Platform · Audit engine
The audit engine and the Reputation Health Index
A reputation score is only useful if you can see what went into it. The Spy Monk audit shows every source it checked, every source it could not, and the exact arithmetic behind the number — so a 62 means the same thing today as it will in six months.
Updated By Drive Digital Editorial
No card needed
Published formula · Coverage printed in every report · AI-estimated, human-judged
Direct answer
How does a Spy Monk audit score what it finds?
Each new audit produces a Reputation Health Score (model drs-rhs-2.0): a 0–100 composite of five dimensions with proposed weights, shown only when enough verified data exists. Audits completed earlier keep the v1 Health Index, a published formula that starts at 60 and adjusts for sentiment, ratings and high-risk mentions.
- New audits: five dimensions, a confidence band and explanation lines — or “Not enough verified data to calculate a reliable score”
- Older audits: the v1 Health Index with its risk band, labelled v1 and never compared with v2
- AI influences only the inputs — sentiment and risk — never the arithmetic
What an audit is
A structured look at what accessible sources say
A reputation audit answers a simple question honestly: what do the public sources we can reach say about this name, brand, product, website or URL right now? The emphasis is on we can reach. No tool sees the whole internet. Private groups, closed accounts, messaging apps and anything behind a login are invisible to every audit, including ours — and a tool that does not say so is asking you to trust a number it cannot justify.
The Spy Monk audit therefore does three things in order. It identifies the entity from the name and the aliases you add (AI-assisted disambiguation of people who share a name is coming in Phase 2, so today you should add distinguishing aliases — a city, a profession, a Hindi spelling). It queries each source adapter that is available to it and records a coverage row for each one: covered, unavailable, not connected, or demo. And it normalises everything it finds into mentions — source, URL, date, title, excerpt, language — removing duplicates by URL so one story syndicated across five sites is counted once.
The result is a preliminary audit. We use that word deliberately: it is the start of understanding, not a verdict.
Pipeline
From a name to a scored, sourced report
Identify
You enter a name, brand, product, website, app or content URL and choose its type. Aliases widen the search; on monitoring projects, exclusion keywords narrow it.
Entity kinds: person, brand, product, business, website, app, content URL.
Search
Each source adapter runs independently. News is live today; web search and YouTube run when their providers are configured. Each adapter writes its own coverage row.
No adapter is silently skipped — unavailable sources are listed, not hidden.
Normalise & de-duplicate
Results become mentions with a canonical URL and a SHA-256 hash of that URL, so the same page found twice is stored once per entity.
Dedupe key: tenant + entity + URL hash.
Classify
Each mention receives an AI-estimated sentiment (positive, neutral, negative, mixed), a risk level and topics. With no AI provider configured, an English and Hindi/Hinglish word-list method is used and labelled as such.
Labels always say "AI-estimated"; the risk reason is shown in plain words.
Score & explain
The Reputation Health Score is computed from the five dimensions with the published model, or the insufficient-data state is recorded instead, and stored with its explanation lines — what each dimension measured, which were left out and why.
Two audits of the same data with the same model version always produce the same result.
Report
Findings, high-risk items, the coverage table and recommended next steps appear on screen and in a downloadable PDF that carries the same coverage table.
The PDF never shows a number without the coverage behind it.
The v1 formula
The v1 Health Index, kept for earlier audits
Audits completed before model drs-rhs-2.0 keep this formula and are labelled v1 wherever they appear. New audits use the five-dimension model described on the score methodology page.
| Component | Rule | Effect on score |
|---|---|---|
| Baseline | Every audit starts here | 60 |
| Sentiment balance | 40 × (positive share − negative share); mixed counts half to each side | −40 to +40 |
| Average rating | (average stars − 3.5) × 8, only when review ratings are available | about −20 to +12 |
| High-risk mentions | 5 points per high- or critical-risk mention | 0 to −25 (capped) |
| Recent negativity | If more than 40 % of mentions in the last 30 days are negative | −8 |
| Clamp | Rounded and limited to the range | 0–100 |
| Risk band | < 35 or ≥ 3 high-risk mentions → Critical; < 50 → High; < 70 → Medium; else Low | label, not points |
Source: docs/ARCHITECTURE.md §5 and packages/shared/src/health-index.ts (v1); the v2 model is in packages/shared/src/v74.ts. Either way, the score is a guide to where attention is needed, not an absolute measure of a reputation.
Coverage
The sources we could not check are part of the result
Every audit ends with a coverage table. For each source it states one of four things:
- Covered — the source was queried successfully and its results are in the score.
- Unavailable — the provider could not be reached or returned an error; the source is listed and excluded from the score.
- Not connected — the source needs an account connection (reviews from your own Google Business Profile, your own social accounts); connectors are coming in Phase 2.
- Demo — shown only in demonstration environments, never in a customer's live audit.
The table travels with the result: on screen, in the PDF and in the API response. If you share the PDF with a colleague, an adviser or a board, they see the limits exactly as you did.
Reading the result
What the number does — and does not — tell you
Treat the score as a direction finder. A high score with a high confidence band says that, in the sources we could reach, the picture is positive and nothing recurring was assessed as high risk. A middling score with two high-risk issues says: read those two first. And when there is too little to go on — a handful of mentions, one source — Spy Monk shows no number at all and says why, because being hard to find is a finding in itself, not a score. The full model, with its weights and thresholds, is on the score methodology page.
Three cautions apply to every audit. Sentiment is an estimate. Sarcasm, mixed languages and missing context mislead automated classifiers; the label is a prompt to look, not a conclusion. Same-name results happen. Matching is by name and aliases, so a finding about someone else with your name can appear; review before acting, and add distinguishing aliases. A low score is not a legal finding. Nothing in an audit establishes that a review is dishonest, an article is inaccurate or a post is unlawful. Those are questions for evidence, and sometimes for a qualified professional — which is why findings flow into cases rather than into automatic action.
Availability
Audit capabilities by status
- Entity identification by name and aliasesAvailable now
- News source discoveryAvailable now
- Web search and YouTube discoveryAvailable nowRuns when the search and video providers are configured; otherwise listed as unavailable.
- Multilingual sentiment — LLM when configured, EN + Hindi/Hinglish lexicon fallbackAvailable now
- Reputation Health Score (five dimensions, confidence band, explanation lines); v1 Health Index kept for earlier auditsAvailable now
- Downloadable preliminary audit PDF with coverage tableAvailable now
- Recommended next actions in the summaryAvailable now
- Reviews and social posts from your own connected accountsPhase 2
- AI-assisted disambiguation of same-name entitiesPhase 2
- Comparison over time or with other entities on a harmonised source panelAvailable nowWarned or blocked when source panels or model versions differ.
- Licensed media-monitoring feedsPending feasibilityDepends on licensing; adapter slot exists.
- Available now
- Available now8 items
- Phase 2
- Phase 22 items
- Pending feasibility
- Pending feasibility1 item
Questions
Questions about audits and the index
Why did my score change between two audits?
Because the inputs changed — new mentions appeared, a source that was unavailable became available, or sentiment on existing mentions was re-estimated. The model changes only with a new, dated version, and each audit stores its version and explanation lines so you can compare them.
Is the free audit the same as the paid one?
The scoring is identical. Paid plans add a premium audit tier with more scan credits, monitoring projects, case management and team seats. No plan unlocks sources that need an account connection until those connectors ship in Phase 2.
Can I audit a competitor or a person I do not represent?
Audit yourself or an organisation you are authorised to represent; the audit records that confirmation. A competitor organisation can be added for benchmarking only — compared on the sources every side has in common, marked as an estimate or blocked when coverage differs, and never the subject of an incident or case. Profiling private individuals you do not represent is prohibited by the Acceptable use policy.
What does "AI-estimated" mean on a finding?
That the sentiment, risk level and topics were assigned by an automated classifier — a language model where one is configured, otherwise a word-list method for English, Hindi and Hinglish — and should be confirmed by a person before anything is done about them.
Can I see the sources that were not covered?
Yes, always. The coverage table lists every source adapter with its status, on screen and in the PDF. Sources marked unavailable or not connected are excluded from the score rather than estimated.
See your own coverage table
The free audit shows the score, the explanation behind it and the sources it could and could not check.
No card needed