How Are Public Records Cleaned Before You Use Them? (Watch, Standardize, Verify)
Early morning, a desk, a fresh printout of public records. One name is written backwards. One charge is three letters and a number. The same person shows up twice. The printout is not fake — every line came from a real public source. It is just raw, and raw records eat a professional's morning one guess at a time.
This post follows one record through the three steps that turn raw into ready: watch, standardize, verify.
The short answer: Public records get cleaned in three steps. First, the public sources — county jails, court clerks, code enforcement boards, property records, business listings — are watched on a rhythm, so new records are seen close to when they appear. Second, each record is written into one shared format, spelled into plain English, and counted once (this writing-it-one-way step is called standardizing; removing the extra copies is called de-duplicating). Third, a person checks the cleaned record against its source before it is delivered.
The Data Decoder follows one sample record — clearly marked as a made-up example, not a real person — through all three steps. You will see its flaws up close: the flipped name, the coded charge, the twin copy hiding beside it. Then you will see the step everybody skips, the human check, taken apart into four questions. The video ends with an honest limit: where cleaning must stop, so it never starts changing what a record says.
What does step one, watching, actually look like?
Each public source posts records on its own schedule. A jail posts bookings as they happen. A clerk's office posts filings as cases move. Watching means checking those sources again and again, on a rhythm, so a new record is found close to when it appears — not stumbled on next week.
Think of a watcher doing steady rounds past five doors, tapping each one in turn. Nothing fancy. The rhythm is the skill.
What happens to a record in step two?
Two things. First, it gets rewritten one way: names in the same order, dates in the same format, charges in plain English instead of shorthand. Five offices may write the same fact five ways; your team should only ever have to read it one way.
Second, its twins get merged. If the same person and the same event appear twice in two spellings, they become one record. The fact survives. The double-count dies.
What are the four questions in step three?
A person compares the cleaned record with the public source it came from and asks:
- Is it still current? Cases move on after records are written.
- Is it the right person? Names repeat in every county.
- Does the plain-English version say what the shorthand really meant? Or did someone translate a guess?
- What is missing? A bond amount that was never set shows as blank — and blank is information, not zero. A hold flag (a note on a booking record that can block release) changes the whole conversation.
Only a record that passes travels. One that fails gets set aside and marked, not quietly fixed by hope.
Where must cleaning stop?
At the facts. Cleaning may change how a record looks. It must never change what a record says.
If the source lists no bond amount, the cleaned record shows blank — not a guess, not a "typical" number. If a shorthand charge could honestly read two ways, the record gets flagged for a human instead of flattened into false confidence. A flag is not a failure. A flag is the system telling the truth about its own limits, out loud, before you act.
You can see the finished work for free in the Sample Lead Breakdown on aidcos.world, which takes records apart piece by piece: where each field came from, how it was cleaned, and what got flagged and why.
The 15-minute one-record audit
You can run a pocket version of this whole pipeline today, on a list you already own. Pick one record — any record. Find the public source it came from. Read the two side by side and ask the four questions from step three: still current, right person, right meaning, anything missing?
Then count how many times that person or event appears in your list. One appearance is a record. Two is a warning. Fifteen honest minutes with one record will teach you more about your list's real quality than a month of working it blind — because now you know which step your list is skipping, and a skipped step has a name you can fix.