Skip to content

feat(reports): the per-loop PI report — template + fill contract (#1700) - #1701

Draft
aarontrowbridge wants to merge 10 commits into
mainfrom
pi-report
Draft

aarontrowbridge wants to merge 10 commits into
mainfrom
pi-report

Conversation

@aarontrowbridge

Copy link
Copy Markdown
Member

Closes #1700's first slice. Supersedes amicissimo#478 (closed with a cross-pointer): the agent fills a fixed LaTeX template at every loop boundary instead of a mechanical projection, with the replacement guardrails — verbatim verdicts/numbers, provenance stamps, tectonic build under a 1–2 page cap with a demotion ladder. This PR lands the template + fill contract; the build-gate script + synthetic fixtures + tests follow in this branch before it leaves draft.

Supersedes the parked amicissimo #478 implementation: the agent fills a fixed
LaTeX template at every loop boundary (replacing the mechanical-projection
doctrine, with verbatim-numbers + provenance-stamp + gate guardrails), built
with tectonic under a 1-2 page cap. Example fills for real campaigns render
locally only (repo is public); committed fixtures will be synthetic.
@coderabbitai

coderabbitai Bot commented Oct 3, 2026

Copy link
Copy Markdown

Important

Draft PR not reviewed

Draft PRs are not automatically reviewed by default.

  • Trigger a manual review

To automatically review draft PRs, update your CodeRabbit configuration:

reviews:
  auto_review:
    drafts: true
  • Autopilot · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

Comment @coderabbitai help to get the list of available commands.

…ons, traceable citations, reading slot, no page cap (#1700)

Round-2 review verdict: tables cannot explain, the hard cap forced the
wrong trade, and hypotheses/citations were missing. The template moves
to prose findings (each explains the hypothesis in plain language before
stating numbers), a digest that names each idea, a per-loop Reading slot
(never silent when empty), and numbered traceable references. Length is
a receipt signal, not a gate.
…ponent references, corrections target the source (#1700)

Spec-20261003-123000-formulation-record (manual review, honestly
labeled): every solve emits formulation.toml (retained ProblemSpec via
extract_spec, or the upstream best-effort path + call-site solver
actuals); the classic-form block renders from the record with zero agent
invention in the math; prose component references are gate-checked
against stated_components; a finding with no record is UNBACKED; raw
matrices cap to digest + dims. Six pytest gates, synthetic fixtures.
…rs, raw stamp string, actuals-sourced integrator fallback (#1700)

End-to-end validated on the real OQC DDQ X_L construction: extract_spec on the
live hand-built problem, record rendered into the memo. Two honest gaps
measured and filed upstream (Piccolo#366): R_bend/Dt_bounds/du_bound are
not retained by the template; the integrator block is absent from the
best-effort record (the alg is call-site-sourced, labeled).
…ean display equation, then a 'Here, ...' prose paragraph (#1700)

Per Aaron's read of the Hong-ye appendix: equations carry symbols only —
the integrator, alg, weights, bounds, solver actuals all live in prose,
the way 2508.19075 writes Appendix G. The generated block validated
end-to-end on the live OQC DDQ X_L record. Tests updated to the new
register; 6 passing.
… from the problem template tag, harden solver-sentence escaping (#1700)
…own-block tolerance, emission-capped passthrough, template_actuals integrator (#1700)

The standing inventory alarm renders on every record (a future problem
shape dropping different fields gets caught at render, not by re-running
a checker); unknown top-level blocks are named and preserved, never
fatal; matrices capped at emission pass through as-is instead of being
re-digested into phantom inline matrices; and call-site template actuals
beat template-name inference — a SplinePulseProblem on PWC dynamics
(#540) never renders as spline-faithful. The previously dead
integrator_source string now renders.
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.

feat(reports): the per-loop PI report — a 1–2 page LaTeX state-of-campaign filled from the ledger at every loop boundary

1 participant