[rocklambros]

What changed

This addresses all six of Ariel's code-quality comments on the trace exporter from his review of PR #286, so they ship in 0.1.4. Apart from two intended changes, behavior is the same.

1. The exporter no longer reads or writes process.env. main.ts now reads the proxy environment, prints the loopback warning and applies the exemption before the Guardian starts, the way it already owns every other variable. keepLoopbackOffProxy and loopbackProxyWarning take the environment as an argument. The package exports them, so an embedder decides whether its own environment changes.

2. The warning no longer bypasses onError, because it left the exporter.

3. otel-trace-exporter.ts is split from 1,022 lines into four files, one job each:

   • otel-mapping.ts (366 lines): the mapping loader and attribute sources;
   • otlp-file-exporter.ts (69): the file writer;
   • loopback-proxy.ts (93): the proxy handling;
   • otel-trace-exporter.ts (568): the assembly.

   The code moved verbatim. The old file re-exports every name it had, so existing imports keep working.

4. REPLAY_DETECTED_CODE and the object reader are defined once. The two copies were identical.

5. The OTLP compression type is named once instead of being built inline.

6. close() applies the same bounded drain whether trace export is on or off. Before, it special-cased the null exporter.

The 0.1.4 CHANGELOG gets one bullet for the two behavior changes.

Which issue does this implement

Refs #274

How it was checked

• New tests, each seen failing first:
  • The exporter leaves NO_PROXY alone.
  • close() drains with export off. On Bun 1.3.11 it timed out before the fix.
  • main.ts prints the warning.
  • main.ts keeps a loopback collector off the proxy. That test fails with the call removed (2 proxy connections, no export) and passes with it restored. It's gated to Bun 1.3.14 and later, because 1.3.11 ignores runtime environment edits.
• Suites:
  • bun test: 1234 pass on Bun 1.3.14, and 1231 pass with 4 version-gated skips on 1.3.11.
  • bun run typecheck is clean and bun run conformance exits 0.
  • uv run pytest gives 918 passed, and mkdocs build --strict passes.
• End to end against integration from before the change (5ee326b):
  • The walkthrough's trace output, reduced to span names, decisions, labels, attribute keys and drop counts, is identical to the baseline, with a clean environment and with hostile OTEL_* and proxy variables.
  • Under a counting proxy, steps 1 to 7 and the rerun made 0 proxy connections, and the startup warning still fires.
  • Natively, the collector config validates, a restart keeps both batches at 0600, and disabled after failure still appears.
  • The committed sample trace still passes its freshness test without being regenerated.

Base branch

• integration, because this touches the specification, schemas, a reference
  implementation, an adapter, tests, or CI

Type of change

• Reference implementation or adapter

I tested this

• I synced my branch with the base branch before opening this
• uv run pytest -v passes on my machine
• uv run mkdocs build --strict passes on my machine

Checklist

• Commits are signed off with git commit -s (required by the DCO)
• Prose follows STYLE.md
• No secrets, tokens, or internal URLs in the diff

Security

• This change has no security impact

[rocklambros]

Admin-merged on the project lead's call, so Ariel's six review comments from PR #286 ship in 0.1.4. Every check is green, and the description has the test and end-to-end evidence, including trace output identical to the pre-change baseline. @afogel, these are your comments, so a look after the fact is welcome.