Thorlabs Blender Optical Path

OpticalModeler 2.0 — Thorlabs Blender Optical Path

Turn a measurement requirement, schematic, or photographed apparatus into an editable, explainable Blender scene. Deliver the requested model and views with evidence that matches the claim. Manufacturer geometry, installed identity, visual readiness, and physical qualification are separate facts.

Load the relevant references

• Read physical-gates.md before geometry, CAD placement, or mechanical review.
• Read evidence-contract.md before declaring any PASS or preparing a release.
• Read history-derived-rules.md when revising an existing scene or when old fixes may have regressed.
• Read project-case-study.md for the complete G1/G2 2D-to-3D example.
• Read fresh-design-and-rendering.md when the user requests a new measurement path, uses an old example as a quality reference, excludes electronics, or reports inadequate detail/beam visibility.
• Read photo-reconstruction-and-revisions.md for photographed setups, constrained layout changes, repeated user corrections, or selectable ports/shutters.
• Read presentation-and-delivery.md for material/light changes, image reuse, interactive opening state, and final delivery.
• Read publication-privacy.md before publishing laboratory-derived guidance or examples. Keep the private identity map outside the public tree.
• Read end-to-end-workflow.md before a whole-system build or large forward test. Use its single-run ledger instead of splitting one system into independently authored modules.
• Read multi-run-qualification.md before a Skill release or scale/stress campaign that must compare multiple fresh whole-system runs.

Published detailed prompts, correction loops, and final handoffs: six walkthroughs, English dialogues, and 中文交互案例. They are edited teaching examples, not raw conversations or a new qualification campaign.

Authority and revision rules

1. Apply system constraints first, then the newest explicit user wording or annotated screenshot, then active project rules, then this general skill, and only then older artifacts or PASS labels.
2. Freeze submitted or accepted packages. Make corrections in an active revision without overwriting frozen evidence; group related fixes before submission instead of publishing a version for every preview.
3. Keep one writer, run ID, revision root, deterministic generator lineage, saved-Blend lineage, and workflow ledger for the whole system. Use helpers only as read-only reviewers unless an isolated non-shared asset scope is explicit; never stitch independently authored modules into a whole-system claim.
4. Translate every correction into object families, fixed and editable degrees of freedom, world-space geometry, a measurable gate, and required evidence before editing. "Make it compact" does not authorize changing the optical order, internal spacing, or an accepted upper endpoint.
5. Use BLOCKED or UNVERIFIED when real geometry or evidence is missing. File existence, imported CAD, process success, labels, AABB contact, and self-reported text are not proof.
6. Use existing evidence and photographs before asking a material question. If the user requests a complete review candidate before further measurements, finish that authorized candidate with explicit unknowns; do not invent hardware identity or call it physically qualified.
7. When helpers are authorized, assign isolated scopes and keep ordinary findings in a consolidated queue read at checkpoints. A moving build/render is not stalled. External messages, uploads, and publication require the user's authorization for those actions.

Core workflow

First choose new measurement design, photograph/schematic reconstruction, constrained correction, presentation-only rerender, or audit-only work. Record which inputs are design authority, reusable assets, or visual references. A new measurement at an old example's quality starts from an empty scene. An explicitly retained reconstruction baseline can be reused read-only in a new revision. For optics-only requests, exclude electrical/data/circuit visualization while retaining optical detectors and their mechanical supports.

1. For a new whole-system build, initialize the run spec, state ledger, and append-only event hash chain with scripts/workflow_ledger.py. For a scoped correction, continue the existing lineage with a frozen input lock, declared change set, current evidence and delivery manifest; do not fabricate completion of pending whole-system stages.
2. Build a machine-readable map: schematic node -> experimental role -> real asset -> optical/fiber/electrical ports -> support path.
3. Inventory every directed edge, branch, optional/deferred node, component, beam segment, beam height, aperture, connector, and required detector endpoint. Never invent reciprocal edges.
4. Acquire official CAD only from manifest-locked manufacturer URLs into a private cache. Verify byte count and SHA-256 before atomic placement; record part number, source URL, unit scale, bbox, local optical axis, surface normal, aperture, provenance, and redistribution boundary. Mark modeled or surrogate parts explicitly and never publish vendor geometry without an explicit grant.
5. Treat each source lock and its hashed files as one atomic input bundle. Before geometry, run a producer-to-consumer artifact preflight: every locked source byte, canonical and part-qualified CAD cache key, official drawing, and required runtime must exist at the exact path/key used by the next script. Validate structured authority inputs as typed exact sets before lookup; missing, duplicate, extra, legacy, malformed, or identity-mismatched records must produce a durable structured BLOCKED result rather than an exception or silently collapsed entry.
6. Freeze deterministic source locks and replay them before geometry. Public-package scripts must reproduce their distributed normalized semantic parameters with zero unexplained field differences; encode every override in source with a reason. Private source/identity locks remain private unless separately authorized and curated. For multi-state systems, every edge has exact active_states or a hashed deterministic expansion, and every declared state has explicit ray templates.
7. Solve optical constraints first: centers, surface normals, reflection, splitting planes, branch endpoints, and zero-radius clearance. Keep design coordinates separate from measured reopened mesh/port witnesses; constant zero errors and endpoint-family matches cannot prove aperture or first-hit acceptance.
8. Solve mechanics post-first from verified table holes through real fasteners, clamps, holders, posts, mount faces, and device interfaces.
9. Fix the shared placement or transform root cause. Prove one representative repeated assembly inside the same run before propagation, then reopen and recheck every copy.
10. Render bright audit views before beauty views. Use cutaways or transparency only to expose hidden, already-measured interfaces.
11. For a whole-system claim, complete the applicable saved-scene, optomechanical, visual, export, derived-claim, sanitization, hash and rule-compliance gates in the same ledger. For a scoped correction/presentation, record current checks and exact preservation evidence while keeping missing whole-system gates pending.

Before final rendering, audit intent as well as geometry: retained relative vectors, exact mirror count and ordering, fixed endpoints, allowed support moves, and the actual requested footprint. Device centers inside a platform projection do not prove complete device/support coverage. Audit all changed-to-fixed and changed-to-changed neighbors, including fasteners owned by other nodes.