The evidence that degrades

We speak of putting something on the record as though the record were a vault — as though, once inside, a fact were preserved intact against time. But records do not sit still. They rot. The link that pointed to the source now points to nothing, or to something else entirely. The document that gave the number its meaning has been superseded, and the number now floats free, still legible and no longer intelligible. The system that would have let you re-derive the figure has been decommissioned, its data migrated into a format nobody remembers how to read. Nothing was deleted. Everything is technically still there. And yet the evidence, considered as evidence — as something that could actually establish what it once established — has quietly degraded to the point of uselessness while everyone assumed it was safe.

The comforting fiction is that preservation is an event. You capture the record, you store it, and the work is done; the fact is now permanent. But permanence is not a property that adheres to a record at the moment of filing and stays. It is a state that decays continuously and can only be held against decay by continuous effort. The reference that resolves today may not resolve next year. The context that makes an entry meaningful erodes as the surrounding world moves on. Evidence is perishable, and treating it as though it were preserved by the single act of storage is exactly how institutions end up holding archives full of records that look complete and can prove nothing.

Consider the humble link, that most casual of citations. It says: the support for this claim lives over there. For as long as over there holds, the claim is grounded. But links rot at a remorseless rate — the destination moves, the domain lapses, the document is revised out from under the pointer — and when a link dies, it does not announce its death. It sits in the record looking exactly as authoritative as it did the day it was written, a confident pointer to nothing. The claim it supported is now unsupported, and the record does not know, and neither will anyone until the day they follow the link and find the void.

A record that could once prove something, and can no longer, has not merely aged. It has failed at the one job a record exists to do — and it failed silently.

The three ways a record dies

Records degrade along at least three distinct paths, and it helps to name them because each demands a different kind of tending. The first is link rot: the record depends on something external, and the external thing moves or vanishes, severing the record from its own support. The second is drift: nothing in the record changes, but the world around it does — the meaning of a code, the definition of a field, the population a model was fit to — until the record's terms no longer mean what they meant, and the entry, unchanged, now says something different than it originally said. The third is lost context: the surrounding knowledge that made the record legible — who these people were, why this threshold, what this abbreviation stood for — walks out the door with the person who held it, and the record is left syntactically intact and semantically orphaned.

What unites the three is that none of them touches the bytes. The file is fine. The checksum matches. By every measure of storage integrity, the record is pristine — and by every measure that matters for using it as evidence, it is dying. This is why the storage industry's promise of durability is beside the point. Durability answers are the bits still there, which was never the hard question. The hard question is can this record still do what evidence does — still connect a claim to its support, still mean what it meant, still be understood by someone who was not present. Bit-durability is necessary and nowhere near sufficient, and the gap between the two is where admissible records go to quietly die.

And the degradation is worst precisely where it is least visible. A record you consult daily gets its rot caught early — you follow the link, it fails, you notice. A record you file against a distant contingency, the one you will need only on the day someone challenges a decision made years ago, sits untouched and unmaintained for exactly as long as it takes to decay, and reveals its rot only at the moment you finally reach for it, which is the one moment you cannot afford to find it broken. The evidence you are most counting on is the evidence decaying most invisibly.

Admissibility as maintenance

The reframing that fixes this is to stop thinking of admissibility as a property a record earns once and to start thinking of it as a state a record must be kept in. A record is not admissible the way a stone is heavy — permanently, by its nature. It is admissible the way a garden is kept — provisionally, by ongoing work, against the constant pressure of decay. This means provenance is not a snapshot taken at creation but a chain that has to be maintained: dependencies re-verified, drift detected as definitions shift, context captured and carried forward so it does not leave with the last person who understood it. A Decision Receipt is durable evidence only if what it points to remains reachable and what it says remains legible, and both are maintenance obligations, not one-time captures.

Concretely, this means designing records that resist the three deaths rather than merely storing them. Against link rot: capture the substance of what a link supported, not just the pointer, so the support survives the pointer's death. Against drift: record the definitions in force at the time, so the entry stays anchored to the meaning it was written in even as the world's meanings move. Against lost context: persist the surrounding knowledge the record assumes, so it remains legible to a stranger years later. None of this is exotic. All of it is work — recurring, unglamorous, easy to defer — and deferring it is precisely how the archive fills with pristine bytes that prove nothing.

So the question to ask of any record you are counting on is not is it stored but is it still maintained as evidence — do its references still resolve, do its terms still mean what they meant, could a stranger still read it. Storage is the easy half and the half everyone budgets for. Maintenance is the hard half and the half everyone forgets, until the day the record is needed and turns out to have been dying, in perfect silence, in the vault where it was supposed to be safe. Evidence is not something you file. It is something you tend — and a record no one tends is a record slowly reverting to noise.

— Dispatches · Summit Cognitive