Closed GitHub issues (pre-Backlog history index)

The record of work closed before backlog/ became this repository’s tracker, kept as one document so the history is readable from the checkout alone and so the original ID spaces stay the only ones. Compiled 2026-08-14 during the migration.

Backlog task IDs follow creation order, so importing any of this as Done tasks would have created a second ID space over the same history - an FMO-nnnn that can never be made to match the #N or phase number already cited in commit messages, code comments and the docs. These rows point at the originals instead.

Where the full detail still lives

• GitHub issues - bodies and comments are live at gh issue view <N> -R rknightion/fleet-management-operator. The issues were not deleted, so this section is a pointer, not a replacement.
• .planning/ - the phase tracker was deleted from the working tree in the migration commit. Recover any of it with git show <sha>:.planning/<path>; the tree is intact at 72499c5 (2026-02-09), the last commit that touched it.
• docs/superpowers/ - gitignored, still on disk, never committed. The production-readiness scorecard named below is not recoverable from git on another machine.

Closed GitHub issues

Two. Everything numbered #1-#7 in this repository is a pull request, not an issue - gh issue view <n> silently renders a PR when the number is a PR, which makes the issue set look larger than it is.

#8 was itself the non-breaking follow-up to PR #3 (“Harden pipeline naming to namespace scope”), which was closed unmerged because always deriving the Fleet pipeline name from namespace.name is a silent breaking change - Fleet upsert is name-keyed for idempotency, so the rename orphans the existing pipeline rather than moving it. That reasoning is the reason the feature is opt-in today, and it is the kind of thing that gets re-litigated if only the outcome survives.

#9 resolved the mirror question for PipelineDiscovery / CollectorDiscovery: cross-namespace mirroring is a kept, intended capability, documented as a trust boundary rather than blocked. PR #6 had proposed hard-blocking it; that was rejected.

The one open issue, #46 “Dependency Dashboard”, is Renovate’s own bot-maintained issue. It is not work, it is recreated on every Renovate run, and it is deliberately not represented in backlog/.

.planning/ - v1.0.0 and v1.1 phase tracker (2026-02-08 to 2026-02-09)

7 phases, 12 plans, all complete. Milestone consolidated as v1.0.0 at 6704801; v1.0 closed at b2d223a, v1.1 closed at 3fcc31b.

What it left behind that is still load-bearing in the codebase, and should not be “cleaned up” by someone who does not know why it is there:

• Zero List() calls in reconcile paths - all reads go through the informer cache, enforced by AST-based verification tests, not by convention.
• The grep-able comment prefixes Cache:, Reconcile: and Watch:. They exist so audit tooling can find every justified API call; they are not decorative.
• SyncPeriod deliberately absent from ctrl.Options, and the four documented return patterns for backoff.
• updateStatusError preserving the original error so exponential backoff works, and the single-retry guard on 404 recreation that stops infinite recursion.

Two closed items that were re-opened by later work - noted because the phase tracker records them as done and it is misleading on its own. loggerFor() was flagged as dead code in the v1.0 milestone audit, and phase 1 never produced a VERIFICATION.md. Both were accepted as low-risk tech debt at the time.

docs/superpowers/ - production-readiness audit (2026-04-28)

docs/superpowers/audits/2026-04-28-production-readiness-scorecard.md, against source commit 013f042. Calibration: single tenant, 30,000 collectors, thousands of pipelines, single replica, 3 req/s Fleet API limit. Roughly 70 findings across 10 categories, executed as a 10-way parallel fan-out with an aggregation pass.

Outcome: every category GREEN, no S1 findings ever raised. The remediation log in that file is the authoritative per-finding record, with a SHA on most rows. Headline fixes that changed production behaviour: Fleet API burst made configurable and defaulted to 50 (2e8aaf5, 7124397, 6a7ba07); status.matchedCollectorIDs / ownedKeys capped at 1000 with a Truncated condition (62cf4a3, c801c66); selective Collector watch via IndexField on .spec.selector.matcherKeys; per-controller MaxConcurrentReconciles (d9ba8ef); chart memory limits raised from 128Mi to 1Gi; Fleet API metrics, rate-limiter wait histogram and OTEL tracing added; Grafana dashboard, PrometheusRule alerts, five per-alert runbooks, docs/troubleshooting.md, docs/versioning.md and docs/webhook-setup.md created.

Two findings survived and are now live tasks, not history - they were recorded only in this gitignored file, invisible to any other clone or CI run, which is the specific reason they were promoted during the migration:

• PERF-02 - CollectorDiscovery pagination, blocked on the Fleet SDK.
• API-09 - scale subresource, deferred to v1 graduation.

Find them with backlog task list --plain.