Strengthen portfolio accuracy and reproducibility

[SpaceDevEngineer]

What changed

• remove the hidden label-roll prior from the end-to-end solver by separating CD scale/shear calibration from physical body roll and seeding all attitude axes from RANSAC/Wahba
• use Euclidean per-star pixel residuals in the quality gate and make result metadata explicit
• speed up public Triangle ID with a catalog-keyed pattern cache, vectorized third-star filtering, and FOV-cone verification
• make Streamlit Replay robust for both successful and rejected solves
• add auditable full-test results for all 120 frames, corrected visualizations, and honest documentation of exclusions, refusals, limitations, and the roll-convention correction
• add geometry/result/Streamlit regression tests plus GitHub Actions CI
• remove machine-specific data paths and improve reproducible setup

Why

The previous portfolio version described true lost-in-space solving while the plate-solve roll seed still came from label WCS. It also showed only a 16-frame demo despite a larger evaluation being available, and one rejected replay case crashed. This change aligns the implementation with the claim and makes the reported evidence reproducible.

Results

• full evaluation: 104/107 valid frames solved (97.2%), 0 false locks
• corrected median attitude error: 8.98 arcsec; p90: 19.41 arcsec
• 95/104 solved frames below one pixel; all 104 below one arcminute
• bundled replay subset: 15/16 solved, 6.63 arcsec median, one honest refusal
• real one-frame blind regression: 55 matches, 4.88 arcsec attitude error

The previously quoted 7.35 arcsec full-test median used the FITS CD angle as physical roll. The corrected body-roll convention is now documented and applied consistently.

Validation

• python -m compileall -q Code tests
• pytest -q — 7 passed
• git diff --check
• Streamlit AppTest covers both successful and rejected Replay paths

Scope

Pyramid matching is intentionally not included in this PR; it will be added only after its thesis-side evaluation is complete.

[Copilot]

Pull request overview

This PR strengthens the project’s “lost-in-space” attitude-solve claims and makes evaluation evidence reproducible by (1) separating camera calibration from physical body roll, (2) standardizing result metadata and residual metrics, and (3) bundling auditable full-test outputs plus CI and regression tests.

Changes:

• Refactors the core solver to remove label-derived roll leakage (CD angle) by factoring CD rotation out of intrinsics and seeding all attitude axes from RANSAC/Wahba.
• Standardizes quality gating on per-star Euclidean pixel residuals and adds explicit result metadata + normalization utilities for legacy artifacts.
• Improves demo/evidence reproducibility via bundled full-test CSV/JSON outputs, Streamlit Replay robustness, new tests, and GitHub Actions CI.

Reviewed changes

Copilot reviewed 31 out of 41 changed files in this pull request and generated 4 comments.

Show a summary per file

File	Description
tests/test_streamlit_replay.py	Adds Streamlit AppTest coverage for both successful Replay and quality-gate refusal paths.
tests/test_results_contract.py	Adds contract tests asserting bundled demo-set completeness and auditable full-eval counts/summary files.
tests/test_core_geometry.py	Adds geometry regression tests for pose/rotation round-trip, CD factorization, and radial residual computation.
Results/star_id_run/hlsp_tica_tess_ffi_s1751-o1-01222079-cam3-ccd2_tess_v01_img.json	Updates stored refusal artifact to new roll convention + Euclidean residual metadata.
Results/star_id_run/hlsp_tica_tess_ffi_s1751-o1-01222079-cam2-ccd1_tess_v01_img.json	Updates stored solved artifact metadata (residual metric, convention, calibration/seed).
Results/star_id_run/hlsp_tica_tess_ffi_s1751-o1-01222079-cam1-ccd4_tess_v01_img.json	Updates stored solved artifact metadata (residual metric, convention, calibration/seed).
Results/full_test_per_frame.csv	Adds auditable per-frame outcome table for the full 120-frame evaluation.
Results/full_test_metrics.json	Adds machine-readable aggregate evaluation metrics, counts, and provenance.
RESULTS.md	Updates documentation to the corrected roll convention and publishes full-eval metrics + exclusions/refusals.
requirements.txt	Adds pytest pin while keeping runtime deps (including Streamlit) in requirements.
README.md	Updates headline results to full-evaluation numbers and clarifies conventions/limitations and reproducible artifacts.
PROJECT_DESCRIPTION.md	Updates portfolio summary to corrected median error and clarifies statistical claims and roll calibration changes.
Code/Tess_Dataset/process_tess.py	Removes machine-specific default paths; requires explicit FITS dir and uses repo-relative defaults for outputs/catalog.
Code/Streamlit_app/pipeline_app.py	Makes Replay refusal path non-crashing and aligns displayed residuals/pose seeding with updated solver conventions.
Code/Star_ID/visualize_pipeline.py	Updates WCS reconstruction to use attitude-free CD intrinsics.
Code/Star_ID/visualize_inference.py	Updates overlays/residual plots to use Euclidean residuals and adds a summary-only mode.
Code/Star_ID/triangle_id.py	Improves Triangle ID performance via catalog-hash keyed cache, vectorized third-star filtering, and FOV cone culling.
Code/Star_ID/summarize_results.py	Adds a script to export an auditable per-frame CSV from JSON artifacts (with legacy-roll correction option).
Code/Star_ID/normalize_result_conventions.py	Adds a utility to normalize legacy artifacts to the physical roll convention + Euclidean residual metric.
Code/Star_ID/inference_full.py	Core refactor: factor CD rotation out of intrinsics, seed full attitude from RANSAC/Wahba, and gate on radial per-star residuals.
Code/Model_train_code/train.py	Removes machine-specific default dataset path by requiring --data-dir.
.gitignore	Ignores .pytest_cache/.
.github/workflows/ci.yml	Adds CI workflow running compileall + pytest on Python 3.11.
.devcontainer/devcontainer.json	Simplifies devcontainer setup to install requirements on create and run Streamlit on attach.

Comments suppressed due to low confidence (11)

Code/Streamlit_app/pipeline_app.py:257

• st.pyplot does not accept width="stretch" (Streamlit supports use_container_width=True). This is likely to raise at runtime under the allowed dependency range.

            st.pyplot(fig, width="stretch")

Code/Streamlit_app/pipeline_app.py:296

• st.pyplot does not accept width="stretch" (Streamlit supports use_container_width=True). Passing a string width here can break Replay/Live rendering.

    st.pyplot(fig, width="stretch")

Code/Streamlit_app/pipeline_app.py:325

• st.pyplot does not accept width="stretch" in Streamlit 1.30; use use_container_width=True to get the same layout behavior across supported versions.

        st.pyplot(fig, width="stretch")

Code/Streamlit_app/pipeline_app.py:369

• st.pyplot does not accept width="stretch" (Streamlit uses use_container_width=True). Leaving this as-is can cause a runtime exception when the app reaches Stage 2.

    st.pyplot(fig, width="stretch")

Code/Streamlit_app/pipeline_app.py:493

• st.pyplot does not accept width="stretch" (Streamlit uses use_container_width=True). This can crash Stage 3 rendering under the current requirements range.

    st.pyplot(fig, width="stretch")

Code/Streamlit_app/pipeline_app.py:697

• st.pyplot does not accept width="stretch" (Streamlit uses use_container_width=True). This is on the final output path, so a failure here would negate an otherwise successful solve.

st.pyplot(fig, width="stretch")

Code/Streamlit_app/pipeline_app.py:457

• st.dataframe does not accept width="stretch" (for responsive sizing use use_container_width=True).

        st.dataframe(angle_table, width="stretch", hide_index=True)

Code/Streamlit_app/pipeline_app.py:476

• st.dataframe does not accept width="stretch" (for responsive sizing use use_container_width=True).

        st.dataframe(angle_table, width="stretch", hide_index=True)

Code/Streamlit_app/pipeline_app.py:556

• st.dataframe does not accept width="stretch"; use use_container_width=True to avoid runtime errors across the supported Streamlit range.

        st.dataframe(pd.DataFrame(history), width="stretch", hide_index=True)

Code/Streamlit_app/pipeline_app.py:616

• st.dataframe does not accept width="stretch"; use use_container_width=True for the intended responsive layout.

        st.dataframe(pd.DataFrame(history), width="stretch", hide_index=True)

Code/Streamlit_app/pipeline_app.py:666

• This call passes width="stretch" to st.dataframe, but Streamlit expects a numeric width here. Use use_container_width=True for consistent responsive behavior.

}), width="stretch", hide_index=True)

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.