Think² logoThink²
Sign in

Guides (Spanish)

RPA or API: which dependency can you maintain and verify?

A query can return HTTP 200 without the data you expected. We built a synthetic page and API to retrieve D-17: renaming a selector broke the interface read; renaming a JSON field broke the API read. This does not establish a better route. It shows that both need a verifiable contract.

Waiting list; no immediate access or free plan guaranteed. The destination is in Spanish.

Situations to review

The interface changes while the API retains its expected field.

In selectorchange, we renamed the HTML id from status to new-status. The browser still queried #status and returned null, failing the contract despite HTTP 200. The API retained reference:D-17 and status:PENDING.

The demonstrated issue is dependence on one particular selector. We did not test every RPA approach or establish that a different locator would fix this scenario.

The API changes its schema while the interface retains the expected data.

In schemachange, we renamed status to state in JSON. The 200 response contained PENDING but not under the required key. The API-read contract failed; the DOM query retained D-17/PENDING.

Similar text under another key does not automatically satisfy the contract. Use the documented version to decide whether to adapt, migrate or stop rather than inventing a field mapping.

The service returns 401 before usable data is obtained.

In denied, the fixture returns 401 for both routes. The API read obtained error:unauthorized; after its HTTP precheck detected the status, the UI route did not continue to the browser. Both checks failed.

This is a simulated 401, not a test of real authentication, expiry, login or account permissions. Do not switch routes to bypass denied access. Business approval does not create credentials.

Waiting list; no immediate access or free plan guaranteed. The destination is in Spanish.

How to evaluate the work

  1. Choose by need and authorized access: a documented API may fit when it exposes the required fields and operations with known versions, permissions and limits; an interface may be necessary when the workflow is available only there and automated use is permitted. Neither grants permission merely by existing. Consider which contract you can maintain, not a universal promise of lower costs or greater speed.
  2. Contract fixed before the test: reference must be D-17 and status must be PENDING, with HTTP 200. Baseline: both pass. Selector change: API passes / UI fails. Schema change: API fails / UI passes. Denied 401: both fail. These are eight route-by-scenario checks, not exactly eight HTTP requests or a production success rate; prechecks and browser traffic also occur. Changes were deliberately injected, not observed in a commercial service.
  3. Reproduction: our script starts a fixture on 127.0.0.1 using an ephemeral port, without customer systems. Python reads JSON over HTTP; an isolated headless Chromium queries the DOM through CDP and returns document.querySelector('#status').textContent or null. The UI route runs an HTTP precheck and skips the browser on 401. Three 200 DOM snapshots, results, protocol and scripts are preserved. There were no clicks or desktop-app operations: this is a narrow interface-reading pattern, not an end-to-end RPA test. Playwright was not executed.
  4. Selection and maintenance record: purpose; data/operation; API version or UI dependency; access; documented limits; contract; failure and evidence; owner; proposed change. Stop incomplete results instead of recording success from 200 alone. Think² can organize the assignment with collaborators, groups/context, compatible models and authorized tools, with an optional computer per collaborator. This test does not establish connectors, RPA browsing or a Think² API. Keep your current workflow if it already meets the need. Any external actions require applicable authorization; this test only read synthetic data.

Download the evidence

Frequently asked questions

Does API mean guaranteed stability while RPA means fragility?

No. We broke one dependency of each route separately in the fixture. An endpoint can change its contract, version or access; an interface can support different locating strategies. The result does not measure change frequency or actual maintenance effort.

Does HTTP 200 establish that I retrieved the correct status?

No. In the injected changes, the server returned 200 but the expected selector or status key was missing. Our contract checks HTTP status, reference and field; another system requires its own contract and source checks.

Does this test compare speed, cost, service limits or real permissions?

No. Four local synthetic scenarios do not measure latency, cost, load, provider limits or account permissions. The 401 is simulated and does not include a tested authentication flow. The guide requires checking those prerequisites before choosing rather than filling gaps with assumed numbers. The simulated 401 does not implement WWW-Authenticate or establish complete authentication-protocol compliance.

Sources and scope

  • Playwright Locators — Locator recommendations; Playwright was not executed; no guarantee of stability.
  • MDN HTTP401 — HTTP 401 semantics; incomplete fixture for control purposes, not an authentication test.

Keep exploring

    Waiting list; no immediate access or free plan guaranteed. The destination is in Spanish.