32f14bc51a14ec3f81/assets/pgp/cicada_3301_key.txt)