Reading your report

RealSiteWorth returns a range, not a price. This page explains what each part of that output means, which parts you can act on, and which parts are telling you about our data rather than about your asset.

The rule that shapes everything below: a deterministic engine computes every dollar figure, score, and percentage. The AI writes prose about those numbers afterwards. It never produces them.

The range, the midpoint, and why the band is that wide

The engine computes a midpoint first, then draws the band around it:

low  = round(mid × (1 - halfWidth))
high = round(mid × (1 + halfWidth))

halfWidth is not a statistical interval. It is a direct function of confidence:

halfWidth = lerp(0.55, 0.15, confidence)

So the band is ±15% at confidence 1.0 and ±55% at confidence 0. Nothing in between is a percentile. A wide band means "we measured fewer things", not "your site is volatile".

Working the arithmetic out at the label boundaries:

How to read it. Plan around the low end. The low end is also the number the Roadmap optimises against, so the two are consistent with each other.

Two hard edges worth knowing:

• Nothing reads $0. A valued asset is floored at $1. This only stops a literal "$0" readout. It never inflates a real number. A no-data lookup is not floored at all, because putting a dollar figure on an asset we have no data for would be inventing one.
• Above a $5,000,000 midpoint the band goes to maximum width and confidence is forced below MEDIUM. Past that point there are no comparable sales to anchor to, and the report says so with a beyond_comp_range flag. Treat a large number with that flag as directional only.

What confidence actually measures

Confidence starts at 1.0 and loses points for things we could not measure or things that make the asset fragile.

Then the label:

• HIGH at 0.75 and above
• MEDIUM at 0.45 and above
• LOW below that

The missing-signal penalty is the one that dominates. These eight signals are tracked, and each one that comes back null costs 0.15:

domainAge, trafficMonthly, trafficConcentration, referringDomains, domainRank, earliestSnapshotYear, archiveSnapshotCount, searchInterestSlope

Seven missing signals drive confidence to zero on their own. That is usually why a band looks wide. See "Making the estimate better" below, because which signals we even attempt depends on how the run was made.

Caps that override the arithmetic

Some conditions clamp confidence no matter how clean the rest of the data is:

A brand-risk cap is not an accusation. It says the same thing every time: transferability could not be established from public data, so we refuse to show a tight band around a number that assumes a clean sale.

The Free display cap

On a free website valuation, confidence is capped at LOW / 35% before it is shown, even when the underlying computation scored higher. This is a display cap on the free tier, not a finding about your asset.

One exception worth knowing: if a free run finds no traffic at all, it is re-run as a bare-domain valuation and reports the domain-name model's own confidence instead, which ranges from 40% to 70% by name length. So a bare domain can legitimately show MEDIUM on a free run while an operating site shows LOW. That is a difference in which model produced the number, not a statement that the domain is better understood.

A, B, or C: which model produced your number

Every report carries a mode, and the mode changes what the range means.

In modes A and B the domain-name value acts as a floor under the business range. In mode C the name value is the valuation, with a deliberately asymmetric band of mid × [0.5, 1.6] and confidence set by the name's length bucket.

If you expected mode A and got mode C, the usual cause is that no traffic figure came back. On a free run, a null or zero traffic reading routes the valuation to bare-domain treatment rather than failing. On a paid run, the same condition returns a "too thin" error instead of guessing.

RSW Auth: what the open web has invested in you

Authority is a 0 to 100 composite, higher is better. Each component is normalised against a benchmark, capped at 1.5× that benchmark, then combined as a weighted mean.

Grades: A at 85, B+ at 70, B at 55, C at 40, D below that.

Three things to know before you act on this score:

1. Missing components are excluded, not zeroed. The weights renormalise over whatever was actually measured, and the missing component is shown as "Not measured" rather than as a 0. A component list full of "Not measured" with a score of 0 means we had no authority data at all. It does not mean your site has no authority. Read the component list before you read the number.
2. The 1.5× cap is real. Pushing referring domains from 375 to 900 moves this score by nothing, because 375 already hits the cap. If the score is what you are optimising, work on a component that is still below its benchmark.
3. Archive history is scored on continuity, displayed raw. If the domain was dropped or parked for a stretch, the score uses drop-discounted years while the report shows the true archive years. The Memo explains the gap. A large gap between "20 years of archive" and a mediocre archive contribution is that discount at work.

RSW Trust: how clean the profile looks to a buyer

Trust is also 0 to 100 and also reads higher-is-better, but it is computed inside-out. The engine builds a risk magnitude from the components, then exposes Trust = 100 − magnitude. Components are capped at 2× benchmark.

Grades: STRONG at 75, FAIR at 45, WEAK below that.

Two readings that will save you from chasing the wrong thing:

The toxic-backlinks figure is a constant, not a measurement. It is a flagged placeholder fixed at 15% against a 25% benchmark, and it is present on every report. It is labelled "(est.)" in the component list. Do not read it as an audit of your link profile, and do not go disavowing links because of it.

Concentration and channel dependency are 90% of the weight. If your Trust score is low and those two are measured, that is where the score is coming from. Those are also the two components with real counterfactuals in the Roadmap, so they are the ones that can produce a dollar lift.

One consequence of the weighted-mean design: a report where only the toxic-backlinks constant was available scores 70 (FAIR). That is the arithmetic floor from the constant alone. It is not a finding.

The AI Memo and the Value-Gap Roadmap

Both are written by a language model that is handed the finished numbers and explicitly forbidden from touching them. The instruction it receives is direct: every figure was computed by the deterministic engine, and it must not invent, change, recompute, or contradict any number, or state facts it cannot derive from the supplied data.

The Memo

Five sections, always in this order:

1. Summary
2. Revenue and quality of earnings
3. Traffic and authority
4. What we did not include
5. What changes the number

Read section 4 first. The model is given an explicit ledger splitting signals into observed and omitted, and is told that an omitted signal must be acknowledged as missing and never narrated as a measurement. Section 4 is where the wide band gets explained in plain language. It is the most useful part of the Memo for deciding how much to trust the rest of it.

The Roadmap

The Roadmap is not AI-generated advice ranked by an AI. The ranking is arithmetic.

For each weakness the engine flags, it re-runs the entire valuation with that one input patched to its benchmark, then measures how much range.low rose. That rise is the lift. If it clears $250 the item is labelled Value lift. If it does not, the item is labelled Confidence, and no dollar figure is attached. Items are sorted dollar-lifts first by size, then confidence items. The model writes only the title, effort, timing, and explanation, and any $ token it produces is stripped from that prose.

Which weaknesses can produce a dollar lift, and which are confidence-only:

That table is the practical takeaway of the whole Roadmap. If you want the number to move, the four items at the top are the only ones with a mechanical path to moving it. The rest widen or narrow how sure we are.

What each tier sees

Two details the table does not capture:

• A fresh free website valuation returns no Memo and no Roadmap at all. The free response is trimmed to the band, confidence, and signals. The "first three Memo sections, Roadmap headlines only" preview depth is applied by the saved-report viewer to reports that already contain a Memo.
• The free export carries a TEASER watermark. Paid exports do not.

Monthly allowances, from the tier constants:

Anonymous callers are capped at 3 valuations per day per IP, shared across every valuation surface rather than 3 per surface.

Basic's 10 advanced runs are included in the plan, and the in-app way to spend them is rolling out. Advanced runs currently enter only through the agent/API transports, and MCP/agent access is Pro-only, so there is no control in the app that triggers one on Basic today.

Making the estimate better

1. Sign in. It changes which data sources run.

This is the single largest lever, and it is free. Which adapters fire is decided by a pull profile, and the profile is chosen by tier.

An anonymous run attempts one data source. That leaves roughly seven of the eight tracked signals null, which is exactly the −0.15 × 7 that drives confidence to the floor. Signing in adds four more sources at no cost.

Note what the table also shows: domain rank, referring domains, and traffic concentration require the deep pull, which is Pro only. On Free and Basic those three stay in the missing-signal list, the Authority score falls back to page authority for its rank component, and Trust runs without measured concentration. That is why the RSW Auth component list on a Basic report shows "Not measured" against referring domains.

And remember the Free display cap. On Free, signing in improves the underlying signal receipt and the scores. It does not raise the confidence number you see, because that is clamped at LOW / 35% before display.

2. Supply owner numbers for a store or business valuation

For an e-commerce or business asset, you can hand us real financials instead of having earnings inferred from traffic. This surface requires Basic or above. Anonymous and free signed-in callers get a 403.

The band width is set purely by how many of five specific owner fields you filled in:

The five counted fields are monthlyRevenue, monthlyCogs, monthlyEmployeeCost, topTrafficSourcePct, and monthlyChurnPct. Only monthlyRevenue is required, and there is no inference path for it. Without it the request is rejected rather than guessed at.

Every other field is optional and either adjusts the multiple or leaves the conservative band untouched:

{
  "monthlyRevenue": 42000,
  "monthlyCogs": 18500,
  "monthlyEmployeeCost": 6200,
  "monthlyChurnPct": 0.04,
  "topTrafficSourcePct": 0.55,
  "platform": "shopify",
  "businessModel": "held_inventory",
  "ecommerceCategory": "beauty",
  "averageOrderValue": 78,
  "monthlyOrders": 538,
  "inventoryValue": 60000,
  "ltvCacRatio": 3.2,
  "topSkuRevenuePct": 0.31,
  "ip": { "trademark": true, "patentOrFormulation": false },
  "ageYears": 4
}

Three things happen to that payload that are worth understanding:

• Omitted costs are inferred and labelled. If you leave out COGS, a category-typical fraction of revenue is used and marked inferred so the Memo can disclose it. Omitted employee cost defaults to 25% of revenue. Your own number always wins. Supplying it also moves you up the confidence table above.
• Risk inputs cut the multiple, they do not pad it. Monthly churn above 3% costs 8%, above 6% costs 18%, above 10% costs 30% of the band midpoint. A single traffic channel above 40% costs 5%, above 60% costs 15%, above 80% costs 25%.
• Negative earnings are refused, not floored. If revenue minus COGS minus employee cost is zero or below, the run fails with a stated reason instead of returning a number.

Both business and domain valuations then take the same 0.9 conservatism haircut after the multiple. The multiple bands themselves are calibrated to sit at or below the median of real broker sold-comps for each category. This tool is built to come in under a broker quote, not over it.

3. Re-run when the data ages

Valuations cache for 14 days. A saved report is flagged stale in the last 3 days of that window and expired after it. Re-running a valuation is what refreshes the cached report.

4. Read the signal receipt before you argue with the number

Every metric that could not be measured renders as "Not measured", never as 0, "—", or "0%". If a component you care about shows that label, the fix is a deeper run or an owner input, not a different interpretation of the number.

Pulling the useful fields out of a saved report:

jq '{
  mode,
  range,
  confidence,
  flags,
  weaknesses,
  beyondCompRange,
  authority: {score: .authority.score, grade: .authority.grade},
  trust: {score: .trust.score, grade: .trust.grade},
  unmeasured: [ (.authority.components + .trust.components)[]
                | select(.value == null) | .label ]
}' report.json

The unmeasured list is the honest answer to "why is the band that wide".

The disclaimers, in full

These ship with the product. They are not boilerplate we are paraphrasing here.

On every page that displays a valuation figure:

This value range is an automated estimate. The Memo and Roadmap explain the estimate after it exists; they do not generate the number or invent facts.

Directly under a valuation band:

Not financial advice. This is an automated estimate, not a formal appraisal, investment advice, or a recommendation to buy or sell. Real Site Worth is not a financial advisor, broker, or appraiser.

The full version:

Not financial or investment advice. Anything here about value, "worth", multiples, or asset comparison is an automated estimate and editorial opinion, never a formal appraisal, financial or investment advice, or a recommendation to buy, sell, or hold. Real Site Worth is not a financial advisor, broker, or appraiser. Do your own due diligence.

Shipped with every store and business valuation result:

This is a market-informed algorithmic estimate, not a formal valuation or offer to buy. The calculation uses a conservative multiple band based on real broker comparable-sale data and is not a substitute for a licensed business valuation. Results should prompt, not replace, negotiation with a qualified buyer or M&A advisor.

That last line is the right frame for the whole report. Use it to open a conversation with a buyer, a broker, or your own books. Do not use it to close one.