[valzargaming]

Frontier has a source manifest and historical design notes but no maintained research index connecting external claims to project evidence and bounded experiments. Add that index in the existing docs/research structure, a reusable entry template, the Threat Interactive rendering lead, and a controlled baseline/shared-wall-resource experiment proposal. The README links the knowledgebase; seven appended manifest rows preserve the existing CSV contract and all prior references.

The seed separates the parent-reviewed partial caption evidence from primary Godot/AMD guidance and pinned Frontier source observations at b19e99f. It records the incomplete video/outro coverage, mutable documentation, pacing-limited 523x504 habitat baseline, and the 32-wall canonical limit. Static call-path inspection identifies ten mesh/material constructions per new wall; 100/1,000-wall cases are proposed isolated presentation stress fixtures. No rendering improvement or video benchmark is claimed as reproduced.

The weekly procedure revisits changed evidence, avoids duplicates and proposes one consolidated docs-only draft PR per pass. Parent coordination owns scheduling; no automation is created here.

Validation at 3eec777:

• Cloud source reads: root AGENTS.md, existing research/design organization, manifest/matrix, relevant presentation/benchmark paths, baseline report and current CI policy.
• In-memory checks passed for 34 relative Markdown links, final newlines/trailing whitespace, ten-column CSV structure (22 rows including header), seven appended rows, preserved original manifest prefix and one README insertion.
• GitHub comparison confirms exactly six documentation files, one commit ahead of the inspected base and no runtime/configuration/workflow changes.
• Ordinary repository CI is separate: the manifest CSV selects full scope under the existing policy, despite documentation-only content. No CI, runtime or gameplay success is asserted by these source checks.
• No desktop checkout/process changes, local game/build/benchmark execution, GPU workload, paid service, external outreach, merge, release or deployment.

Remaining work: separately authorized runtime A/B evidence, moving-image readability, representative hardware/VR coverage, video publication metadata and final 49 seconds/outro verification. Keep draft.

Changelog: