Skip to content

Simulation template continuation handoff

Historical integration/qualification notes: source-specific origin exports and shared delivery sidecars described below are retired. For current authoring and delivery, use compact metadata and origin packaging.

Recorded October 12, 2026. Use this as the starting point for two separate chats; retain the original chat for debugging history. Read the linked source contracts rather than copying old tutorial notebooks.

Baseline and ownership

  • Repository: doplaydo/simulation-templates; current worktree C:/Users/eve/Documents/Codex/worktrees/simulation-template-spec, branch codex/simulation-template-spec. Do not confuse this with a GFP app branch.
  • Earlier pushed checkpoint: 1b00be5. The subsequent October 12 checkpoint on codex/simulation-template-spec includes zapFDTD, later Meep refinements and these notes. Use that checkpoint as the shared baseline; 1b00be5 alone misses them. Preserve manual notebook edits and paired .py files.
  • Fork worktrees: C:/Users/eve/Documents/Codex/worktrees/simulation-meow-femwell (codex/meow-femwell-templates) and simulation-zapfdtd-refinement (codex/zapfdtd-refinement). They initially received uncommitted copies of the same changes. Integrate the shared checkpoint while preserving any subsequent fork edits before independently committing; do not commit duplicate baselines.
  • New-template chat owns MEOW first, then Femwell. Zap-refinement chat owns the current zapFDTD origin notebook/binder/tests and numerical qualification. Coordinate changes to shared scripts/notebook_inputs.py, wavelength handling and specs; integrate shared fixes once rather than editing them concurrently.
  • Eve owns GFP menu/launcher/viewer implementation; neither chat should implement that UI. Keep this original chat available for historical questions.

Read first

File Owns
Full spec Inputs, PDK/material contracts, audit, release gates and engine priorities.
Eve handoff UI entry points, source choice, bound paths, Config initialization and project defaults.
Development notes Meep evidence, SDK bootstrap/guards, retired BAD loader and LF workaround.
zapFDTD notes Native API, defaults, material/cladding behavior and verification limits.

Non-negotiables: true generic templates, not walkthroughs; Meep 3D only; one hidden Config followed by Settings, Prepare/preview, Run and Results; no tags or execution toggle. .venv kernelspec throughout. Config holds imports, PDK activation, bound paths and geometry. GSCH layout rebuilds GDS, rejects dropped connection warnings and uses existing SDK LVS before import. Schematic-only inputs require a compatible solver; never pass connectivity as physical geometry. Seed editable wavelength center/full span from project pyproject.toml at generation. Preserve dotted PDK module imports and PDK material/layer information. Keep test workarounds out of clean notebooks. No new Meep 2D/cross-section copies.

New-template chat: MEOW, then Femwell

Read current gsim/nbs and source APIs, not only older simulation-templates examples. Establish each engine's actual accepted geometry/cross-section inputs, material handling and output contract before declaring source compatibility. Build one readable generic baseline per engine; keep device recipes as future presets. Reuse the current notebook structure, binder/metadata conventions and project defaults where applicable, without imposing FDTD-only settings. Verify preparation and failure handling, then an appropriately authorized solve and readable results. Solver computation permissions from historical jobs are not blanket authorization for new jobs. Low priority: Palace, SAX, CirculaX, BeamZ. Exclude Tidy3D and Ansys from new templates.

Zap-refinement chat

Start with simulation_templates/zapfdtd_template.*, scripts/bind_zapfdtd_template.py, tests/test_zapfdtd_template.py and scripts/verify_zapfdtd_notebook_inputs.py. Python/GDS/valid GSCH preparation passes; two broken GSCH fixtures stop. Latest recorded shared suite: 65 passed. Run/Results have synthetic coverage; our two real jobs failed before results. Eve reports CRLF caused the crash: apply the recorded LF patch only to development execution, once per fresh kernel; qualify a successful run before claiming success.

Validate the current backend schema/version and show actual port responses and field-plane heatmaps. Convergence must be reported separately from job completion. Do not resume unrelated crash-fixing work already handled by the solver team. PDK cards currently become fixed real indices at the center wavelength in the writer; the submitted jobs contained no dispersion blocks. Uniform PDK-selected cladding touches the core; it does not synthesize substrate/BOX volumes. Keep single-source results honest: one column, implicit all-port measurements. Multi-source batching and source eigenmode exports remain future work.

Local execution and evidence

  • Python: C:/Users/eve/Documents/test_projects/photonics-full-init/.venv/Scripts/python.exe (3.12.14 at last check). Project PDK: gpdk in that project.
  • Default installed GFP was 1.8.20. Select the extension's 2.1.0 SDK before any imports in a fresh process/kernel, including sibling nyancad; use the harness bootstrap and verify gfp.__file__. Last extension: gdsfactory.gdsfp2-beta-2026.1012.0-win32-x64. Discover/recheck installed versions; do not assume this dated path stays current.
  • GFP_API_KEY is inherited from environment. Never print, persist or forward it. Disable solver/cloud guards in a fresh kernel before authorized real execution.
  • Local source repos: C:/Users/eve/Documents/GitHub/{gsim,FDTD,SAX,CirculaX}. Latest recorded gsim: 0.6.0 / 05c6c93; FDTD: a7b4175. Recheck before testing.
  • Fixture directory: C:/Users/eve/Documents/test_projects/photonics-full-init/gpdk: meep_two_bends_both.gsch, meep_two_bends_schematic.gsch, meep_two_bends_layout_busted.gsch. Leave these source files untouched.
  • Session outputs are outside the worktree, under C:/Users/eve/Documents/Codex/2026-10-07/https-github-com-doplaydo-simulation-templates/outputs. zapfdtd-failed-job-inputs-20261012 holds exact config/mesh ZIPs for fdtd-7fc98b3d (bend) and fdtd-0a7cd1ad (straight). zapfdtd-aeb787a2-metadata-20261012 holds another user's successful-run output: 777 s solver runtime, timestep cap reached, not converged; no image/environment/ submitting-gsim provenance. Do not confuse that with our failed-job inputs.

Starter prompts

New templates: "Read docs/simulation-template-continuation.md and its linked contracts. Continue true simulation templates, starting with MEOW, then Femwell. Use your own worktree from the checkpoint containing the latest zap/Meep work. Keep notebooks simple and Eve's app UI out of scope. Report progress in chat."

Zap refinement: "Read docs/simulation-template-continuation.md and the zap notes. Refine and qualify the existing zapFDTD template in your own worktree. Preserve our source routes and notebook design. Use development-only SDK/LF workarounds where still needed, and distinguish preparation, completion and convergence. Show me results in chat; obtain authorization for new cloud jobs."