shim to have agents ignore p1 and p2 in observation space

[relh]

Summary

• Default the agent obs shim to ignore inventory power tokens (inv:*:pN) and remap feature IDs to legacy ordering
• Restore legacy inventory normalization (100.0) for those remapped features
• Fix power-token detection regex so the shim is effective

Why

Inventory power tokens introduced in #4356 shifted feature IDs and broke older policies trained on the pre-split inventory encoding. We only care about counts up to 255 for agents, so remapping to the legacy scheme restores compatibility.

Changes

• ObsShimTokens now defaults ignore_inventory_power_tokens=True
• Power-token names are detected correctly and remapped to UNKNOWN
• Legacy feature ID ordering + legacy normalization restored for inventory features

Testing

• uv run cogames run -m machina_1.open_world -c 4 -v heart_chorus -p s3://softmax-public/policies/relh.machina_bc_dinky_sup.hc.1221.10b.0.35/relh.machina_bc_dinky_sup.hc.1221.10b.0.35:v2700.mpt -e 1 -s 1000 --format json

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[datadog-official]

✅ Tests

🎉 All green!

❄️ No new flaky tests detected
🧪 All tests passed

_{This comment will be updated automatically if new data arrives.
🔗 Commit SHA: 1b057ff | Docs | Was this helpful? Give us feedback!}

[relh]

• shim to have agents ignore p1 and p2 in observation space #4531 👈 (View in Graphite)
• main

This stack of pull requests is managed by Graphite. Learn more about stacking.