Quartz 5

ONGOING

3 min read

ONGOING TO DO LIST

Nominal GNC

Explore / formally discard Tikhonov regularization

  • Disposition now DOCUMENTED in tikhonov_regularization (topic page, 2026-06-23): the low-onset 0.020 variant was REJECTED (it “improves” p_c only by destroying the singular margin — targeting-ON collapse); incumbent onset 0.05 retained. Keep this item open only to explore a NEW variant (e.g. directional/filtered damping, per terminology entry).

Later - GNC

Null-space / posture controller integration later

Pinocchio 4.0

Wiki & code hygiene

Topic pages — Batch 2 backlog (cross-file concept pages) wiki

Dedup the 5 baseline validators’ shared harness (~389 lines) wiki

  • The five validate_*_baseline.py share ~389 lines of scaffolding (main/run/check/diff_against); detector is analysis/ast_duplication.py (NOT validation/ — fixed in d6d5cc0). WANTED (Toni Jun24), promoted from low-priority. Plan: extract ONE shared harness (e.g. validation/_baseline_harness.py), keep all 5 DISTINCT pins (base/ee_initial/ee_targeting/ee_coverage/ee_tracking are different regimes — do NOT cut to fewer). GATE CODE: after refactor, --pin nothing; run test_gold_standard.py + each validator must still report max_abs_diff=0.000e+00. Do this FRESH, not at end of a long session. NB validation/tests is NOT bloated — 136 dispatch + 64 data-primitive tests are legit coverage; this harness is the only real DRY target there.

Retire

Inconsistent scoring in ANCHOR mode

  • ANCHOR is inactive at the moment - may return later in risk - retire/segregate anchor mode stuff
  • target_finder score_target has no stable no-target score leaf (pre-existing MAJOR ISSUE) wiki
  • parameters: confirm anchor_length=0.50058 provenance (USER NOTE: nil, and not used); decide on falsified aim_floor_rate/aim_lead_max levers wiki

Auto-surfaced (Dataview — renders in Obsidian)

Pages awaiting review (status: draft)

TABLE updated AS "Updated", code_path AS "Source", code_paths AS "Spans"
WHERE (code_path OR code_paths) AND status = "draft" AND !contains(file.folder, "_")
SORT updated ASC

Most recently touched pages

TABLE updated AS "Updated", file.folder AS "Kind"
WHERE (code_path OR code_paths) AND !contains(file.folder, "_")
SORT updated DESC
LIMIT 15

Automations (not yet actually scheduled)

  • wiki-ingest
  • wiki-lint
  • validation-purge

DONE

Staging for finished items — move a completed board item here with a date. The weekly wiki-lint drains
this block to logs/DONE.md. Last drained 2026-06-25 (4 items → logs/DONE.md ## [2026-06-25] archived).

Purge worklist — dormant code marked defunct → PURGED 2026-06-25

  • Removed 2 modules (quintic_pose, detour_planner) + 5 legacy-only importers + 8 function-level members (curvature/quaternions/mesh envelope half). Pages + inbound links cleaned; gold gate + 408 non-legacy tests green. Detail in wiki/log.md.

Boresight + omega figures: retire the render_line_panels shim, make them run-spec-native wiki

  • analysis/boresight_roll_figure.py + analysis/omega_ripple_figure.py render via the render_line_panels shim with hand-set styling (stopgap; colours/titles/legends fixed 43e96af). Proper fix = drive both from a run-spec YAML so they inherit pipeline conventions. Needs: (a) the derived signals added as analysis-time catalog metrics via values_for_spec (computed from existing logs, NO re-run) — nu_e error-angle [deg], world-frame velocity error (SignalAnalysis.velocity_error, already in data_analyzer), frame angle [deg], z_e→deg; (b) per-run zoom-window support We have this! (the overlay zoom is shared-window only). NB the omega y-axis is dimensionally murky (a generalized-velocity projection onto a Γ singular direction) — the honest pipeline-native version just plots omega_b [rad/s]. Lesson captured in memory report-figures-via-own-pipeline (pay the pipeline cost up front).

Graph View