test(cyborg): audit example pack against RAES and CAGE-2

[Brad-Edwards]

Goal

Audit the installed cage2-research example pack delivered by #21 / PR #78 against both the published RAES contracts and the original CAGE Challenge 2 Scenario2 evidence.

Audit baseline

• The RAES package and published contracts used by the adapter, with exact versions recorded.
• cage-challenge/cage-challenge-2 at the qualified upstream commit 26ce1c1253fa9e2e73f25e6a7f2da32860c11257.
• The public CAGE-2 evaluation description and the repository qualification ledger.
• The installed raes_adapters.cyborg example-pack artifact, not only its source checkout.

Work

• Build a clause-by-clause evidence matrix mapping pack files and claims to RAES contracts, validators, and architecture constraints.
• Compare topology, participant roles, action/observation boundaries, reward and evaluation semantics, runtime assumptions, and reproducibility claims with original Scenario2.
• Identify omissions, approximations, semantic losses, unsupported equivalence claims, provenance gaps, and native-simulator leakage risks.
• Exercise the published pack validation/release commands and the CybORG qualification path against a clean installed distribution.
• Distinguish the bounded smoke example from the complete Scenario2 SDL work tracked by docs(cyborg): author the validated CAGE-2 (Scenario2) SDL document #77 without duplicating that implementation.

Acceptance criteria

• Exact RAES and upstream CAGE-2 versions/commits are pinned in the audit record.
• Every portable pack artifact and material claim has traceable RAES and upstream evidence, or an explicit documented limitation.
• Scenario topology, roles, actions, observations, rewards, evaluation protocol, and runtime behavior are compared with source citations.
• Pack content CI, release checks, adapter qualification, and relevant clean-install tests pass with recorded commands and outcomes.
• Findings are classified as contract violations, semantic losses, documentation gaps, or intentional bounded simplifications, with concrete remediation decisions.
• The audit states which results belong to this example pack and which are satisfied by docs(cyborg): author the validated CAGE-2 (Scenario2) SDL document #77.
• No native CybORG objects, hidden truth, reward vectors, raw logs, action identifiers, credentials, or full tracebacks enter portable RAES artifacts.

Related work

• feat(cyborg): add a CAGE-2 example environment pack #21
• docs(cyborg): author the validated CAGE-2 (Scenario2) SDL document #77
• PR feat(cyborg): complete CAGE-2 example pack #78