docs(traces): evict aged spans only at the live-span bound

[turnipdabeets]

Summary

Amends traces › Live span bounds so over-age spans are evicted only when startSpan finds maxLiveSpans reached. Below the bound, a span that runs longer than maxSpanAgeMs and then ends is exported.

Why. The requirement's own reason for age eviction is leak recovery at the count bound. Evicting below the bound doesn't serve that, and it drops real long spans (batch jobs, migrations): their already-exported children arrive as orphans. The loss was reproduced in a review of PostHog/posthog-python#953.

Where the two implementations stand.

SDK	When over-age spans are evicted
posthog-node	on every startSpan (current text); fixed in PostHog/posthog-js#5018
posthog-python	only at the count bound (in the unmerged traces stack ending at PostHog/posthog-python#957)

Changes

• Live span bounds: sweep at the bound, then return a no-op handle if, and only if, the bound is still reached; age eviction is forbidden at any other time; maxSpanAgeMs is described as the leak-recovery threshold; the mobile-background note is narrowed.
• New scenario: a long span below the bound is exported. The two existing scenarios still hold.
• Behavior note: leaks below the bound are no longer counted as drops or warned about until the bound is reached.
• Archived change: openspec/changes/archive/2026-09-18-evict-aged-spans-only-at-live-bound/.

Created, validated and archived with @fission-ai/openspec@1.4.1 (the version CI pins): openspec new change → openspec validate evict-aged-spans-only-at-live-bound --strict → openspec archive -y, which applied the delta to openspec/specs/traces/spec.md. All four artifacts are complete and every task is checked. openspec validate --specs --strict --no-interactive (the CI job) and git diff --check pass.

🤖 Generated with Claude Code