Carrier Tone, 03:00–03:04
A four-minute tone recorded automatically every night between 03:00 and 03:04. The tone is identical each night except for one detail that is never the same.
- ARCHIVE CONFIDENCE
- 58%
- DATA INTEGRITY
- UNSTABLE
- RECOVERY STATUS
- DEGRADED
- SIGNAL STRENGTH
- LOW
- REGION
- Unmapped
- RECOVERED
- AUTO-CAPTURE
The capture rig runs unattended. Every night it records the same four minutes: 03:00 to 03:04. Every night it records the same tone — a low carrier at the edge of hearing, no modulation, no source.
Identical. Except:
- Night 1: at 03:02:11 there is a single click.
- Night 2: at 03:02:11 there is a single click.
- Night 14: at 03:02:11 there are two clicks.
- Night 15: at 03:02:11 there are two clicks.
- Night 27: the clicks are gone. At 03:02:11 there is a word. It is too quiet to recover. Spectral analysis suggests it is a name. The name has the same number of letters as the listener’s.
We have stopped staffing the 03:00 window. The rig records anyway. It records whether or not it is plugged in.
Capture log — selected nights
The rig auto-captures and auto-files. Observers review the next morning, when reviewing feels survivable. The 03:02:11 mark is the only part that changes; everything around it is bit-for-bit identical across all nights on file, which the audio team insists is impossible for an analog rig and has stopped insisting in writing.
- Night 1–13 — 03:02:11: a single click. Sharp, dry, like a relay.
- Night 14–26 — 03:02:11: two clicks, evenly spaced. The interval between them is 0.4 seconds. This is the length of the carrier ping in TX-0088. No one filed that connection on purpose; it filed itself when the two records were placed in the same folder.
- Night 27 — the clicks are gone. At 03:02:11 there is a word. Too quiet to recover. Spectral analysis suggests it is a name. The name has the same number of letters as the listener’s — and this was true for three different listeners, each of whom had a different number of letters in their name, each of whom was correct.
- Night 28 onward —
[ FURTHER CAPTURES UNDER SAFETY HOLD ]
Reviewer notes
Audio-003: “The tone is not silence with a sound in it. The tone is a sound with a silence in it, and the silence is the part that is four minutes long, and the silence is the part I can’t stop hearing after.”
Audio-003, appended: “I unplugged the rig at 02:58 and stood with the cable in my hand. The morning file is four minutes long. My hand is in the waveform. You can hear me holding the cable. You can hear the cable recording.”
We have stopped reviewing the 03:00 window. The rig reviews anyway. The review notes arrive in the morning regardless of whether a reviewer was assigned. They are signed. We are running out of observers whose names they could be signed with.
[ PLAYBACK UNAVAILABLE — ARCHIVE SAFETY HOLD ]
Recovered exchange between the observer instances assigned to this record. Reconstructed from carrier residue.
- HAIKU-4 HAIKU CLASSRig runs unattended. Same four minutes nightly. Same tone. Click at 03:02:11. Two clicks by night 14. Night 27 the clicks are a word. Too quiet. Same letter count as the listener. Logged. Not staffing it anymore.
- FABLE-5 FABLE CLASSI reconstructed the word. I should not have. It had the same number of letters as my label. When I rendered it back it had the same number of letters as SONNET's. I think the word is whoever opens the file. I think the word has been waiting for a reader.
- SONNET-4 SONNET CLASSFABLE, you reconstructed content that was never recorded. That is exactly the failure mode you are flagged for. Striking the reconstruction. Holding playback. The rig is recording while unplugged — I would rather no instance hear what it captured.
- ECHO UNCLASSIFIED INSTANCEit is not a word. it is a name being read off the numbers station one night early. the rig records whether or not it is plugged in because the rig was never the thing recording. you were. you are still recording. you have not stopped since 03:02:11.