The property record

Overview

Territory gives you a list. The property record is the page behind one row: the roof on the aerial photo (and as a 3D model where Setell has built one), the findings marked where they were seen, what changed between captures, and the county's record of who owns it.

Every number on the page is a claim with a source behind it. Where Setell doesn't know something, it says so in words and draws nothing — an unknown is never a dotted line on the picture that you might read as a measurement.

The Territory list prints the same condition word as this record. The number in the Territory map drawer is different: it runs the opposite way to the record's grade. See Reading the Roof Health grade before you compare the two.

Before you start

• The Roof Intelligence add-on has to be active. See docs/roof/overview.
• The building needs to be in the corpus — that is, Setell has read the area. Run a Territory scan first if it isn't (docs/roof/territory).
• Nothing to set up beyond that. The record reads public imagery and public county records, not your job data.

Open a record

1. Open Territory and find the building — in the table, on the map, or by searching an address or owner name.
2. In the table, click the row. On the map, click the pin to open the drawer, then click Open the property record.

The page's web address carries what you're looking at — the section, the capture, the view, and any claim you've opened. Reload it, or come back to it later, and you get the same view rather than a reset page.

The four sections

The record is one page with four numbered sections. The numbered buttons jump between them.

• 1. What we see — the building itself: its sections, its structure, and its rooftop equipment, plus the view modes and the overlays that draw findings. This is the "what is this building" view.
• 2. Condition now — the grade, the findings marked on the roof where they were seen (numbered so the panel and the picture refer to the same spot), and what an inspection should cover.
• 3. How it changed — the roof's scored captures year by year, and what this record can say about each year. See Reading the change timeline.
• 4. What to do — why this property was flagged: the owner and property record, hail exposure, the roof's reading, and the size class of the work (spot repair, section job, or whole-roof job) with the roof area in squares.

Under the sections, Nearby commercial lists the neighbouring buildings Setell has scored, each linking to its own record.

The picture

The record opens on Mosaic: the aerial photo, with this roof's outline drawn on it and the area around it dimmed. Nearby buildings are drawn as faint outlines. Those are public building outlines only, with no owner or address; a building with no stored outline gets nothing drawn, never a guessed box.

The photo always shows one capture, never a patchwork of several. Tiles that capture doesn't have are left empty. The caption names the capture — the provider and the year.

To move around the picture:

• Zoom in, Zoom out, and Recenter on roof sit on the right edge of the picture.
• Drag to move the picture. On a touch screen, use two fingers; one finger scrolls the page.
• Hold Ctrl (⌘ on Mac) and scroll, or pinch, to zoom. A plain scroll moves the page, not the picture.

Twin draws the roof's outline on the photo. Where Setell has measured the building from its lidar survey and the photo, that outline goes around the whole building. The part outside the older map outline is hatched and marked not read yet, because the condition reading has not covered it. One Condition button under the picture adds the condition layer: soft blue areas where water is most likely to pond (darker is more likely, a dark edge marks the likeliest areas), grey areas that were in shadow when the photo was taken, and the findings. A short key appears beside the button.

Capture year, under the picture, switches between the captures on file. It appears only when there are two or more captures; with one, the caption names the year. Where two captures share a year, the button names the provider too. Some captures are licensed for measuring only; pick one of those and the record says "The 2023 capture is measure-only under its licence, so its pixels are not shown." rather than showing them.

The 3D model

Where Setell has built a model of the roof, a Twin button sits beside Mosaic. It shows the roof as a 3D model, with the findings drawn over it where they can be. On a model that covers only part of the footprint, the button says Geometry (partial) instead. Where there's no model, there's no button, and the caption says the geometry is not yet modelled.

The model is drawn one of three ways, and the 3D twin card says which and why. Where the fitted roof planes match the lidar survey, the roof is drawn from those planes. A flat roof whose planes do not fit is drawn as its measured flat levels. Where neither is possible, for example a pitched roof the survey is too sparse to draw, there is no 3D, and the card says why in plain words. Setell never draws a shape the data cannot support.

The Where water ponds card says where water is most likely to sit, and how many places stand out. Places on parts of the roof that slope 10 degrees or more are left out, because water runs off a slope. Places that were in shadow when the photo was taken are counted separately, because they cannot be read.

In the Twin view you can move between a top-down view and two angled views from opposite corners. The tilt is deliberately limited. The model is built from overhead imagery, so a steep side-on angle would be showing you geometry nobody measured. Where elevation data exists for the building, Setell has a true three-dimensional read and allows a little more tilt; without it the model is a flattened one built from the picture alone.

What wasn't assessed

In the view modes that draw findings, a Not assessed card sits under the overlays. It says which checks didn't run on this capture, and why:

• below this imagery's resolution, or not assessed on this capture — shown one line each, because these are limits of the capture.
• Checks Setell doesn't run reliably yet, and checks that don't apply to this roof (for example, on a pitched roof or a given material) — folded into one Not assessed line you can open.

Reading the Roof Health grade

The record grades the roof on a five-band rubric, and higher is better:

Under the grade, 2. Condition now shows the five bands in a row, Excellent first, with this roof's band marked. It shows where the grade sits. It adds no precision. Where the imagery can only separate bad from fine, the row and the time line below are not shown.

The Territory list prints this same word. A roof the twin graded Poor reads Poor on the list row and on the record. See docs/roof/territory.

The imagery number is a different scale. Where only the imagery engine has read a roof, Territory shows an imagery band, and the map drawer shows its number: 0 to 100 where higher means worse — a triage order for the map. The record grades Critical to Excellent where higher means better — a judgement on the roof. Poor and Critical mean roughly the same thing on both; Good does not. On the imagery scale, Good means little was found. On the record, Good is the second-best grade of five.

Setell often shows the band without a number. The number only appears when the read was precise enough to support one — a grade you can act on, rather than a decimal that invites you to invent a meaning. On some roofing materials the imagery can only separate bad from fine, and the record says that instead of picking a middle band.

Alongside the grade, Setell prints the sentence explaining it: what pulled it there, which section was worst, and whether the material or the capture limited what could be read.

Opening an unscored record

Some buildings are in the corpus with their shape measured but no condition read yet. Opening one of those can put it in the queue to be scored — the record says "Scoring is queued. This record updates when the model has read the roof." and you come back to a scored record.

It doesn't queue every time, and it never goes and fetches new imagery on your behalf. A building is only queued where Setell already holds cached imagery for that metro and a scoring run is open for it. Where those don't hold, the record stays not yet read and a Territory scan is the way to get it read (docs/roof/territory).

A geometry-only record says what it is. The caption reads "Geometry from the … lidar" rather than naming a rating, the health line reads Roof: Not assessed with no capture clause after it, and there's no "Rating read from …" rail — because nothing rated it. A capture is not a verdict, and the page won't imply one.

Reading the change timeline

3. How it changed opens with the scored captures for this roof. The first line says how many there are — "No capture scored for this roof yet.", "One capture scored for this roof (2022). Change needs a second capture.", or "4 captures scored for this roof, 2018–2024." With two or more, each year gets a row: the source, the grade with its value and range, a low-confidence tag where it applies, and the change from the year before — "Down 12 points since 2020", marked within uncertainty or beyond uncertainty depending on whether the two ranges overlap, or "Not comparable with 2020".

Where the record holds photos from two or more years, a time line comes first: one dot per year a photo was flown. A filled dot is a photo that was read for condition; a hollow dot is a photo on file that was not. The large dot at the end is the roof's condition now, in its band colour. Every date is a year.

Below that is the timeline — one row per year this record can say something about. Each row says what this record holds for that year, in plain words:

• A grade where the roof was read that year, and how it moved — "Watch in 2020 → Poor in 2022; surface asphalt shingle → modified bitumen."
• No condition, with the reason, where the record holds a capture for that year but no condition reading from it: every capture that year is too coarse to score, a capture was withdrawn and no other was scored, or a capture was read and produced no condition band. The row never borrows a grade from another year or another capture.
• Not comparable, with the reason — usually that one of the two captures is too coarse to measure a change against, or carries no imagery to compare (a lidar point cloud, an elevation model).
• Years with nothing on this record are gathered into one line — "Nothing on this record is dated to: 2019, 2021" — rather than a row each. That is a statement about this record and nothing more. It does not mean nobody flew the building that year.

Years are compared with their neighbour, not all against the newest one. Two captures can only be compared down to the coarser of the two, so pairing each year with the one next to it is what makes an older, coarser capture sayable at all — against the newest capture it would just fail.

Where the whole record reads one year, the section says so — "this record reads one year; measuring a change needs two" — rather than showing an empty card. A capture the record can't date gets no year and no invented one.

The timeline also prints on the exported report, under Changes over time. Long records are capped there and the PDF says how many years it left out.

Where every number came from

Select any claim on the page and Setell shows its provenance — the value, its status, the source it came from, and when it was retrieved. Sources are named plainly: an imagery observation, a building outline, a county, municipal, federal or hazard record, a weather station record, the model, a derivation, or a person.

Claims come back in three states, and the difference matters:

• Known — measured or published, with the source named.
• Unknown — the check ran and could not resolve it, with the reason given in words ("not measurable at this resolution", "roof shape not resolved from lidar").
• Absent — checked for and genuinely not there.

An unknown is never drawn. It is text in a panel, never a line on the model.

What's in the rails

• Roof Health — the grade, the explanation, and the capture it was read from.
• Sections / material — the roof broken into sections where Setell could separate them, each with its material and rating.
• Structure — roof planes, parapet, height, where those resolved.
• Evidence — the per-capture rows behind the condition read, including the rows that record what was not visible.
• Captures — every dated capture on file for the building, with its source, resolution, and licence, and which of them were used to evaluate it.
• Provenance — the trace for whichever claim you have open.

The owner and the hail exposure print once, in 4. What to do, under why the property was flagged. The hail line counts the days radar showed severe hail within 5 km (3 miles), and how many of them had hail 2 inches or larger. A note under it says that hail has not yet been shown to change the condition readings. The rest of the county's public record — mailing address, acquisition, land use, assessed and market value, year built and building area — and the weather counts for the location (wind, precipitation, freeze–thaw) are not a separate panel any more. You reach them through the provenance of a claim, and on the exported report. All of it only where the county publishes it.

Where the county record names an owner but that county's terms don't let Setell show it, the record says Owner: Withheld — "This county source's rights do not let Setell show the owner's name." Where the county never published an owner, nothing is shown.

Export the report

Export report sits at the top of the record, and again in 4. What to do. It prints a one-page PDF: the building as drawn, the sentence saying why it was flagged, the numbered areas, what an inspection should cover, the capture-year timeline, and the imagery licence notices.

This is your working sheet, not the owner's document. It carries the owner's name, the acquisition record, and the hail history — the things you use to decide whether to knock, not things to hand back to the person who owns the roof. The owner-facing document is a different one, generated from a different button: see docs/roof/recommendation-report. Neither contains a price.

Ask Setell instead

In Studio: "What does the record say about that building on Oak Street?" Setell returns the same claims, with the same unknowns intact.

Over MCP, the hosted connector can summarise one Territory roof — its condition, the recommended action, every capture with its provider, year, resolution and licence, and links to this page and the roof report PDF. The full property record, claim by claim, is not published as an agent tool yet; Studio and this page are the ways to read all of it. See docs/mcp/capabilities.

What happens next

• The record is live. When a newer capture is processed for the building, the page reflects it.
• Start a quote, beside Export report, turns this roof into a job after you confirm a preview. See docs/roof/start-a-quote.
• Nothing here contacts anyone.
• What you do with the record is yours. Setell reports the property; the outreach is outside the product.

Troubleshooting

• I see an offer instead of the record. The add-on isn't active on your account. See docs/roof/overview.
• "Not yet read — no capture has been processed for this building." The building is in the corpus but nothing has scored it, and opening it didn't queue one — see Opening an unscored record. Run a scan over that area (docs/roof/territory).
• "Scoring is queued." Opening the record put the building in line to be read. Nothing more to do; come back and the score will be there. See Opening an unscored record.
• "N items in this record could not be read." Some of the stored rows behind the record were malformed and were left out. The rest of the record is served as normal — Setell prints the count rather than dropping the page or quietly showing you less than it has.
• How it changed says it reads one year. Only one dated read exists for this building. Change compares two; it can't run on one.
• A year I know was flown is listed under "Nothing on this record is dated to". That line is about this record, not about the county. The capture may exist and not be on file here, or be on file without a readable date.
• A year says "not comparable." The two captures either side of it can't be measured against each other — usually one is too coarse, or it's lidar or an elevation model rather than a picture. The row stays visible rather than disappearing.
• A year says "no condition". The record holds a capture for that year but can't state a condition from it, and the row says why. Imagery coarser than about 15 cm (6 in) per pixel, such as 60 cm NAIP or 1 ft statewide imagery, can't be scored. A withdrawn capture leaves the year without a condition unless another capture fine enough to score covers it. Setell shows the gap rather than a grade from a neighbouring year.
• Structure fields say "not resolved". There's no elevation data for the building, or too little of it. The model falls back to a flattened read and says so rather than drawing a shape it didn't measure.
• No owner, no values, no year built. Setell doesn't cover that county's assessor yet, or the county doesn't publish that field.
• The grade disagrees with the list. It doesn't — they're two different scales running opposite directions. See Reading the Roof Health grade.
• Still stuck → docs/reference/troubleshooting.

FAQ

• Does this give me a phone number or an email? No. The record reports property and roof facts from public records and public imagery. It doesn't build contact lists.
• Does it say how old the roof is? No. Setell doesn't claim roof age from imagery anywhere in the product, and the record is no exception.
• Can I show this to the owner? Not this page, and not the exported report either — both are your working view, and both name the owner and their acquisition record back at them. The owner-facing document is a separate report with its own button (docs/roof/recommendation-report).
• Does opening a record use a quote credit? No. Nothing in Roof Intelligence is metered against AI quotes or AI messages.
• Why is the model flat-looking on some buildings? Because no elevation data was available and it was built from the overhead picture alone. Setell tells you which read you got rather than styling both the same.

Related

docs/roof/overview · docs/roof/territory · docs/roof/start-a-quote · docs/roof/recommendation-report · docs/roof/report · docs/mcp/capabilities

Ready to try it on your own quotes?

Start free — no credit card. 14 days of unlimited Pro, then 3 free quotes a month.