Platform · Score methodology
How the Reputation Health Score is calculated
A score you cannot inspect is just an opinion with a number on it. This page sets out the whole model — the five dimensions, the proposed weights, what happens when data is missing, when Spy Monk refuses to show a number, and how much confidence each score deserves.
Updated By Drive Digital Editorial
See your own score and its explanation
Proposed weights, published · No score without enough verified data · Not a credit score or a legal finding
Direct answer
How is the Spy Monk Reputation Health Score calculated?
It is a versioned 0–100 composite of five dimensions: review quality, public sentiment, issue recurrence and severity, response and resolution, and search and brand presence. Dimensions without data are left out and the remaining weights rescaled — but only when enough data remains. Otherwise Spy Monk shows no score and says why.
- Proposed weights 25 / 25 / 20 / 15 / 15 per cent, model drs-rhs-2.0 — to be validated in a pilot before they are treated as settled
- No score below 8 mentions, 2 sources or 60 % of the model weight with data: “Not enough verified data to calculate a reliable score”
- Every score carries a confidence band, its sources, its explanation lines and a stale marker after 14 days
Principle
A direction, not a verdict
The Reputation Health Score exists to give you directional feedback you can act on: is the public picture of you, your brand or your organisation getting better or worse, and why? It is calculated the same way every time from the mentions Spy Monk could actually read, so two people looking at the same snapshot reach the same understanding of it.
It is deliberately modest about what it knows. The score only describes the sources that were checked in that run, and the coverage table says which ones those were. A dimension with no usable data is never quietly counted as zero — which would punish you for a source Spy Monk could not reach — and never quietly counted as perfect, which would flatter you. It is left out, and the page tells you so.
The weights below are proposals. The v7.4 specification sets them as defaults for experiment, and governance may change them per sector after a pilot shows how they behave. Every change creates a new model version, and scores from different versions are never compared with each other.
The model
The five dimensions
Each dimension produces a value from 0 to 100, or no value at all when the data for it is missing.
- Phase 2
Review quality · 25 %
Star ratings from review sources that are valid for the entity, such as a business profile. Review connectors arrive in Phase 2, so for most entities today this dimension has no data and is left out rather than guessed.
- Available now
Public sentiment · 25 %
The balance of positive and negative mentions across the sources that were checked. Sentiment labels are automated estimates; you can open every mention behind the figure.
- Available now
Issue recurrence & severity · 20 %
How often high- and critical-risk issues recur and how serious they are. One loud item and a steady pattern of the same complaint are treated differently.
- Phase 2
Response & resolution · 15 %
Whether criticism receives a reply and whether issues are resolved. It needs owner replies from connected accounts, so it is usually left out until those connectors exist.
- Available now
Search & brand presence · 15 %
Whether your own legitimate pages appear among the top web-search results for your name. Left out when no web-search source is configured for the workspace.
- Available now
Versioned and explained
Each snapshot stores the model version, the weight actually applied to each dimension, the sample size, the sources used and one plain explanation line per dimension.
Weights
Proposed weights and what happens without data
Defaults from the v7.4 specification, §125. They are configuration proposals that need pilot validation.
| Dimension | Proposed weight | Measured from | When there is no data |
|---|---|---|---|
| Review quality | 25 % | Ratings on source-valid review sources | Left out; weight shared among the rest |
| Public sentiment | 25 % | Sentiment share across checked mentions | Left out; weight shared among the rest |
| Issue recurrence & severity | 20 % | High and critical issue clusters | Left out; weight shared among the rest |
| Response & resolution | 15 % | Owner replies and resolved cases from connected accounts | Left out; weight shared among the rest |
| Search & brand presence | 15 % | Owned results in the web-search top 10 | Left out; weight shared among the rest |
Source: DEFAULT_SCORE_WEIGHTS and DEFAULT_SCORE_POLICY in the platform's shared contract (model drs-rhs-2.0). Operators can change weights per sector only through a new model version.
Step by step
From mentions to a score
Read the eligible sources
At the end of an audit, Spy Monk takes the mentions from the sources that returned data in that run. Sources that were blocked, unreachable or not configured are listed as unavailable — they are not estimated.
Measure each dimension
Each of the five dimensions gets a value from 0 to 100 with its sample size, its sources and a one-line basis — or no value when there is nothing usable to measure.
Check that there is enough
A score is only published with at least 8 mentions, from at least 2 distinct sources, and with dimensions covering at least 60 % of the model weight. Below any of these, the result is the insufficient-data state.
Rescale the weights that remain
The weights of dimensions without data are dropped and the remaining weights are rescaled to make up the full weight. This is the only adjustment, and the applied weights are shown.
Assign a confidence band
High, medium or low, depending on how much of the model has data and how many mentions sit behind it.
What you seeHigh needs ≥ 95 % of the weight and ≥ 32 mentions; medium ≥ 80 % and ≥ 16; anything else is low.
Publish the snapshot
The score, its version, the time it was calculated, the last successful fetch, the explanation lines and links to the evidence are stored together, so the number and its breakdown always come from the same snapshot.
Worked example
Renormalisation, worked through
These figures are illustrative — they describe no real entity.
Suppose an audit finds 20 mentions across 3 sources. Public sentiment measures 64, issue recurrence and severity measures 80, and search and brand presence measures 72. Review quality has no data (no review connector) and response and resolution has no data (no owner replies). The dimensions with data carry 25 + 20 + 15 = 60 % of the model weight — exactly the minimum — so a score can be published.
The three weights are rescaled to share the full weight: sentiment 25 ÷ 60 ≈ 41.7 %, issues 20 ÷ 60 ≈ 33.3 %, presence 15 ÷ 60 = 25 %. The score is 64 × 0.417 + 80 × 0.333 + 72 × 0.25 ≈ 71. With only 60 % of the model covered, the confidence band is low, and the explanation lines say that two dimensions were excluded and why.
If web search had not been configured as well, the dimensions with data would cover only 45 % of the weight. Spy Monk would then show no number at all — only “Not enough verified data to calculate a reliable score” and the reasons.
Rules
When Spy Monk shows a number, and how much to trust it
Thresholds are the current scoring policy. Like the weights, they are proposals to be measured in a pilot.
- Minimum sample: 8 mentions across eligible sourcesAvailable nowBelow it: insufficient-data state, no number
- Minimum breadth: 2 distinct sources returned dataAvailable nowOne source alone cannot describe a reputation
- Minimum coverage: dimensions with data carry ≥ 60 % of the weightAvailable nowOtherwise rescaling would stretch too little data
- Confidence band (high / medium / low) on every scoreAvailable nowShown with the gauge, never hidden in a tooltip only
- Stale marker when the last successful fetch is over 14 days oldAvailable nowThe score stays visible but is labelled stale
- Comparison only with ≥ 80 % weight coverage, the same model version and fresh dataAvailable nowOtherwise the comparison is warned or blocked
- Weights tuned per sector after validationPending feasibilityNeeds pilot data; until then the defaults apply everywhere
- Available now
- Available now6 items
- Pending feasibility
- Pending feasibility1 item
Limits
What the score is not
- Not a credit score. It says nothing about creditworthiness, financial standing or ability to pay, and must not be used for lending, hiring, tenancy or insurance decisions.
- Not a legal finding. A low score does not mean anything said about you is defamatory or unlawful, and a high score does not mean it is accurate. Only a court or the relevant platform can make that kind of determination.
- Not a universal truth. It reflects the sources Spy Monk could read in that run, in the languages it could classify, over that period. Content it could not see is not in it.
- Not a ranking. There is no league table and no “average for your sector”; Spy Monk has no benchmark data it could source.
- Not a promise. No one can guarantee that a score will rise, and that includes Spy Monk. Improving the underlying experience and responding well are what move it over time.
Related views
Comparisons, scenarios and the video summary
Comparisons. You can compare an entity with itself over time, or with up to three others, only on the same model version and on the sources every side has in common — the harmonised source panel. If the panels differ, the deltas are marked as estimates; if the versions differ or fewer than two sources are shared, the comparison is blocked. See Compare.
Improvement scenarios. The scenario planner lets you ask “what if our response rate improved?” and shows a broad range, labelled Scenario estimate — not a forecast or guarantee. It needs a scored baseline, assumes the same sources and model, and never models synthetic reviews or rating manipulation.
Video summary. An optional short video narrates the approved report, labelled AI-generated summary, with captions and a transcript. It is generated from the report and never adds a finding of its own; when narration is unavailable you get the captions, transcript and PDF instead. See Reports.
Versions
Versions and the older Health Index
The current model is drs-rhs-2.0. Audits completed before it keep the earlier Health Index, labelled v1, with its own explanation lines; it is described on the audit engine page. The two are different formulas, so a v1 number and a v2 number are never placed side by side as if they measured the same thing.
When governance changes a weight or a threshold, the change creates a new version with a date, earlier snapshots keep the version they were calculated with, and the explanation lines name the version. Terms used on this page are defined in the glossary.
Questions
Score methodology questions
Why does my dashboard say “Not enough verified data to calculate a reliable score”?
Fewer than 8 mentions were found, fewer than 2 sources returned data, or the dimensions with data cover less than 60 % of the model. The panel lists which. Adding sources, keywords or a longer period usually helps; Spy Monk will not fill the gap with a guess.
Why are two of the five dimensions often missing?
Review quality and response quality need data from connected accounts, such as a business profile. Those connectors come in Phase 2. Until then the two dimensions are left out and the others are rescaled, which is why many scores today carry a low or medium confidence band.
Is the Reputation Health Score a credit score?
No. It describes public online content about an entity, from the sources Spy Monk could read. It says nothing about finances and must not be used for credit, hiring, tenancy or insurance decisions.
Can Spy Monk raise my score?
No one can promise that. The score moves when the public picture changes — for example when recurring issues stop recurring or criticism gets a fair reply. Spy Monk helps you find and respond to what matters; it never removes lawful criticism to lift a number.
Who decides the weights?
The defaults come from the v7.4 specification and are proposals. Drive Digital's governance can adjust them per sector after a pilot, and every change creates a new, dated model version.
Why is my score marked stale?
The last successful fetch is more than 14 days old, often because monitoring is paused or out of scan credits. The number stays visible with the stale label until a new run completes.
See the score — and its reasons — for yourself
The free audit ends with a snapshot, its explanation lines and the coverage table, or an honest insufficient-data state.
No card needed