Skip to content

feat(trace_log_tracker): dose/actuator provenance on step stats - #49

Draft
quint-mbbdigital wants to merge 2 commits into
CalculatedContent:mainfrom
quint-mbbdigital:pr/rg-c-trace-log-provenance-fields
Draft

feat(trace_log_tracker): dose/actuator provenance on step stats#49
quint-mbbdigital wants to merge 2 commits into
CalculatedContent:mainfrom
quint-mbbdigital:pr/rg-c-trace-log-provenance-fields

Conversation

@quint-mbbdigital

@quint-mbbdigital quint-mbbdigital commented Aug 10, 2026

Copy link
Copy Markdown

Type

  • A no-def
  • B docs-only
  • C embeds a definition choice (logging only)

Summary

Extend local_delta #34 provenance grammar to trace_log_tracker step stats:

  • actuator_id = trace_log_tracker
  • ecs_backend = midpoint_pl_detx (lineage label for joint tables)
  • dose_definition = correction_frobenius_over_base_step_delta_frobenius (matches existing correction_ratio)
  • dose_value = that ratio when a correction applied; null on skipped / geometry_failed
  • is_first_apply = first successful correction per parameter (not first schedule-due step)
  • is_first_due = first schedule-due clock step (may have null dose)

Also add docs/PROVENANCE_INVENTORY.md listing which packages emit these fields.

Why

Inventory: only wwpgd_local_delta had #34-style provenance. Trace-log is the midpoint sibling; joint tables need the same fields under F1–F5.

Definition surface (C)

  • Choice: logging fields only; correction math unchanged
  • Default: no change to apply cadence or correction mode
  • Open to change: yes (field names / dose_definition string)
  • Feedback filters: F1 scheduled≠applied (is_first_due vs is_first_apply); F4 null dose on no-op; F5 explicit definition string

Prior / SoT

Test plan

  • PYTHONPATH=. python -m unittest discover -s tests -v in optimizers/trace_log_tracker
  • no scientific efficacy claim

Out of scope

Other optimizers' provenance (see inventory); changing defaults; efficacy claims.

Logging-only fields (actuator_id, ecs_backend, dose_definition, dose_value,
is_first_apply) mirror local_delta CalculatedContent#34 grammar: null dose when no correction,
schedule-aware first apply. Correction mathematics unchanged. Inventory doc
lists remaining packages.
Split schedule clock (is_first_due) from first successful correction per
parameter (is_first_apply) so null-dose due steps are not first-apply (F1).
Persist applied_parameters in state_dict for resume-safe logs.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant