Skip to content
VizBolt

Methodology

A Tableau Server health check that measures, not asks

Most assessments send you a questionnaire and score your own guesses. A VizBolt review is measured — by a read-only collector your own administrator runs — under four guarantees that hold on Tableau Server and Tableau Cloud alike. The method itself is published in full, every threshold attributed, in How to Audit a Tableau Environment.

Four guarantees

01

Read-only, provably

The collector opens its database session read-only and runs SELECT statements — nothing else. There is no code path that writes to your environment.

SET TRANSACTION READ ONLY

02

No internet access

It makes no outbound connection of any kind. It reads, writes one file to your disk, and exits.

writes to ./vizbolt-summary.json

03

You preview before you send

The summary is a few hundred KB of counts, durations, and percentiles. You open it, read every line, and decide whether to send it.

aggregates only

04

No names, no credentials, no raw data

User names, email addresses, workbook and project names, connection strings, query text, and business data are never collected. Object names can be replaced with codes — the key stays on your machine.

key stays on your machine

What crosses your network boundary — and what never does

How VizBolt collects data without your data leaving your networkDiagram: inside the customer network, a read-only collector queries the Tableau repository and writes a summary file to disk. The customer previews the file, and only after approval is it sent to VizBolt for analysis. User names, workbook names, connection strings, query text, and row-level data are never collected and cannot cross the boundary.YOUR NETWORKVIZBOLTTableau repositoryworkgroup db · readonlySELECT onlyCollectorplain SQL + Pythonno internet accesswrites to diskvizbolt-summary.json312 KB · counts & percentilesYou previewthen decidesent only afteryou approveAnalysisrules · benchmarksNever collected — so it cannot cross:✗ user names & email addresses✗ workbook & project names✗ connection strings & credentials✗ query text & row-level data

Collected — as aggregates

  • Job outcomes, durations, and queue waits, bucketed by hour
  • View load-time percentiles (TP50 / TP95) by view class
  • Extract sizes and last-access age bands
  • Session and peak-concurrency counts
  • Schedule frequency and overlap density
  • Object totals by type — counts only

Never collected

  • User names and email addresses
  • Workbook, project, and data source names (codes optional)
  • Connection strings and credentials
  • Query text of any kind
  • Any row of business data
  • Free-text fields — comments, descriptions, tags

The collection, in three steps

01

The collector

Reads your repository. Plain, readable scripts your team reviews before anything runs — no compiled binary, no agent, no network access.

02

The Environment Pack

Records your platform's shape: topology, version, licensing, key settings, and per-node hardware. Same rules — plain-text output, runs on your machine.

03

You preview, then send

One summary file of counts, durations, and percentiles. Nothing that identifies people or content. You read every number, then sending it is a decision you make — not a default.

read-onlyno network accessaggregates onlypreview before send

Want the detail? The full method — including what each script reads — is published on the method page.

What you get back

Every finding is labeled measured, inspected, or attested — so you always know how we know.

Scored report

Every finding and its recommendation across the six pillars, each citing its evidence, under a Health Score always shown with its data coverage.

Executive deck

The findings that matter, in the order they matter, written for the people who fund the fixes rather than the people who run the platform.

Tableau workbook

Your aggregates, explorable in the tool your team already uses — filter, drill, and re-check every number behind every finding yourself.

Prioritized action list

Owner-ready recommendations, ordered by impact against effort.

Coverage report

Exactly what was measured, over what window — and what was not, stated plainly so no finding claims more history than it has.

Tableau Server and Tableau Cloud

The method is the same; only the source changes. Tableau Cloud has no repository, so the collector reads the equivalents Tableau publishes for it.

data source

Tableau Server
The repository (workgroup database)
Tableau Cloud
Admin Insights and the Metadata API

access needed

Tableau Server
The repository's readonly user, enabled by your administrator
Tableau Cloud
A site-scoped access token you create and can revoke

what runs

Tableau Server
SQL you can read, against repository tables
Tableau Cloud
Python you can read, against published APIs

history read

Tableau Server
Whatever your retention settings have kept
Tableau Cloud
What Admin Insights retains for your site

output

Tableau Server
The same aggregated summary file
Tableau Cloud
The same aggregated summary file

Questions administrators ask

Do you need access to our systems?

No. Your administrator runs the collector inside your network; we never connect to your environment. On Tableau Server it uses the repository's readonly user. On Tableau Cloud it uses a site-scoped token that you create and can revoke at any time.

What if we can't send anything outside our network?

Nothing is ever sent automatically. The collector writes one file to your disk and stops. If policy blocks even the aggregated summary, the file can go through whatever data-transfer approval process your organization already uses — the engagement waits for it.

Will it affect our server's performance?

The collector runs read-only queries against the repository, not against VizQL or your dashboards. The queries are windowed and take minutes. If you prefer, run it outside business hours — the numbers are the same.

Does it work with Tableau Cloud?

Yes. Tableau Cloud has no repository to query, so the collector reads Admin Insights and the Metadata API instead. The output is the same aggregated summary file, and the same preview-before-send rule applies.

How much history do you need?

More history means better trend findings, but the collector simply reads what your retention settings have kept. Ninety days of job history is ideal; thirty is workable. Failure concentration shows up even in short windows.

Who owns the output?

You do. The report, deck, workbook, and action list are yours to share inside your organization as you see fit.

Next step

See what your environment is actually doing

Start with the free Health Snapshot: run the collector, preview the file it writes, and get your top five findings within a business day.

Get the free snapshotContact us