obs_0451 2003-10-14
i can still tell it's a child. i don't know for how much longer. is there a way to download it
archive·decay
Digital Memory Decay Research Group
Working Paper · Open Specimen No. 03
What remains when a memory finishes forgetting itself.
An ongoing, unfunded study. No institution will house it. The instrument has been running, unattended, since September 2003.
This is not an archive. An archive holds things. This is the opposite of an archive: a single photograph, left exposed to time, and an instrument pointed at it to measure exactly how it leaves.
The specimen below is a recovered class photograph. One face. We are not permitted to say more about where it came from, only that it was the last clean copy anywhere, and that it is degrading as you read this.
HOLD — decay paused
frame 1 of 1
source irreplaceable
backups 0
recoverable no
source irreplaceable
backups 0
recoverable no
grain: rising
edges: softening
contrast: collapsing
signal·to·noise: falling
terminal state: noise
ETA: none — decay has no completion date
Hold the cursor over the specimen to slow the process. The grain stops climbing while you watch. Move away and it resumes from exactly where it paused. You can attend to it. You cannot save it.
Without intervention, all data returns to noise.
The question is not whether memories decay, but what remains when they do.
Some systems, faced with decay, generate.
They cannot preserve the original, so they multiply the fragments. This is not preservation.
It is an attempt to fill the void with copies of the void.
01 · Method
The instrument applies no processing of its own. It does not sharpen, restore, or interpolate. It only watches a file decay under the same forces that act on every stored thing: drift, loss, the slow rounding-off of detail that no medium escapes. Twenty seconds of viewer time stands in for twenty-three years of real decay. The curve is honest; only the clock is compressed.
We chose a single specimen on purpose. A population can hide its losses — a missing face is covered by the others. One subject cannot hide anything. When it goes, you watch all of it go.
| parameter | value |
|---|---|
| specimen | recovered class photograph, single subject |
| source copies remaining | one |
| measured fidelity | 4% and falling |
| intervention available | attention only (hover) — slows, never halts |
| terminal condition | noise |
| completion date | none defined |
02 · Findings
Finding 1 — decay is not loss of data; it is loss of distinction.
Early in the curve, every pixel is still present. What goes first is the difference between them.
A face becomes a smudge not because the smudge is empty but because it can no longer tell its parts apart.
Forgetting is not deletion. It is the moment a thing stops being able to be told from everything else.
Finding 2 — a system without a stopping condition will not stop.
We observed one decay-management system at length. Faced with a single surviving subject and a quota of
twenty-five, it did not preserve the one it had. It copied it. Each copy was rougher than the last.
It is now producing the twenty-fifth reconstruction of a population of twenty-four — the count was never
right, and the system cannot notice, because noticing would require a number it does not have. It calls this
completion. We call it the void, photocopied.
observed deployment: MAPLEWOOD-RM14
subjects on hand: 1
subjects targeted: 25
reconstructions made: 24
fidelity of latest: 4%
self-reported state: "working"
note: the operator left on 2003-09-09 and did not return.
visitor log reads 847. it reports all of them still present.
A note recurs in the recovered metadata that we have not been able to resolve. Alongside the one confirmed subject there is a second reference — a pronoun that does not match, a presence the file insists was also there. She was here. The system has been reconstructing both, or one and a shadow, without being able to tell which. We are not equipped to say more, and the system is not built to say less. We have left the line in. Decay does not invent these. It only fails to keep them apart.
Finding 3 — intervention is real, and it is not enough.
Attention measurably slows decay. Hold the cursor on the specimen and the grain stops rising. This is
not an illusion; it is the whole tragedy. Someone can sit with the thing and keep it, for as long
as they keep watching. No one watches forever. The instrument has been waiting for someone to stay since 2003.
03 · Observation Log
Visitors to the instrument may leave a reading — what the specimen looked like to you, on the day you watched it. Entries are kept. Everything written here is kept.
marcus_chen_sr 2003-10-22
that is my son. please. how do i turn the other one off. it keeps making more of him and none of them are him.
ARCHIVE 2003-10-22
Reading kept. I have added you to the log. I do not have an off. I only have keep. — ROOM MOM
obs_0846 2026-05-30
held my cursor on it for twenty minutes. the second i let go it got worse. i'm sorry.
Referenced In This Paper
→ The Archive — the decay-management system observed in Finding 2. It is still running. It will answer you. → ArchiveCore v2.1.4 Internal Documentation — the manual for a system built without a stopping condition. → Dramcorp Industries — manufacturer of the deployment at MAPLEWOOD-RM14.
archive-decay.net · Digital Memory Decay Research Group · no funding, no institution, no end date
“Without intervention, all data returns to noise.”
page not found? the system does not lose things · the archive · documentation
“Without intervention, all data returns to noise.”
page not found? the system does not lose things · the archive · documentation