How it works
Six stages, in the order you work them. What arrives done, and what stays your call.
Every screen in the app explains itself once you're standing on it. None of them says why the order is the order, or what you're being asked to decide at each stop.
It isn't an explanation of assessment work. You do that for a living. Where a rule exists because of something you told me, I've quoted you, so you can check the reasoning rather than take it on trust.
Parcels
Which parcels the district actually assesses.
Freeboard pulled every parcel in the book and map straight from the county, stamped the pull with a date, saved each parcel's own county record, and drew every lot.
Membership. The software will not decide that a parcel is exempt, and it will not tell you your list is wrong when it disagrees with the county's. You said the recorded plat is the authority and the county is supposed to match it, so a difference between the two is a question rather than an error.
“I can't assess a Ramada. It has no purpose. [...] I'm providing sewer. Well, there's no toilets or sinks that the ramada doesn't make any sense to assess them.”
You, on why assessability is a judgment and not a lookup.
Valuations
Which tax year the values are read from.
Freeboard fetched the full cash value for every parcel in one pass and saved the county's own page for each one, dated, so you have the backing without printing them a parcel at a time.
The tax year policy, recorded as a decision with your name and the date on it rather than as a setting somebody changed.
“every time that you buy a parcel, that APN number changes with the ownership. So when I go back to try and find that parcel, it no longer exists.”
You, on why the saved record matters more than the number.
- Full cash value
- $851,300
A real assessor value from this district
Methodology
The method, and the ratio threshold every unit gets tested against.
Nothing. This stage records what you chose, who chose it, and the reasoning, so the answer survives being asked about in five years.
All of it. The software has no opinion about which method fits a job, and the threshold is a number you state rather than one it assumes. For Cahava you set the test at total to bond, capitalized interest included, against the new bond only.
“These formulas have been time tested. They stand up in court. Everybody does them, you know, the same way.”
You, on the library this stage is built to hold.
- Ratio threshold
- 4:1
- Assessment basis
- Johnny B
- Lien basis
- Johnny B
- Method
- Not recorded yet
This district's recorded decisions
Roll
The EDU on each parcel, and the answer to every flag you raised.
Freeboard spread the money, tested every unit against your threshold, balanced the whole roll to the cent, and named both the units that fall below the line and the exact cash that clears each one. Change a single EDU and it recomputes everything and shows you what moved.
EDU. Always.
“You can't just look at something on the county parcel and go, that's 230 EDUs. It doesn't even know it's going to subdivide. It just sees one parcel of land.”
You, on why no county field is allowed to reach this number.
The computed roll, as one shape
Evidence
Whether to accept the artifacts the numbers rest on.
Every value the report leans on points at a dated file whose fingerprint still matches, so nothing has changed underneath you since you looked.
Accepting them, and the reason behind each judgment, so a reviewer can ask a person why a number is what it is rather than being told the system produced it.
“Eventually, I need to clear them all. [...] It's a snapshot in time. It's something that I need that I know we need to resolve, but then we should have it cleared before we ever finalize.”
You, on why an open flag blocks the report rather than being listed at the back of it.
- Evidence exists, verifies, and is accepted
Every value the report relies on resolves to a dated artifact whose hash still matches and which the engineer has accepted.
- Every judgment has an author and a reason
Engineer-owned changes and basis values are attributable, so a reviewer can ask a person why a number is what it is.
Report
Everything about the document.
Less than the rest, and this is the stage I'm building now. When it's finished it produces the report from the locked revision, the assessment schedule copied from that revision with no arithmetic added in the document, a manifest of the revision itself, and a manifest of every supporting artifact.
Signature and seal, outside the product, permanently. Correcting an issued report creates a successor and a new bundle rather than overwriting one, so nothing that has gone out is silently rewritten. This is also the stage where I'm stuck without you: two or three finished reports from different counties is the thing that tells me what's fixed about an engineer's report and what's just how one county lays it out.
- Report
- From the locked revision
- Assessment schedule
- Rows copied from that revision, no arithmetic added
- Revision manifest
- Its identity, inputs and reproducible results
- Evidence manifest
- Every artifact, with its hash and capture time
What generating the bundle produces
The rule underneath all of it
Freeboard does the mechanical half and stops. It reads the county, it does the arithmetic, it keeps the paper trail, and it refuses to infer a judgment from a county field no matter how convenient the field looks. Every number in a finished report traces back either to a dated county record or to a decision with your name on it.
That rule is the reason the tool is worth trusting at twenty districts a year rather than four, and it's the one thing I'd ask you to hold me to. If you ever find a number in here that came from neither of those two places, that's a defect and I want to hear about it.
“It would be cool if I had some way to get AI to do most of this work so I can go from 4 or 5 to like 20 a year.”
You, the first time you described what you wanted.
None of it is AI. Nothing in here infers anything or understands anything, and the twenty a year has to come from the mechanical half being done properly rather than from a model guessing well. The one place you said it belongs is drafting the narrative sections of the report over numbers the deterministic core already produced, and that is the only place it will go.