[ChrisRackauckas-Claude]

Summary

Downgrade Sublibraries fails on lib/DiscreteMeasures because the IntervalArithmetic extension still called Base.intersect and compared with ==. Those APIs throw on every IntervalArithmetic 1.x release (including the floor 1.0.11 and latest 1.0.12), so this is an incomplete 0.x→1.x migration rather than a missing floor bump. This PR switches to intersect_interval / isequal_interval, which are the supported IA 1.x APIs and already work at the declared floor. A follow-up commit also makes the extension's imports explicit so ExplicitImports QA passes when the extension is loaded.

Verification

CI-faithful local floor run (julia +lts, julia-downgrade-compat v2 via dg_local.jl, mode alldeps, effective skip = Julia stdlibs ∪ in-repo sublibs, Pkg.test(; allow_reresolve=false)):

Before (HEAD without fix), IntervalArithmetic v1.0.11 — matches CI:

  Measure Restrictions   |   14      5     19   3.6s
    Interval Ext         |           5      5   3.5s
ERROR: LoadError: Some tests did not pass: 40 passed, 0 failed, 5 errored, 0 broken.

After (interval API fix), same floor Manifest:

Test Summary:     | Pass  Total  Time
Discrete Measures |   45     45  9.8s
     Testing DiscreteMeasures tests passed

Explicit-imports follow-up (julia +1, JULIA_REVISE_POLL=1):

QA-only (OPTIMALUNCERTAINTYQUANTIFICATION_TEST_GROUP=QA):

Test Summary:     | Pass  Total     Time
Quality Assurance |   18     18  1m03.7s
     Testing DiscreteMeasures tests passed

All / extension loaded (OPTIMALUNCERTAINTYQUANTIFICATION_TEST_GROUP=All; Core loads IntervalArithmeticExt then QA runs ExplicitImports):

Test Summary:     | Pass  Total   Time
Discrete Measures |   45     45  10.5s
Test Summary:     | Pass  Total     Time
Quality Assurance |   18     18  1m01.5s
     Testing DiscreteMeasures tests passed

Commands:

export RESOLVER_PATH=<scratch>/resolver
julia +lts dg_local.jl "<stdlib+sublib skip>" lib/DiscreteMeasures alldeps 1.10
julia +lts --project=lib/DiscreteMeasures -e 'using Pkg; Pkg.test(; allow_reresolve=false)'
JULIA_REVISE_POLL=1 OPTIMALUNCERTAINTYQUANTIFICATION_TEST_GROUP=QA julia +1 --project=lib/DiscreteMeasures -e 'using Pkg; Pkg.test()'
JULIA_REVISE_POLL=1 OPTIMALUNCERTAINTYQUANTIFICATION_TEST_GROUP=All julia +1 --project=lib/DiscreteMeasures -e 'using Pkg; Pkg.test()'

What I did NOT verify

• Full OUQBase / CanonicalMoments downgrade suites in this PR (CanonicalMoments was already green in the cited CI run; OUQBase ambiguity is a separate draft PR Fix OUQBase 𝔼/ℙ SymbolicUtils Operator ambiguity (separate from #54 floor) #59).
• Root-package Downgrade workflow.
• 32-bit / nightly.

Reviewer pushback

• Confirm isequal_interval is the preferred public comparison for decorated intervals (vs isequal / isthin checks).
• The extension still println("Load Ext") — left untouched as out of scope.
• No IntervalArithmetic compat bump: the floor already resolves to a version that works once the API calls are fixed.

Closes #54

Please ignore this draft until reviewed by @ChrisRackauckas.

Risk assessment

• Risk: low
• Blast radius: DiscreteMeasures IntervalArithmetic extension + its Interval Ext tests only; weakdep path for interval-valued clamp_domain/clamp_weight. No public API rename. Downstream OUQBase only sees correct interval clamping. Explicit-import change is extension-module only.
• Evidence: local floor fail-before (40 pass / 5 error, same intersect ArgumentError as CI run 35162761668) and pass-after (45/45) at IntervalArithmetic 1.0.11; QA 18/18 and All (Core+QA with extension loaded) 45+18 after explicit imports
• Independent review: Codex gpt-6-astra rated round 1 (head 9251200) MERGE, risk low, high confidence. Claude Code (claude-opus-5-5, a different harness from the Cursor author) rated round 2 (head 7bcb7d1, after the explicit-import cleanup) MERGE with no blocking findings. The two red checks, OUQBase downgrade and umbrella QA, are pre-existing and fixed by Fix OUQBase 𝔼/ℙ SymbolicUtils Operator ambiguity (separate from #54 floor) #59 and Fix QA: explicit imports, OUQBase docs, approved reexports #57.
• Merge: auto-merge candidate after CI green on explicit-imports follow-up head — independent review MERGE/low/high on the interval fix; follow-up is mechanical ExplicitImports hygiene verified locally

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

Risk assessment

• Risk: low
• Blast radius: Only the lib/DiscreteMeasures IntervalArithmetic weak-dependency extension, i.e. clamp_domain(::Interval, lb, ub). The other method, clamp_weight, is not changed. On IA 1.x, which is the only range the IntervalArithmetic = "1.0.11" compat allows, clamp_domain used to throw ArgumentError on every call. It now returns intersect_interval(x, interval(lb, ub)). That turns a method that always errored into a working one, so it is a patch-level fix and no working caller can break. Nothing outside the extension is touched: no public names, core modules, compat bounds or the root package.
• Evidence: Head 7bcb7d1 has 2 failing checks, and both fail the same way on the latest default-branch commit, 013f14a (runs 35162761668 and 35162761805):
  • Downgrade Sublibraries / test (lib/OUQBase): MethodError: promote_symtype(::OUQBase.𝔼_, ::Type{Real}) is ambiguous, "0 passed, 0 failed, 1 errored" on both head and main. Pre-existing.
  • tests / QA (julia 1) / Tests - QA: the implicit import of @random_variables from OUQBase, plus the undocumented-public-API and reexport failures, "18 passed, 2 failed, 1 errored" on both head and main. Pre-existing.
  • Downgrade Sublibraries / test (lib/DiscreteMeasures) was red on main (5× ArgumentError: `intersect` is purposely not supported, 40 passed / 5 errored) and is green on head, so the PR fixes it.
  • Every other check passes, including DiscreteMeasures Core on lts/1/pre, DiscreteMeasures QA, root Core on lts/1/pre/x86, Downgrade Core, docs, Runic and typos.
  • No regressions.
• Independent review: Devin CLI (Mac) / fusion-claude-opus-5-5-high-sidekick-swe-2-medium rated it low risk with high confidence. The 10-line diff moves from APIs that IA 1.x deliberately makes throw to its exported replacements (intersect_interval, isequal_interval, interval, Interval), so no non-public dependency internals are used. The test edit swaps ==, which throws InconclusiveBooleanOperation on non-thin intervals, for isequal_interval while keeping the same five cases and expected values. That is not a weakened test. It flips a CI job from red to green without making any job red. Minor points the description does not raise:
  • There is no version bump in lib/DiscreteMeasures/Project.toml, so a separate release step is needed to ship the fix.
  • The two earlier agent reviews that the description cites are not on GitHub (0 reviews, 0 comments), so I couldn't verify them.
  • The println("Load Ext") side effect is still there; the description does say this.
• Merge: auto-merge candidate

---

🤖 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)