The old system traced every service to OTel and kept a queryable audit log; the new modules have neither. That is the difference between a system you can operate and one you can only watch.
Add OpenTelemetry tracing through the same injected logger paths (spans per run, per model call, per connector call, per approval), and an audit log owned by one module that records who did what and what the system did on whose behalf, independent of the decision trail it can be read from.
Done when: a run appears as one trace through the collector, and the audit log answers who approved an action, what changed a setting, and when a connector was installed.
The old system traced every service to OTel and kept a queryable audit log; the new modules have neither. That is the difference between a system you can operate and one you can only watch.
Add OpenTelemetry tracing through the same injected logger paths (spans per run, per model call, per connector call, per approval), and an audit log owned by one module that records who did what and what the system did on whose behalf, independent of the decision trail it can be read from.
Done when: a run appears as one trace through the collector, and the audit log answers who approved an action, what changed a setting, and when a connector was installed.
sam
added the featurehigh labels 2026-11-16 01:24:27 +00:00
Part of epic #1977. The dependency could not be recorded: this Gitea's dependency write endpoint answers 404 and the connector that used to add it is disconnected, so this comment carries the link.
Part of epic #1977. The dependency could not be recorded: this Gitea's dependency write endpoint answers 404 and the connector that used to add it is disconnected, so this comment carries the link.
Strengthened per the maintainer on 2026-11-16: audit is a compliance requirement, not a debugging aid. ARCHITECTURE.md 2.10 now says what is recorded for every change and mutation on the platform (not only runs), with who or what made it, what changed, why (cause, verdicts and reasoning, the approval and the evidence seen), and the outcome; the payload and response of every call that changed something; queryable and exportable; append-only by grant and chained so a gap is detectable; retention per kind, with configuration changes kept for the life of the organisation. The decision trail is the run-scoped view of the same evidence. Recorded eval results and advisory standing belong here too, tied to the extension digest.
Strengthened per the maintainer on 2026-11-16: audit is a compliance requirement, not a debugging aid. ARCHITECTURE.md 2.10 now says what is recorded for every change and mutation on the platform (not only runs), with who or what made it, what changed, why (cause, verdicts and reasoning, the approval and the evidence seen), and the outcome; the payload and response of every call that changed something; queryable and exportable; append-only by grant and chained so a gap is detectable; retention per kind, with configuration changes kept for the life of the organisation. The decision trail is the run-scoped view of the same evidence. Recorded eval results and advisory standing belong here too, tied to the extension digest.
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
The old system traced every service to OTel and kept a queryable audit log; the new modules have neither. That is the difference between a system you can operate and one you can only watch.
Add OpenTelemetry tracing through the same injected logger paths (spans per run, per model call, per connector call, per approval), and an audit log owned by one module that records who did what and what the system did on whose behalf, independent of the decision trail it can be read from.
Done when: a run appears as one trace through the collector, and the audit log answers who approved an action, what changed a setting, and when a connector was installed.
Part of epic #1977. The dependency could not be recorded: this Gitea's dependency write endpoint answers 404 and the connector that used to add it is disconnected, so this comment carries the link.
Strengthened per the maintainer on 2026-11-16: audit is a compliance requirement, not a debugging aid. ARCHITECTURE.md 2.10 now says what is recorded for every change and mutation on the platform (not only runs), with who or what made it, what changed, why (cause, verdicts and reasoning, the approval and the evidence seen), and the outcome; the payload and response of every call that changed something; queryable and exportable; append-only by grant and chained so a gap is detectable; retention per kind, with configuration changes kept for the life of the organisation. The decision trail is the run-scoped view of the same evidence. Recorded eval results and advisory standing belong here too, tied to the extension digest.