# Integration Pulse

> Audit what an external developer can find out about their own failing integration, and find whose fault each
> breaking integration is. Classify the integration support tickets and paste each integration's numbers; the browser
> checks every failure class against what integrators see today, splits error rates into 4xx, 5xx and 429 against each
> integration's own baseline, catches silent integrations, reads blame against your deploys and their releases, ranks
> the breaking ones by installs and reads the gates. Metered runs write the visibility design, and the outreach. Built
> on the SkillSafe skill @samber/integration-error-observability, from samber/developer-platform-skills (MIT), by Samuel Berthe.

https://integration-pulse.skillsafe.ai/

## What is free
- The audit: the ticket pile by failure class (synchronous errors, webhooks that never arrived, auth, rate limits, silent data mismatch, an integration that stopped calling) against the surface integrators have today, the self-serve baseline, and which classes no surface lets them diagnose; failure counts alone never count.
- The signals: each integration's 4xx, 5xx and 429 rates against its own baseline, breaking at twice its baseline and five points above it, or silent when calls fall under a tenth of baseline; blame - ours after our deploy or on 5xx, theirs on 4xx (especially after their own release), neither on 429; breaking integrations ranked by installs.
- The rungs: delivery log, request log, replay, request-ID echo, searchable IDs, propagation, digest, threshold alerts with a cooldown, state-change alerts, the fleet view.
- The gates: failure-class coverage, request-ID resolvability, tagging at ingestion, retention against time to notice, no alert per failure; plus an audit table, a fleet view (.csv), the checks (.csv) and everything (.json).

## The metered lanes
Two tasks on gpt-terra, metered per run, routed by `task`. Input string fields: `task` (surface or outreach), `platform`, `signals`, `carry` (outreach only: the design's decisions) and `facts`.
- `surface` replies with `summary`, `audit` (the gap per class), `rungs` (ten, as ranked), `log_fields`, `correlation`, `buckets` (4xx, 5xx, 429), `notification`, `redaction` and `caveats`.
- `outreach` replies with `summary`, `contacts` (each breaking integration with its cause, channel, message and next step), `ours` and `caveats`.
The page holds every reply to its own facts: the gap per class, the ranked rungs, a full log row, a searchable request ID, three separate buckets, the ranked posture with a cooldown, redaction and no reliability promise; the breaking integrations in order with the page's causes, our faults owned, rates only by family.

## Links
- API tutorial: https://integration-pulse.skillsafe.ai/api.html
- Source skill: https://skillsafe.ai/skill/@samber/integration-error-observability