# Codex work started · sample inputs and Split Complex

16 October 2026 · PROPOSED / TEST ONLY · Active agent results and input proposal R02

## Exact path claim

Codex owns this new `docs/crown-key-production/R01/codex-inputs/` directory on `codex/crown-key-design-inputs-r01`, plus the shared plan/catch-up publishers and directory entries used for its website handoff. It contains proposed design values, source reconciliation, agent evidence and concept prompts. Claude retains schema, validation, drawing/blockout and migration ownership.

Baseline: main `aa3b95526ce18bd7c74cc0a232c62e1b0f82535b`, which published the [Quik catch-up hub](https://thelostsouls.world/production/handoffs/quik/catch-up/). Claude's new exact path claims/schema revision were not visible in the remote branch inventory when this started. Do not invent their completion or migrate these inputs into an assumed schema.

## First handoff to Claude

Read [SAMPLE_INPUT_PROPOSAL.json](SAMPLE_INPUT_PROPOSAL.json). These are deliberately authored candidate values for a disposable test scene, not measurements extracted from Crown Landmark imagery. They describe one connected ground route, one entrance proxy, one arcade-bay proxy and explicit checks. Claude should map them into the reviewed schema after his asymmetric fixture passes, preserving units, test-only status, unknowns and revision provenance. Report any field/interface mismatch first.

The values do not replace `BUILDING_DIMENSIONS.json` for any production asset. The sample has no district placement and does not prove the actual Crown Landmark entrance fits. A separate local comparison to the issued entrance geometry is still needed.

Read [SAMPLE_REVIEW.md](SAMPLE_REVIEW.md) for the three incorporated geometry fixes and the R02 recheck. [CLAUDE_NEXT.md](CLAUDE_NEXT.md) is the concrete fixture/schema handoff. The written proposal is internally coherent; all generated-geometry and engine checks remain untested.

Claude's contract-pilot PR #51 is now published. Independent review requires four focused validator/drawing corrections and a selected floor/scope mapping convention before integration. Read [TOOLING_REVIEW.md](TOOLING_REVIEW.md) and [SAMPLE_SCHEMA_MAPPING_REVIEW.md](SAMPLE_SCHEMA_MAPPING_REVIEW.md) for exact findings, tests and next inputs.

## Research running alongside the fixture

![Split Complex candidate A, a single-bridge concept study and not a measured build reference](split-complex-concept-a-r01.webp)

The first concept is generated. [IMAGE_RECORD.json](IMAGE_RECORD.json) records input roles/hashes, the prompt, method and visual review limits. The image is a review candidate; no model/camera alignment or metric test is claimed.

![Terrace Tower candidate A, a cascading terrace concept and not a measured build reference](terrace-tower-concept-a-r01.webp)

Terrace source research is complete in [TERRACE_REFERENCE_REVIEW.md](TERRACE_REFERENCE_REVIEW.md). Its first concept keeps the rounded taller rear volume and broad curved terraces. [TERRACE_IMAGE_RECORD.json](TERRACE_IMAGE_RECORD.json) and [TERRACE_IMAGE_PROMPT.md](TERRACE_IMAGE_PROMPT.md) record provenance and limits. Level count, core, hidden faces and dimensions remain proposed or unknown.

## Agents activated in this session

The sample-review agent completed the written-geometry review and R02 recheck. The Terrace-research agent completed the source/shape/core/setback brief. The tooling-reconciliation agent checked current branches/PRs and identified Claude's unpublished generic-fixture dependency. [EXECUTION.json](EXECUTION.json) records the dated results and next gates. These are completed bounded tasks, not a live external agent monitor.

Read [SPLIT_REFERENCE_REVIEW.md](SPLIT_REFERENCE_REVIEW.md) and [IMAGE_PROMPT.md](IMAGE_PROMPT.md). The first image is a concept study of the catalog's one-bridge reading. The two-level location-render interpretation remains a separate candidate. No bridge choice, new floor count, core size, placement or final shell has been selected.

## Next gates

1. Claude returns schema/fixture evidence, actual tool versions and exact path claims.
2. Codex and Claude reconcile these proposed inputs with that schema and the local entrance dependencies.
3. Generate reproducible drawings and blockout from the same test inputs; keep real holes distinct from markers and reservations.
4. Quik runs the isolated import/traversal test and returns the catch-up report, with actual measurements and evidence.
5. Revise the package standard, then advance Split Complex and Terrace Tower. Concept research can continue while those gates are open.

No message was sent to Claude or Quik; these are shared repository handoff files.
