# Portfolio content audit

This audit covers the public portfolio copy and the sanitized evidence artifact. Implementation claims were checked against the V1 source, tests, docs, and archived run. Personal origin and availability wording follows the project-specific narrative/status supplied by the owner.

| CLAIM | SOURCE | STATUS |
|---|---|---|
| AI can propose work, but Houdini execution and deterministic validation decide whether a candidate passes. | `src/autonomous_houdini_developer/v1_orchestrator.py`; `verification.py`; `validator.py` | VERIFIED as the V1 design and acceptance rule |
| The example task is Box → RBD Material Fracture → Null. | `src/autonomous_houdini_developer/tasks/rbd_development.py`; archived `output/run_result.json` | VERIFIED |
| The first candidate executed in Hython, then failed deterministic checks because both required links were false even though geometry cooked. | Archived run `output/run_result.json`, attempt 0 | VERIFIED |
| The repair restored the Box-to-Fracture and Fracture-to-Null connections in `candidate_rbd.py`. | Archived run `output/repair_attempt_1.json`; `output/run_result.json` changes | VERIFIED |
| Post-repair immediate validation passed; regression passed with 1 passed, 0 failed, 0 inconclusive; verification was VERIFIED. | Archived `output/verification.json`; `output/regression.json` | VERIFIED |
| V1 release run passed. | Archived `output/run_result.json` | VERIFIED |
| Candidate SHA-256 is `306c6a6c3389030927ec713a36866c3fd4c1ec71ff0a89bce6ffb023b6c08ccb`. | Archived `output/verification.json`; `output/run_result.json` | VERIFIED |
| The supplied terminal capture shows VERIFIED, immediate validation PASS, regression PASS (1/0), candidate verification VERIFIED, and V1 release PASS. | `portfolio/media/v1-verification-terminal.png`; matching archived `runtime/runs/20261003T190816Z_Autonomous_RBD_Development_bcab598c/output/run_result.json` | VERIFIED; this is a separate run from the canonical 17:17 run above. Its run ID and hash remain as displayed in the original image. |
| The supplied deterministic validation capture shows VERIFIED/PASS, regression counts 1/0, required node and geometry checks, Houdini version 22.0.368, and piece/point/primitive counts. | `portfolio/media/verification-evidence.png`; archived `runtime/rbd_autonomous_repair/evidence/rbd_autonomous_repair_result.json` and corresponding run record | VERIFIED; capture is from a separate repair run, not the canonical release run. |
| The repair screenshot shows a diagnosis and patch connecting Box → Fracture → Null in `candidate_rbd.py`. | `portfolio/media/repair-evidence.png`; `portfolio/media/repair-evidence.txt`; RBD repair implementation and archived repair evidence | VERIFIED as the displayed repair operation; it is presented as a related evidence capture, not attributed to the canonical run. |
| The Houdini network captures show the broken and repaired network states and the fractured geometry. | `portfolio/media/houdini-broken-state.png`; `portfolio/media/houdini-final-verified.png` | VISUALLY VERIFIED; original supplied screenshots embedded without alteration. |
| The final regression breadth shown is one configured RBD case. | `src/autonomous_houdini_developer/tasks/rbd_development.py`; archived `output/regression.json` | VERIFIED; explicitly limited to one case |
| Hython is the real Houdini execution environment used in the run. | `src/autonomous_houdini_developer/houdini_executor.py`; archived run execution fields | VERIFIED |
| Houdini Inspector is read-only and reports node structure, connections, parameters, and geometry. | `src/autonomous_houdini_developer/houdini_inspector.py` | VERIFIED |
| Repair preflight checks allowed target file, syntax, observed parameter schema, and validator protection. | `src/autonomous_houdini_developer/patch_preflight.py`; repair tests | VERIFIED |
| Patch application stages changes and syntax-compiles modified Python before commit to the candidate. | `src/autonomous_houdini_developer/patch_applier.py` | VERIFIED |
| The verifier requires immediate validation PASS, full configured regression PASS, and captures a candidate hash. | `src/autonomous_houdini_developer/verification.py` | VERIFIED |
| Candidate snapshots use SHA-256; backups and change reports are supported. | `src/autonomous_houdini_developer/code_manager.py` | VERIFIED |
| Failure Lab implements `BROKEN_CONNECTION`. | `src/autonomous_houdini_developer/failure_lab.py`; `tests/e2e/test_failure_lab.py` | VERIFIED |
| Failure Lab implements `INVALID_PARAMETER`. | `failure_lab.py`; `tests/e2e/test_failure_lab_invalid_parameter_autonomous_repair.py`; `test_hallucination_defense_autonomous_recovery.py` | VERIFIED |
| `MISSING_INPUT`, `BROKEN_OUTPUT`, and `REGRESSION` injectors are operational. | `failure_lab.py`: each injector raises `NotImplementedError` | FALSE; portfolio marks these RESERVED |
| Failure Lab executes Houdini or performs repairs. | `failure_lab.py` docstring and implementation | FALSE; it mutates a candidate and records fault evidence only |
| Test baseline is 18 passed, 0 failed. | Previously established baseline in project brief; existing test suite scope checked under `tests/` | ACCEPTED EXISTING BASELINE; suite not rerun in this environment |
| The 18 tests prove arbitrary Houdini workflows are supported. | No implementation, test, or documentation supports this inference | FALSE; portfolio says the scope is bounded |
| V1 is a completed experimental milestone. | Archived final verified V1 release run and task-specific portfolio direction | VERIFIED as the demonstrated milestone; not a claim of universal capability |
| V1 demonstrates bounded RBD development/repair, Houdini execution, deterministic validation, inspection, constrained repair, regression, verification, and provenance. | RBD task adapter; orchestrator; executor; inspector; preflight/applier; regression; verifier; archived final run | VERIFIED within the demonstrated task scope |
| V1 establishes universal autonomy, all Houdini workflows, production deployment, multi-DCC, broad Pyro/VDB support, or complete Failure Lab coverage. | `docs/V1_SCOPE.md`; `failure_lab.py`; implemented task and tests | FALSE; explicitly listed as out of scope or future research |
| The project is personal R&D and is not offered for purchase, public download, or external testing. | Owner-provided project positioning for this page | CURRENT STATUS COPY; not inferred from code |
| The personal origin story describes a progression from Houdini workflow complexity to Python automation and then AI experimentation. | Owner-provided project-specific narrative direction | APPROVED STORY CONTEXT; not presented as a source-code claim |
| V1.1 and V2 roadmap topics are implemented or promised. | No such implementation evidence; roadmap uses potential/future wording | FALSE; directions are labeled future research only |
| Screenshots show authentic Houdini UI or terminal captures. | No screenshots were supplied in inspected evidence | NOT PROVIDED; only conceptual network illustration and clearly marked screenshot placeholders |
| Demo video is available. | No video supplied | NOT PROVIDED; placeholder only |

## Privacy and evidence provenance

The public HTML, CSS, JavaScript, and `evidence/final-run.json` were scanned for private absolute Windows paths, usernames, API key patterns, bearer credentials, and machine-specific environment values; those text assets contain no matches. The terminal screenshot is embedded exactly as supplied and visibly contains its original local PowerShell prompt (including a machine-specific path and username). It was not altered, cropped, or hidden because the media integration instruction required using the original capture. The page does not repeat that path in HTML text or captions. The terminal and detailed verification screenshots have different run IDs from each other and from the canonical 17:17 V1 release run; captions and this audit keep those records distinct.