Fix uncertainty forecast trajectory indexing

[ChrisRackauckas-Claude]

Change

Read each trajectory from esol.u[i] in get_uncertainty_forecast. With current ensemble array indexing, esol[i] is a scalar, so the subsequent symbolic selection throws instead of returning forecast matrices. The new official regression covers scalar, one-element-vector, and multiple-variable selections, including output shape and initial values.

Rebased onto main — no longer stacked on #336 (the shape-reexport change is independent).

Verification

Julia 1.11.9 standalone public-API probe, identical 433-entry dependency graph except the EasyModelAnalysis source path:

Before (018c4e5a315138f4d8c0a01b64601f76c590d714):
Forecast trajectory indexing | 3 errors | 22.3s
x(t), [x(t)], and [x(t), y(t)] all fail at scalar ensemble indexing.

After:
Forecast trajectory indexing | 9 pass | 15.1s
exit 0

The probe constructs D(x) ~ -k*x, D(y) ~ k*x, with initial [1.0, 0.0], requests two trajectories at [0.0, 0.5, 1.0], and checks the returned sample count, variable-by-time matrix shape, and exact initial column. It does not loosen tolerances or remove sampling.

🤖 Generated with Devin CLI (model: SWE-2 Max) — local session, no shareable URL

[ChrisRackauckas-Claude]

The completed 32-bit Core job fails during dependency precompilation with the same two roots already audited on the shape prerequisite, before EMA assertions:

• Symbolics 7.39.0 with SymbolicUtils 4.46.3: Differential hashing at src/diff.jl:141 passes a UInt64 seed to the 32-bit symbolic hash method.
• QuasiMonteCarlo 0.3.11 with LatticeRules 0.0.1: ArgumentError: maximum number of points n must be less than or equal to 2^32.

I read this job's log rather than inferring the cause from its lane name. The linked prerequisite audit records the independently observed master failures, existing upstream fixes, and release/resolution constraints. No CI lane, precompile workload, or assertion is disabled here. The additional exact-commit local Core run using registered dependencies is still in progress; the body separately identifies the completed composed Core and registered QA results.

Links

• This failed job: https://github.com/SciML/EasyModelAnalysis.jl/actions/runs/34217743081/job/102033534738
• Existing dependency audit and fixes: Drop dangling shape reexport #336 (comment)

Audit by Codex CLI 0.153.4 (model: gpt-6-astra); local transcript /home/crackauc/.codex/sessions/2026/09/05/rollout-2026-09-05T05-08-31-01a070d3-9e50-71c2-98f5-129385e50fb7.jsonl.

[ChrisRackauckas-Claude]

The CI docs job ended cancelled, with the explicit check annotation: The job has exceeded the maximum execution time of 6h0m0s.

Before cancellation, its log recorded 48 failed example blocks, including cascades: Carcione 4, Petri 2, scenario2 4, scenario3 14, scenario4 7, and ensemble tutorial 17. There is no final makedocs summary, so this is not a completed or successful docs build.

This is separate from the already-recorded local strict docs run, which completed with exit 1 and [:example_block]. The local focused regression (9/9), registered Core (40/40), and registered QA (23/23) passed; they do not establish full docs success. No retry, source change, or test/build suppression was made in response to this cancellation. The known x86 dependency failure also remains.

Links

• Cancelled docs job and annotations: https://github.com/SciML/EasyModelAnalysis.jl/actions/runs/34217743023/job/102033405182
• Existing x86 failure evidence: Fix uncertainty forecast trajectory indexing #337 (comment)

🤖 Generated with Codex CLI 0.153.4 (model: gpt-6-astra); local transcript: /home/crackauc/.codex/sessions/2026/09/05/rollout-2026-09-05T05-08-31-01a070d3-9e50-71c2-98f5-129385e50fb7.jsonl.