docs: version-context: monitoring.md's 'Unreleased: current main' zero-publication paragraphs state main directly

[github-actions]

What this PR does

On the working Monitoring page, are the start-up zero-publication facts for the overload-shed, retry and circuit-breaker families still framed as 'Unreleased: current main publishes ...', or stated directly as the selected version's contract now that the site version selector identifies the documented version?

Why we need it

Source change: smg-project/smg@0ef57c8

Documentation targets SMG main; the site version selector supplies version context.

Version-context cleanup verified against the pinned main checkout (baseline commit 1, 0ef57c89, the newest supplied source commit). One page and topic: getting-started/monitoring.md's two 'Unreleased: current main' paragraphs, the monitoring-page twins of the paragraphs the pending metrics.md cleanup (version-context-metrics-zero-publication) already covers on reference/metrics.md. monitoring.md:211 — 'Unreleased: current main publishes smg_worker_overload_shed_total at zero for every stage from start-up... only the gauge still waits for protection to flag a worker'; :520 — 'Unreleased: current main publishes the overload-shed and retry families at zero from start-up... and each worker's circuit-breaker transition counters at zero when that worker is registered. ... On a gateway built from main, absence of the start-up families points at a scrape problem...'. The selector and banner identify the SMG version and release snapshots keep their own text, so each paragraph is rewritten as a direct statement (drop 'Unreleased:'/'current main'/'built from main' framing), and the general rule at :512 ('A series appears only after the first event that records it, so a metric can be absent on an idle gateway') is reconciled with the eager exception in the same pass rather than contradicted by a trailing unreleased note. Retained claims re-verified at the pin: observability/metrics.rs:1019-1069 init_startup_series — smg_worker_overload_shed_total at zero for stages selection/dispatch/pd_admission, the selection-stage fallback counters smg_worker_overload_fallback_total and smg_worker_liveness_fallback_total at zero, smg_worker_retries_total and smg_worker_retries_exhausted_total at zero for worker_type regular/prefill/decode times the endpoint values (chat, generate, completions, rerank, responses, decisions, systemone, messages, audio_transcriptions, other), and the smg_worker_retry_backoff_seconds attempt="1" series registered — so the external-provider (worker_type="external") exception in :520 stays true, as does 'only the gauge still waits' (smg_workers_overloaded is not pre-published; no workers_overloaded write in init_startup_series); each worker's five breaker-transition series publish at zero at registration (worker/circuit_breaker.rs init via Metrics::init_worker_cb_transition — the writer re-verifies the exact call sites before rewriting). Strict boundaries: do NOT touch :209 or :514's 'admission metrics need --max-concurrent-requests' preconditions — that functional correction is the release-deferred admission-default concern (closed PR #152), blocked on the held reference/configuration.md, and must not be folded in or contradicted; do NOT touch :86 or :506 ('Requests to gRPC workers do not carry trace context...') — a functional trace-context correction from commits outside this scan's focus (smg#2936/smg#3126, another shard's; noted in remaining_work); :185 and :446 are the pending removed-worker-series-dropped concern's lines — untouched. Coordination: the pending version-context-metrics-zero-publication concern rewrites the metrics.md twins (:11, :749, :786, :818) — same topic on a different file, and the cross-page links stay valid either way; land one, rebase the other; several pending observability concerns touch other monitoring.md lines — if any of their PRs is open when this is attempted, defer per the blocked-page rule. Dedup: merged #135 wrote these paragraphs correctly under the earlier release-pinned regime; no open, merged or declined PR removes the narration; no open PR holds monitoring.md. Two paragraph rewrites plus the :512 reconciliation, far under 999 lines.

Scope: observability / version-context-monitoring-zero-publication. Other concerns are deferred.

How to test

• Passed the documentation path and size guard (under 1000 added plus deleted lines; no file-count limit).
• Passed an independent accuracy and single-concern review.
• Passed git diff --check and pnpm check and the production pnpm build.

Checklist

• Tests added/updated (if applicable)
• Docs updated (if applicable)
• Commit includes a DCO sign-off

[coderabbitai]

Important

Review skipped

Bot user detected.

To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

• Configuration used: Organization UI
• Review profile: CHILL
• Plan: Advanced
• Run ID: f221fcf8-ca09-4a86-95bc-b4ee62a480b2

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

_{Comment @coderabbitai help to get the list of available commands.}

[github-actions]

Documentation maintenance: ready. Unsuccessful content rounds: 0/3. Operational attempts: 0/3.

Head: 480445f112899305f2800a1db725e51ca8bf5413; reviewed main: a3477283ec884c24b1cc4847f189813919e3f2cc.
Reviewed SMG source: 374a755da5e99f1d398fa5ef20dbfec532bb6571.

The PR is exactly the scoped version-context-monitoring-zero-publication cleanup: monitoring.md's two 'Unreleased: current main' paragraphs (:211, :520) rewritten as direct statements and the :512 general rule reconciled with the eager exception, three changed lines in one file. All claims verified at the pinned checkout (which contains baseline 0ef57c89's muse alias, so it is at/past the baseline; for version-context-* the commit is a baseline only): init_startup_series (observability/metrics.rs:1019-1069, called unconditionally after the recorder installs, server.rs:1583) zeroes smg_worker_overload_shed_total for the only three stage constants selection/dispatch/pd_admission (worker/overload.rs:54-61) plus the fallback counters and smg_worker_retries_total/smg_worker_retries_exhausted_total for regular/prefill/decode across the endpoints, independent of overload protection; smg_workers_overloaded is written only on flag transitions (registry.rs:741, metrics.rs:2123-2126), so 'only the gauge waits' holds; each worker's five breaker-transition counters publish at zero at breaker construction during worker build (circuit_breaker.rs:130-143, builder.rs:382); external-provider retries use worker_type="external" (external_router/src/metrics.rs:123, openai/chat.rs:244,251) and increment only on events, so the first-event exception holds; metrics_eager_registration_test.rs asserts this exact contract. Boundaries respected: :86/:506 (trace context), :209/:514 (admission preconditions) and the removed-worker lines are untouched; no version framing, PR/code references or failure narratives remain in monitoring.md. The #overload-protection-metrics anchor resolves (metrics.md:723). The reference/metrics.md twins (:11, :749, :786, :818) state the same facts and belong to the separately tracked pending concern version-context-metrics-zero-publication, so they are not a missing same-concern correction; circuit-breakers.md and overload-protection.md are consistent, and control-plane-auth.md:110's 'current main' is an unrelated JWT-auth concern. Feedback contains no threads, so none are addressed.

Workflow evidence

Human review threads and CODEOWNER approval remain under repository policy.