The Half-Life of a Note

In the spring of 2026 an agent walks a house in Charlotte and writes one line in the file.

Staining under the upstairs bathroom, seller says it was fixed in 2021, no invoice.

That sentence is doing three separate jobs and nobody separates them, because the field does not ask. It records something the agent saw. It records something the seller said. And it records a gap, which is the absence of any document confirming either.

Three years later the house comes back to market with a different agent, and the note is still in the file. The seller who described the repair has moved to another state. The agent who saw the staining is at another firm. What survives is the shortest and most damaging version of what was known, stripped of who observed it, how they knew, and whether it was ever true.

It is now a fact about the house. Nobody asserted it. It outlived everyone who could have qualified it.

Fields are a theory of what matters, written in advance

Every record keeps what its fields were built to hold, which sounds obvious and has a consequence almost nobody designs around.

Somebody decided, before any of this information existed, which facts a property would have. Bedrooms. Bathrooms. Square footage. Year built. Those were reasonable choices and they encode a claim: that these are the things worth knowing about a house.

Everything outside that list survives only in whatever free text field exists, and free text is where structure goes to die. It cannot be searched across properties, counted, compared or checked. It is preserved and unusable at the same time, which is a particular kind of loss because it looks like retention.

So the knowledge that fits no field is not rejected. It simply never enters in a usable form, and its absence leaves no trace. Nothing in the record says here is where we stopped recording.

That is survivable while the person who knew is still nearby. It stops being survivable when the record outlives the knower, because then the record is all there is, and the record is silent about what it never held.

Schemas are conservative because changing them is expensive

The obvious response is to add fields. In practice this is close to impossible at any speed.

A national data standard is an agreement between hundreds of organisations. Adding a field means every software vendor rebuilding an integration, every participant updating a process, and years of partial adoption during which the field exists and is half empty. So standards lag practice by a decade or more, and the gap between what is known and what is recordable widens continuously.

There is a concrete instance arriving this autumn. On 2 November 2026 the American residential appraisal form is replaced for loans sold to Fannie Mae or Freddie Mac. The forms being retired have been in service close to twenty years.

The replacement is better in most respects. Free-form commentary moves into structured fields and defined sections. Appraisers gain room-level condition ratings, energy efficiency features and disaster mitigation improvements as proper fields rather than sentences in an addendum. All of that is more checkable than a paragraph, which is the whole argument for it.

It is also a trade, and worth naming as one. The appraiser gains fields and loses the blank space where an explanation used to go. More of what they observe is captured as data. Less of their reasoning is captured at all.

What gets kept becomes what is true

This is where the problem stops being about property.

A record is not a neutral container, as a recorded sale price already demonstrates. It is an argument about what counted, made by whoever designed it, and it becomes more authoritative as everyone who could contradict it moves on.

Historians have lived with this for a long time. We reconstruct the ancient world disproportionately from tax records, legal documents and grave goods, not because those were the most important things about those societies but because those were the things somebody thought worth recording and the things that happened to endure. Two filters, both severe, and everything downstream inherits the result.

The difference now is that we are building the filters deliberately, at speed, and machines will read what comes out as the complete account rather than as the residue of a series of design decisions.

Decay is a harder problem than provenance

Provenance is the easier half. Who observed this, when, and on what basis. That can be a field, and adding it would fix the Charlotte note, because the reader would know that the staining was observed and the repair was reported.

The other half cannot be solved the same way, because the same field holds notes that behave completely differently over time.

The half-life of a note is how long the thing it records stays true. A field records the note. It does not record the half-life, so a note that survives long enough is read as current by default.

Consider three notes written on the same day, in the same field, in the same format. Water damage seen during a showing. A shared driveway with the property next door. A sale that went between family members and never reached the open market.

The first expires, probably within a year or two. The second never expires. The third describes a completed event and is permanent. Same shape, three behaviours, and in 2029 all three read as current.

Which means accuracy degrades even when nothing in the record changes. The database is not wrong on any given day. It is wrong by an increasing amount, silently, and no data quality process is looking for it, because nothing about the data is malformed. Everything is correctly formatted, correctly stored, and quietly out of date.

How long does information stay accurate in a database?

It depends entirely on what the information describes, and almost no system records that. A note about condition can expire within a year. A note about a boundary or an easement never expires. A note about a completed transaction is permanent. All three usually sit in the same field with the same date format, so a reader three years later has no way to tell which kind they are holding.

What is data provenance and why does it matter?

Provenance is the record of where a piece of information came from: who observed it, when, and on what basis. It matters because the same sentence can be a firsthand observation, something reported by another party, or a judgment, and those carry very different weight. Written without provenance they look identical on the page, and read years later all three become facts.

Why do records become less reliable over time?

Because information ages at different rates and records store it uniformly. Nothing has to change or corrupt for a database to become less accurate. It simply contains an increasing proportion of statements that were true when written and are no longer true now, with nothing distinguishing them from the statements that remain true.

What would fix it

Not recording more. Recording two things about each note that we currently do not, both decided at the moment of writing by the person best placed to decide.

Where it came from. Observed firsthand, documented with a source, or judgment. Three values. This separates I saw staining, the seller told me about staining, and I suspect there is a damp problem, which are three different claims currently written identically.

Whether it can change. Fixed or perishable. Two values. A shared driveway is fixed. Condition is perishable. A completed transaction is fixed. One choice, made by somebody who already knows the answer.

Four values across two fields, and the record stops lying by omission.

What that buys is specific. A perishable note surfaces with its age foregrounded and framed as what was true then rather than what is true now. A fixed note surfaces plainly. A judgment surfaces as somebody's opinion with their name attached. The reader stops having to reconstruct which kind they are holding, which is work they currently do badly or not at all.

What it does not buy is truth. A perishable note flagged as perishable is still unverified. This makes a record honest about its own reliability rather than accurate, and those are different things that get conflated constantly.

There is a harder version of the same proposal. Structure and context should be separate categories rather than points on a spectrum. If a thing can be defined consistently, sourced and applied repeatedly, it belongs in structure. If it depends on observation or interpretation, it stays context and is labelled context rather than dressed as data.

Every system that mixes the two drifts the same way. Context slowly gets read as fact, because the reader has no marker telling them where one ends and the other begins, and the only reliable defence is a boundary the system enforces rather than one the reader is expected to remember.

The record that outlives the knower

The Charlotte note is a small case and the mechanism is not small.

Most of what any professional knows about the thing they work on has no field waiting for it. Some of that gets written into free text, where it is preserved and unusable. Most of it stays in one person's head and leaves when they do.

What survives is what fit, and it survives without any marker saying that it was only ever part of what was known.

A note in CMAflow carries text and a date, and nothing else. It cannot say whether it was observed, documented or judged, or whether it is the kind of fact that expires. Nothing in the system is old enough for that to have bitten, so it is a design position rather than a shipped feature. I would rather describe a gap I can see coming than claim one I have closed.

---

This article is general information and analysis, not financial, lending, or appraisal advice. Verify any home value with a licensed professional before acting.

The Independent Agent
Substack | Spotify | CMAflow FAQ | YouTube | Free CMA | Home valuation | Insights | Blog