# Audit — 2017 beep-WAV transcript and residual tail ## Scope This audit starts from the preserved 2017 textual transcript, not an unrecovered audio file. It verifies the duration-token mechanics, the A/x onion decode, and the limits of the remaining token tail. ## Source and preservation status - Transcript/table source: `archive/recon/audio-files/03_beeps-wav_OCTAL-TRANSCRIPT-and-letter-table-VERBATIM.md` - Source SHA-256: `17ec93163e486d15b705b87833959c2e4795e6cf8ef2f0f3cf1c9cf1404b9cff` - Contemporary public solver record: - Public-record boundary: The March 2017 discussion retains the reported Instaudio route and duration transcript, but it is a solver record rather than the lost WAV bytes. - Original Instaudio object: `https://instaudio.s3.amazonaws.com/live/private/7fa2c325c0a7395c8856c5bb4f11b5616c6f60c5.wav` - Object status: Current S3 object is a preserved 319-byte NoSuchBucket stub; original WAV bytes remain unavailable. - Public-CDX check: 2026-07-27 returned `[]` for the exact object and `[]` for the object prefix. The archive check is evidence of a failed recovery attempt, not proof that no private browser cache or unindexed mirror exists. ## Mechanical reproduction - `2994` base-9 digits on `60` preserved lines. - Splitting at runs of `0` yields `710` non-empty packets. - All `27` documented token classes map; unknown tokens: `[]`. - The published meta-transcription matches exactly: `True`. - Its structure is three identical blocks plus `djhedjhedjh`; token-label first occurrences are `abcdefghijklmnopqrstuvwxyzA`. - The repeated A/x region has `176` bits and decodes MSB-first to `666666m7x6x5regc.onion`. ## The sentence and the correction | Candidate second word | One-to-one substitution status | Result | |---|---:|---| | `MIND` | `False` | Fails because `g` and `j` both map to `N` once KNOWS / WHEN / OPEN are applied. | | `BIRD` | `True` | Consistent with a one-to-one mapping and semantically apt, but not uniquely forced by the token pattern. | `BIRD` is a valid and semantically apt contemporary correction, but its validity does not uniquely select it from every possible four-letter word consistent with the residual substitution positions. ## What `rssstuvwx` actually establishes The repeated pre-A/x token run is `abccadefbgheijklmelndjecndeopcdebmekqdjerssstuvwx`. The decoded sentence plus its terminal divider occupies `abccadefbgheijklmelndjecndeopcdebmekqdje`; the exact remaining tail is `rssstuvwx`. It contains no current word-divider token `e` and has equality pattern `ABBBCDEFG`. The letters `r` through `x` are not author-supplied alphabet characters. Norbert's table assigns each distinct duration packet its next Latin label in first-occurrence order. All seven tail labels first occur inside this tail; their raw base-9 packet runs, in order, are `1141, 1212, 1212, 1212, 12, 15, 2121, 8, 141`. Alphabet positions, Caesar shifts, or ASCII values applied to `rssstuvwx` would therefore operate on the transcriber's canonical labels, not on a source-selected character stream. The source supplies no numerical packing or delimiter rule for those packet runs beyond their role as distinct tokens. Under the extra assumption of a single injective substitution completed from the `BIRD` reading, `rstuvwx` are still unmapped and may draw from `cfjmquvxyz`. That leaves `604,800` possible assignments before any external language model, key, or author instruction is supplied. This does **not** make the tail meaningless; it shows that the surviving transcript alone does not determine a plaintext, password, or next action. A claim to decode it must add a source-led codebook or a recovered primary artifact and then pass a control, rather than choose a convenient word.