[ChrisRackauckas-Claude]

Summary

OUQBase Core fails on current SymbolicUtils (≥4.38, including latest 4.48) with MethodError: promote_symtype(::𝔼_, ::Type{Real}) is ambiguous against promote_symtype(::Operator, ::Type{T}). This is the same failure seen in Downgrade Sublibraries (CI run 35162761668) and Sublibrary CI Core. This PR resolves the ambiguity, switches Term construction to SymbolicUtils.term, and fixes follow-on group-lookup / raw-moment parsing that only become reachable once Operators construct successfully.

Independent review: Codex gpt-6-astra rounds 1-4 CHANGES (each fixed); Claude Code claude-opus-5-5 (a different harness from the Cursor author) round 5 on head 630a206: VERDICT MERGE, no blocking findings. Codex is out of quota until 2026-10-03, so round 5 used Opus.

Round-2 follow-up on this branch: get_raw_moment_order now checks isinteger before Int(...) so non-integer exponents raise the descriptive raw-moment error instead of InexactError.

Verification

Non-integer raw-moment exponent (round-2 finding)

Before:

BEFORE_ERR_TYPE=InexactError
BEFORE_ERR=InexactError: Int64(2.5)
Test Summary:                               | Fail  Total  Time
non-integer exponent before isinteger guard |    2      2  4.0s

After:

Test Summary:                                        | Pass  Total   Time
OUQBase operator regressions (after isinteger guard) |   16     16  45.8s

Probability ≳/≲ regression (round-1 finding)

Before (pre-Inequality unwrap): 10 pass, 2 fail, 2 error (Any[] / AssertionError).
After: 14/14 (then 16/16 with the isinteger assertion).

OUQBase Core (TEST_GROUP=Core, julia +lts)

promote_symtype operators | 14/14 (now 16/16 with extra assertions)
Flood Problem (Q only) | completes
Flood Problem (canonical moments objective) | 1 error: isapprox(::BasicSymbolic, ::Float64)

OUQBase QA (TEST_GROUP=QA)

Quality Assurance 16/16 on julia +lts (prior head); reviewer also saw 18/18 on +release.

Out of scope: canonical-moments objective_f / symbolic isless

Not fixed in this PR (and no upstream issue filed here). Standalone Symbolics-only reproducer + version bisect live in the job REPORT.md:

• Current Symbolics 7.41.1 / SymbolicUtils 4.48.0: evaluate(H, d) is numeric; evaluate(substitute(H ≳ h, …), d) leaves isless(2.0, (1000.0 / (9000.0sqrt(0.0009)))^0.6) unfolded.
• Symbolics 6.58.0 folds the same ≳ comparison to true::Bool; breakage begins at Symbolics 7.0.0.

What I did NOT verify

• Full green OUQBase Core (blocked on upstream evaluate/isless for ≳).
• Root-package Downgrade workflow end-to-end.
• Whether dropping the getdefault via-owners ignore (reviewer finding 3) is safe on every CI Julia lane.

Reviewer pushback

• Preferred Term constructor vs Differential-style.
• underlying_random_variables placement.
• Follow-up for Symbolics ≳ evaluate / isless fold (documented in REPORT; not filed upstream from this job).
• Sibling PR Fix QA: explicit imports, OUQBase docs, approved reexports #57 will conflict on qa.jl ignore cleanup.

Part of #54

Please ignore this draft until reviewed by @ChrisRackauckas.

Risk assessment

• Risk: medium
• Blast radius: OUQBase Operator construction (𝔼/ℙ), RV group resolution, raw-moment parsing (including non-integer exponent errors).
• Evidence: fail-before/pass-after for promote_symtype, ≳ discovery, and non-integer moment order; QA green locally; Core still blocked only on disclosed Symbolics isless.
• Independent review: Codex gpt-6-astra round 1 CHANGES (fixed); Claude Code Opus 5.5 round 2 on 5242c31 MERGE.
• Merge: approved by round-2 independent review; wait for @ChrisRackauckas.

Harness: Cursor Agent CLI 2026.09.26-dd393fe; model: auto; transcript: /home/crackauc/sandbox/agent-jobs/OptimalUncertaintyQuantification.jl/jobs/54-cursor/log.txt on amdci2.julia.csail.mit.edu

Update (Codex gpt-6-astra round-4 CHANGES)

Removed the QA all_qualified_accesses_via_owners ignore for :getdefault and the comment that preserved the non-owner access. getdefault (and qualified getbounds) now go through their owner ModelingToolkitBase, declared as a direct dependency (compat = "1.50", matching ModelingToolkit 11.31).

Ownership fail-before / pass-after

Before (e09a134, no ignore):

OWNERSHIP_CHECK=FAIL
QualifiedAccessesFromNonOwnerException
getdefault has owner ModelingToolkitBase but it was accessed from ModelingToolkit
.../canonical_moments.jl:198:48

After (630a206, no ignore):

getdefault access sample: ModelingToolkitBase
OWNERSHIP_CHECK=PASS

Re-verification (OPTIMALUNCERTAINTYQUANTIFICATION_TEST_GROUP)

julia +lts (1.10.12):

promote_symtype operators |   25     25
Flood Problem (Q only) | completes
Flood Problem (canonical moments objective) | 1 error: isapprox(::BasicSymbolic, ::Float64)  [disclosed; fixed in #60]
Quality Assurance |   16     16
Testing OUQBase tests passed  (QA)

julia +1 (1.13.0):

promote_symtype operators |   25     25
Flood Problem (canonical moments objective) | 1 error [same disclosed]
Quality Assurance |   18     18
Testing OUQBase tests passed  (QA)

Harness: Cursor Agent CLI 2026.09.26-dd393fe; model: auto; transcript: /home/crackauc/sandbox/agent-jobs/OUQ-b/jobs/54-cursor/log.txt on amdci2.julia.csail.mit.edu

Risk assessment

• Risk: medium
• Blast radius: Only the OUQBase sublibrary. Affected: 𝔼/ℙ/𝟙 operator term construction (now built with SymbolicUtils.term, plus a new promote_symtype(::𝔼_/ℙ_, ::Type{T}) method), random-variable group lookup (get_ordered_group_names now goes through the new internal underlying_random_variables), raw-moment parsing (get_raw_moment_order rewritten; the RHS is now coerced with Float64(...)), and a new direct dependency on ModelingToolkitBase (compat 1.50, SciML, MIT). CanonicalMoments, DiscreteMeasures and the umbrella source are untouched. Draft PR OUQBase: evaluate Inequality probability events as Bool comparisons #60 edits the same 9 files, and PR Fix QA: explicit imports, OUQBase docs, approved reexports #57 edits qa.jl, interface.jl and operators.jl, so conflicts are certain.
• Evidence: 5 checks fail on head 630a206. None is a regression:
  • tests / QA (julia 1): the same result on main 013f14a (run 35162761805): 18 pass, 2 fail, 1 error (no_implicit_imports, public-API docstrings, unapproved reexports). Pre-existing.
  • Downgrade Sublibraries / test (lib/DiscreteMeasures): the same 5 intersect ArgumentErrors on main (run 35162761668). Pre-existing, and this PR doesn't touch it.
  • Downgrade Sublibraries / test (lib/OUQBase): main fails earlier, at Flood Problem (Q only) with the promote_symtype(::𝔼_, ::Type{Real}) is ambiguous error. The PR gets past that and fails later at Flood Problem (canonical moments objective) with isapprox(::BasicSymbolic, ::Float64). Pre-existing red that the PR partly fixes.
  • sublibraries / lib/OUQBase / Julia 1 and / Julia lts / Tests - Core: these jobs are skipped on main pushes (path detection), so there is no main baseline. I compared against PR Fix QA: explicit imports, OUQBase docs, approved reexports #57 instead (run 36274860212, no fix), where both fail at Q only with the same ambiguity. With this PR they fail only at the isapprox error, which the description discloses as out of scope and handled in OUQBase: evaluate Inequality probability events as Bool comparisons #60. Pre-existing, improved.
  • Newly passing: lib/OUQBase [QA] / Julia 1 (PR Fix QA: explicit imports, OUQBase docs, approved reexports #57's run had 1 failure).
  • Everything else passes: Docs, Runic, typos, root Core lanes, Downgrade Core, benchmark.
• Findings from the diff:
  • No tests were weakened. A new Core testset was added. The QA ignore list got stricter (:getdefault removed; access now goes through its owner ModelingToolkitBase, and the sublib QA check passes).
  • The QA ignore list is now stale, and the description doesn't mention it. :Term and :symtype stay in all_explicit_imports_are_public even though operators.jl no longer imports them. The comment still says the compat caps are "SymbolicUtils 3.x / ModelingToolkit 9.x", which doesn't match the current compat.
  • Operator is now imported from SymbolicUtils instead of Symbolics.
  • get_raw_moment_order now throws ErrorException for non-integer or non-power forms, where it used to hit a different path. The OUQBase version stays at 0.1.0 and the changed functions aren't exported, so this is a minor SemVer concern.
• Description: The body contains two risk sections that contradict each other (an older "round-2 on 5242c31" block plus an update). It says Core is not fully green. It has no review comments on GitHub (0 review, 0 issue comments). It asks to be ignored until @ChrisRackauckas reviews it.
• Independent review: Devin CLI (Mac) / fusion-claude-opus-5-5-high-sidekick-swe-2-medium rated it medium, high confidence: CI shows no regressions and it fixes the documented ambiguity, but OUQBase Core is still red, it adds a direct dependency, rewrites parsing logic, leaves QA ignores stale, and overlaps open PRs Fix QA: explicit imports, OUQBase docs, approved reexports #57 and OUQBase: evaluate Inequality probability events as Bool comparisons #60.
• Merge: needs human review. Reasons:
  • It is a draft that explicitly waits for Chris's review.
  • The OUQBase Core and Downgrade lanes are still red and depend on OUQBase: evaluate Inequality probability events as Bool comparisons #60.
  • It adds ModelingToolkitBase as a new direct dependency.
  • The :Term/:symtype QA ignores and their comment are stale.
  • It will conflict with Fix QA: explicit imports, OUQBase docs, approved reexports #57 and OUQBase: evaluate Inequality probability events as Bool comparisons #60, so the merge order needs deciding.

---

🤖 Risk assessment posted by an AI agent (fleet master) — harness: Devin CLI (Mac) / fusion-claude-opus-5-5-high-sidekick-swe-2-medium; dispatched by Claude Code head, model claude-opus-5-5[1m]
Conversation: local Claude Code session 3cd6500a-1f81-46b5-ac0b-c466e15b6a53 on Chris's Mac (session ID, no URL)