What a Five Star Confidence Score Promises

What a five star confidence score means

A five star confidence score on an RPR valuation means the sale price is likely to land within 10% of the estimate 90% of the time. That figure is not an outside estimate or an inference. RPR publishes it in its own support documentation, alongside the bands for every other rating on the scale, and five stars is the top of that scale.

One question runs underneath everything below, and it is narrower than an accuracy contest: can a tool show where its number came from. Plenty of valuation products are accurate enough for the work they do, and few are built to be explained out loud to a seller who disagrees. RPR is a useful place to work that question through, partly because it arrives as a membership benefit rather than as a purchase, and mostly because it publishes more of its own mechanics than the category usually does, which means the explaining can be done from documents instead of from guesswork.

What the model reads

RPR draws a clean line between its two estimates, and the line is worth reproducing exactly. RVM values, it states, use on-market and off-market MLS listing data plus publicly recorded sold data. AVM values are based solely on publicly recorded sold data. Everything else about the product follows from that sentence, and the MLS layer is a genuine advantage rather than a positioning claim.

RPR is unusually direct about why it matters. Karen France, quoted on its own blog, describes public records as known for being incomplete and slow to react to changing market trends, and calls MLS data the material that fills in the blanks those records leave. A vendor willing to publish the weakness of the input every competitor relies on is telling you something true about the category. An estimate assembled from county records is reading a document that was filed after the fact, written for tax purposes, and updated on a schedule nobody chose for valuation.

So on the first half of that question, where the data came from, the answer is on the page. You can read what classes of information went in, and you can read which of the two estimates you are looking at.

The bands RPR publishes

The confidence score runs from zero to five stars. Its stated purpose is to describe the expected accuracy of the estimated value, and RPR attaches a published margin of error to each rating.

Confidence score	Published band (sale within this range 90% of the time)	Range on a $600,000 estimate	Width of that range
5 stars	plus or minus 10%	$540,000 to $660,000	$120,000
4 stars	plus or minus 15%	$510,000 to $690,000	$180,000
3 stars	plus or minus 20%	$480,000 to $720,000	$240,000
2 stars	plus or minus 25%	$450,000 to $750,000	$300,000
1 star	plus or minus 30%	$420,000 to $780,000	$360,000

Source: RPR support documentation, What are RVM and AVM estimated values. The dollar columns apply the published percentages to a $600,000 estimate.

Read down the right-hand column before reading anything else. At the best rating the product offers, on a mid-priced American house, the window is $120,000 wide and one sale in ten is expected to land outside it. At one star the window is $360,000 wide on the same house. Nothing here is hidden, and publishing a band at all is a stronger disclosure than a single number standing on its own, because a figure with no interval beside it makes a quieter claim about its own precision than this table makes.

What the table does is separate two things that look identical on screen. A valuation carrying five gold stars reads as a verdict. It is a band, and a wide one, and the stars are a statement about the band rather than about the house.

The score measures agreement, not accuracy

Here is the part that changes how the number should be used. RPR defines the score in terms of its models rather than in terms of outcomes. A high confidence score, the support page states, indicates that the results of multiple automated valuation models yield similar estimated values. A low score indicates that those models yield a wide range. Elsewhere RPR puts it as a score based on the outcomes of multiple automated valuation models, with higher scores defined by narrower valuation ranges.

That is a dispersion measure. It reports how far apart several models landed, which is a real and useful signal, and it is not the same property as being right. Models drawing on overlapping inputs will converge when those inputs are consistent, including when they are consistently wrong. A renovation that never reached the public record is invisible to all of them at once. A condition problem behind a front door is invisible to all of them at once. In both cases the models agree, the range is narrow, and the score rises. Agreement among readers of the same flawed page is not evidence that the page is correct.

RPR does not dispute any of this. Its own writing states that AVMs do not factor in a property's condition, that unusual homes and new construction distort an estimate, and that low inventory in a market affects the quality of the output. It also says, in a sentence worth repeating to any seller who arrives with a printout, that these models are not meant to serve as an appraisal. The score is therefore doing exactly what RPR says it does. The failure mode is in how it reads to an agent in a hurry, because five stars out of five is the visual language of a rating rather than of an interval.

What RPR does about the gap

This is the part that matters more than an accuracy contest. RPR does not ask the agent to accept the model number. Its Refine Value tool lets the agent set the condition, correct the bedroom, bathroom and square footage details the record got wrong, and add upgrades or repairs, with regional renovation figures from Zonda applied at zip code level. The design concedes the thing the score cannot cover and hands the correction to the person who has been inside the house.

Two consequences follow, and both belong in a fair review. The first is that a refined value is no longer the model output, which is the right outcome and should be stated as such rather than blurred. The second is that the refined number inherits a new problem: it is now the agent judgment expressed as a figure, and whether the seller ever learns which part came from the model and which part came from the agent depends on what the finished report chooses to print.

What the documentation covers, and what it leaves out

Half of the question is covered, and covered well. RPR publishes what data classes feed each estimate, publishes what each confidence rating means as a numeric band, publishes that the score is a measure of model agreement, and publishes the limits of the whole approach in its own voice. On transparency about its method, RPR has published more than it was under any obligation to.

What it does not publish is the derivation for the house in front of you. There is no printed list of which models were consulted, which sales anchored them, what was adjusted and by how much, or why this property drew four stars rather than five. There is also no published accuracy figure. RPR states that the RVM has shown consistent high performance in independent testing, and points to tests rather than to a median error, so the band table remains the only quantified promise available. For an agent who needs to explain a number across a kitchen table, the honest reading is that RPR tells you how much to trust the figure and not how the figure was reached.

That distinction holds for any tool in this category, including this one. CMAflow computes its confidence level from the dispersion of the adjusted comparable values rather than from the spread of several models, with the coefficient of variation across those values setting a band of 5%, 8% or 10%. It is still a dispersion measure and it carries the same structural caveat: comparables agreeing with each other is not proof that the subject belongs where they put it. What changes is what the reader can audit, because each adjustment arrives on the page with its reason printed beside it and each comparable carries the recommendation to use or set it aside along with why. A tight cluster earns a tight range, a scattered one earns a wide range, and the reasoning is on the page rather than behind the score.

Context Blindness™ is the gap every tool in this category is working against, and the measure of a tool is not whether it closes the gap by itself but whether it shows the agent where the gap sits.

What does a five star confidence score mean on an RPR valuation?

RPR publishes the answer in its own support documentation. Five stars means the sale price is likely to fall within 10% of the estimated value 90% of the time, and five stars is the highest rating on the scale. On a $600,000 house that is a window running from $540,000 to $660,000, and one sale in ten is expected to fall outside even that. The published bands widen from there, reaching plus or minus 30% at one star.

What is the difference between an RVM and an AVM?

RPR states that RVM values use on-market and off-market MLS listing data plus publicly recorded sold data, while AVM values are based solely on publicly recorded sold data. The MLS layer is the real distinction and it is a substantive one, because RPR itself describes public records as incomplete and slow to react to changing market trends. Both remain estimates produced without anyone looking at the house.

Does a high confidence score mean the valuation is accurate?

It means the models agreed. RPR defines a high confidence score as several automated valuation models yielding similar estimated values, and a low score as those models yielding a wide range. Agreement and accuracy are different properties. Models reading the same incomplete record can converge on the same wrong number, and the score will report that convergence as confidence.

How to use the score on Monday

Treat the stars as an instruction about process rather than as a quality grade. At four or five stars, the models are reading a consistent record, which means the public and MLS data on that house is probably clean and the figure is a reasonable starting point for a conversation. At two or three stars, something about the property or the market is inconsistent enough to pull the models apart, and the useful move is to find out what, because the cause is usually a specific and knowable thing: a record that disagrees with the house, a shortage of recent sales nearby, an unusual floor plan, new construction with no closed history. A low score is not a worse number. It is a flag naming where the work is.

The practical test for any valuation tool is what the seller can be shown. A band is something an agent can defend, because it states the uncertainty plainly and it comes from the vendor rather than from the agent. A single figure with a star rating beside it is harder to carry into a disagreement, because the only available answer to why is a brand name. When the adjustments, the comparable decisions and the reason for the range are all printed where the seller can read them, the conversation stops being about whose number to believe and becomes about whether the reasoning holds, which is a conversation an agent who has seen the house will win.

---

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