Skip to content

Platform module

Supplier control center

Onboarding, credentials, mapping runs and honest health data.

Design-partner previewBuilt and demonstrable; onboarding a limited number of design partners.

How does TravelCore handle a slow or failing supplier?

Searches fan out to every supplier in parallel under a per-supplier time budget. A connector that exceeds its budget is actually cancelled, named in the response, and the result set is marked partial rather than the search failing. Connectors that burn an error budget are shed automatically for a cooldown, then readmitted with a clean window.

A distribution platform is only as good as its worst supplier on its worst day. The control center makes supplier behaviour visible instead of anecdotal: response time distributions rather than averages, booking success rate, error taxonomy, and the share of searches where a supplier returned nothing in time.

Onboarding a supplier is a defined path: credentials stored in a secrets manager and never in the database, sandbox verification, a content and rate mapping run, a shadow period where the supplier's results are collected but not sold, and only then production traffic.

Each connector carries an error budget. When a supplier burns through it, the platform sheds it from the search fan-out automatically rather than letting it drag every search down, and retries it after a cooldown so a recovered connector returns on its own.

The aggregation layer is real code and it runs on this page — the same modules the test suite imports, covered by 22 tests. It is also the only place on the site where all three engines work together: suppliers fan out in parallel with per-supplier time budgets, the matcher resolves their offers into unified properties, and the pricing engine turns each net rate into a selling price. Make a supplier hang and watch the result set shrink while the page keeps working.

Stated plainly:This module is built and demonstrable, and is onboarding a limited number of design partners. It is not yet carrying production volume across many customers.

Capabilities

  • Supplier onboarding workflow with a mandatory shadow period
  • Credentials in a secrets manager, never in application storage
  • Sandbox and production credential separation
  • Content, rate and room mapping runs with diff review
  • Response-time percentiles by nearest rank, so the number is a latency a request actually experienced
  • Booking success rate and error taxonomy per supplier
  • Error budgets with automatic shedding from the search fan-out
  • Per-supplier revenue and margin contribution

Run it yourself

Make a supplier slow and watch the result set, not the page, degrade

This is the only place on the site where all three engines run together: four suppliers fan out in parallel, the matcher resolves their offers into unified properties, and the pricing engine turns each net rate into a selling price. Break one and search again.

Stated plainly:The suppliers are fakes with configurable latency, not live connectors — the point is the behaviour of the aggregation around them. The fan-out, error budgets, shedding and percentile reporting are production code covered by 22 tests, in design-partner preview rather than carrying live volume.

Supplier behaviour

Break a supplier and search again

Four connectors, each with its own latency. The one that hangs will miss its budget, and after enough failures it burns its error budget and stops being called at all.

sup_alpha

120ms

sup_beta

120ms

sup_gamma

8000ms

sup_delta

120ms

600ms

A supplier that misses this is aborted and reported. It never delays the response.

0 searches in this session