Fix the decision and criteria

The target is an internal application where authenticated staff list, filter, create, update, and audit records from an existing data source.

Evaluate:

  1. deployment and network boundary — 15%;
  2. data-source and credential boundary — 15%;
  3. identity and action authorization — 20%;
  4. environments, review, and release — 15%;
  5. backup, restore, and incident evidence — 15%;
  6. app and data portability — 10%; and
  7. licensing, staffing, and operating burden — 10%.

Reject any candidate that fails a hard requirement before interpreting the weights.

Normalize public evidence

Candidate Documented operating-model signal Verify hands-on
Retool Current pricing documents cloud or self-hosted availability and plan-dependent permissions, audit, source control, and external apps Exact plan, resource permission behavior, release recovery, export, and self-hosted operations
Appsmith Official introduction describes cloud or self-hosting, database/API connections, JavaScript logic, Git version control, deployment, and sharing Edition boundaries, granular authorization, secrets, upgrades, restore, and export completeness
ToolJet Deployment guidance distinguishes cloud from self-hosted operation License behavior, identity and permission depth, release promotion, backup, upgrade, and app export
Budibase Hosting documentation describes cloud and multiple self-hosted methods and makes customer operating duties explicit Permission behavior, app versus instance backup, update path, external data sources, and portability

This table intentionally avoids a numeric feature count. The documentation uses different definitions, plans, and levels of detail.

Run one identical build

In every candidate:

  1. connect to the same disposable database;
  2. create list, detail, create, and edit screens;
  3. define viewer and editor roles;
  4. deny one field and one action to the viewer;
  5. store a credential using the documented server-side method;
  6. create development and production configurations;
  7. change a field and promote a release;
  8. inspect attributable action evidence;
  9. simulate an unavailable data source;
  10. restore or revert the previous app version;
  11. export the app and representative data; and
  12. hand the result to a second operator.

Record video or screenshots, configuration exports, timestamps, error messages, plan, version, and operator effort.

Apply hard stops

Remove a candidate when:

  • a UI-hidden action remains callable by an unauthorized role;
  • production credentials must enter client-side code;
  • a builder edit can silently change the live app;
  • the buyer cannot retrieve required business data;
  • no supported backup or recovery path covers required platform state; or
  • the intended operator cannot perform upgrades and incident response.

Interpret without manufacturing a winner

A managed candidate can win when fast operation and vendor support outweigh deployment control. A self-hosted candidate can win when private network access and platform control justify the staffing. An open repository can improve inspectability without making app definitions cross-platform portable.

The output should be two candidates with unresolved questions and a completed test record. A winner is justified only after both receive the same test, version cutoff, and evidence threshold.