Operations Sentry — Skipped Scope

Documint client-side Sentry HTTP plugin

Source: specification/specification.md — outbound HTTP visibility section.

What was skipped: a Sentry-aware HttpClient plugin attached to the Documint Ktor client that would add Documint-specific semantic context (template name, document kind, render outcome) to outbound spans.

Why: the JVM OTel agent already captures Documint calls as generic http.client spans with method, URL, and status code — enough to detect latency regressions and error spikes. Adding the Sentry-aware plugin is an enrichment, not a coverage gap. The verification recipe in the how-to demonstrates that Documint outbounds show up in Sentry traces today.

Tracked as: PDEV-491.

Per-component Sentry adoption in accounts-component

Source: the common-module design always intended every consumer to inherit observability.

What was skipped: actually wiring Component.build() Sentry init and the SentryRequestSession Ktor plugin into accounts-component.

Why: scope discipline. This project explicitly carved accounts-component out at kickoff and parked it under its own ticket so the project closes cleanly when operations ships.

Tracked as: PDEV-533.

Removal of the deprecated SENTRY_ENABLE_AUTO_SESSION_TRACKING env var

Source: the SDK 8.x rename of the env var to canonical SENTRY_AUTO_SESSION_TRACKING.

What was skipped: dropping the deprecated name from the operations Helm chart’s deployment.yaml.

Why: keep the migration safe. The chart currently sets both names so the SDK picks up whichever one it honours at the running version, and a future SDK bump that finally removes the deprecated alias does not regress. Removing the deprecated alias is a one-line change once we are certain no consumer still reads it.

Tracked as: PDEV-538.

Unification of partition secrets

Source: discovered during PR #459 — a pre-existing non-CDK Alpha00X-Secrets stack already holds six legacy partition credentials (Arda API key, Arda Signup, HubSpot client + token, Pylon widget, Amazon Creators API). The new CDK PartitionSecrets stack stands alongside it.

What was skipped: collapsing the two stacks into one and migrating ownership of the six legacy credentials into CDK.

Why: the legacy stack is referenced by amplify env-var injection paths in amm.sh and by direct AWS console workflows. Migrating in-flight while shipping new partition secrets multiplies the risk of breaking the four-partition deploy.

Tracked as: PDEV-541.

Review-thread follow-ups

Source: PR triage on operations#172 and documentation#94.

What was skipped: substantive review threads classified as “out of scope for the current PR” and deferred with explicit user signoff (per the project’s PR-review rule that deferred fixes need explicit user authorisation).

Tracked as: PDEV-543 (operations#172 thread) and PDEV-544 (documentation#94 thread). PDEV-544 blocked-by PDEV-543.

E2E trace-propagation regression test

Source: the verification work for PR #845 (FE trace propagation) showed empirically that the failure mode is only observable end-to-end.

What was skipped: a Playwright/E2E smoke test that asserts sentry-trace and baggage headers propagate from the browser through the BFF to the BE on a representative request path.

Why: scope. PR #845 closes with unit-level alias-parity tests; the empirical end-to-end verification proved propagation works on a deployed preview, which was deemed sufficient signal for project close.

Tracked as: a candidate Linear ticket — call out during sprint planning. Listed under suggestions.md as well.

Refactoring fire-and-forget patterns elsewhere

Source: discovered while fixing CsvUploadService.

What was skipped: an audit of all launch { ... } invocations inside Ktor request handlers across operations to find other request-scope leaks.

Why: scope. The CSV path was the known failing one and the fix landed in PR #172. A broader audit deserves a dedicated ticket.

Tracked as: candidate Linear ticket — listed under suggestions.md.