Evaporation
How much of your team's reasoning leaves nothing behind the moment the query finishes.
At one estate we assessed, more than ninety percent of analytical work left nothing reusable behind. The SQL text sits in the query log on a retention clock, but the reasoning behind it was never captured, and the query is close to impossible to find again.
Most of what your team worked out this quarter is already out of reach, and nothing was deleted. The SQL is still sitting in the log. What went missing is the reasoning behind it, along with any practical way to find it again, so the next person rebuilds it slightly differently. Every query carries a small act of reasoning: why the washout ran 180 days and not 90, why adherence used PDC and not MPR. Evaporation is what happens to that reasoning once the answer ships. The number is delivered, the thinking that produced it dissolves, and that thinking was the best available starting point for the next question.
You do not know your rate yet. You can, by the end of the day, off your own query history, for free.
What it tells us
The Evaporation Rate is the share of analytical work that leaves nothing reusable behind. The SQL text itself survives. It lands in the query log and stays there for a while. What the log never holds is the reasoning that produced the query, or any realistic way to find that query again before it clears.
Reasoning that evaporates is not cheaper the second time. Someone re-derives it, slightly differently, a quarter later. Capturing it does more than prevent that loss. Once captured, it becomes the floor the next analysis starts from instead of a blank page.
What we found
The modeled layer mostly survives by design. A dbt model that someone tested and documented persists because they built it to. This reading is about the wide ad-hoc layer beneath it: the one-off pulls, the exploration, the questions that were answered but never promoted to a model. That is where almost all of the week's reasoning happens, and where nothing asks you to keep it.
A typical one-off looks harmless. It answered a real question once, then closed.
SELECT patient_id FROM pharmacy_fills WHERE drug_class = 'oral_anticoagulant' AND fill_date = first_fill_date AND enrolled_months >= 12
Finds new oral anticoagulant patients with a full year of prior enrollment. It answered a real question once, then closed. The next person who needs it rebuilds it from scratch, and might get it slightly wrong.
This should have become a named view, defined once and reused. Instead it sits in the log with no reasoning attached, and the next cohort like it gets re-typed by whoever asks next. The log will hold it for 365 days. Whoever wrote it stopped remembering why the threshold was 12 months somewhere around day three.
The surviving 10% is rarely the best work. It tends to be the oldest, most-remembered work, the pulls someone happens to recall from last quarter. Analytical maturity ends up stored in whoever remembers what was done, an oral tradition running alongside a SQL editor. That is Key-Person Risk wearing a different hat. When they leave, the memory leaves with them.
Why it matters to you
Every evaporated query is rework waiting to happen. Your team re-derives last quarter's cohort logic instead of building on it.
The logic your reporting depends on disappears the moment the query finishes. What the number counted is no longer reproducible without the person who wrote it.
You pay to re-derive the same answer again and again. The cost hides inside senior hours and vendor invoices that never get their own line.
An agent built on the surviving 10% learns from fragments rather than your team's best work. It answers from whatever was remembered, not whatever was right.
What good looks like
The reasoning is captured at the source, before the log clears, so the choice of a 180-day washout or a PDC denominator is kept alongside the query that used it. Each analysis starts from the last one instead of a blank page, and the agent answers from the full record rather than the sliver that happened to be remembered. Captured this way, the reasoning stays available to the whole team and stays current as new work lands, instead of sitting with one person or frozen at last quarter's version. That is the difference between knowledge that evaporates and an institutional knowledge asset, a set of files readable by humans and machines that meets five criteria. The 90% that evaporated last quarter becomes the precedent this quarter draws on.
What to do about it
- Run Neuron on your query history.
- It recovers the reasoning that would otherwise evaporate, ranked by how often each definition was reused.
- Review the recovered logic with its owners.
- Promote the logic that survives review into shared, named objects and the semantic model.
- Each week, look at the one-off queries your team ran.
- Ask which of them should have been a view.
- Take the ones that recur and calculate them once, as a named object everyone uses.
- Re-read the history each week to catch the next batch.