open-problem

Status: resolved (2026-11-12). No defect in the code. Two notes were stale or missing, one backlog finding was already handled, and one finding was a hazard (two internal functions of one name with different behaviour), removed by a rename. Related: Feature Backlog (P17, P18), LP-028 Findings of the review of P1.

Sources: a review of the uncommitted moves P17 and P18 by Mistral (vibe --agent review), run with Julia in the helper worktree Lenticulum.jl-mistral (2026-11-12). Its answer is in ../Lenticulum.jl-Review-2026-11-12/P17-18-from-mistral-2.md (the first run stopped at its turn limit); its scripts are scratch/p17-18/check_p17.jl and scratch/p17-18/check_p18.jl in that worktree. code: lib/Mycelium.jl/src/gaussian.jl (reduce_mixture, reduce_message), lib/LenticulumCore.jl/src/gaussian_belief.jl (belief_mean, belief_cov of SampleBelief), lib/AdversarialFactors.jl/src/ratio.jl, lib/Lenticulum.jl/src/Lenticulum.jl (the re-export); tests: lib/Mycelium.jl/test/mixture_reduction.jl, lib/LenticulumCore.jl/test/sample_moments.jl.

Problem. Two moves out of type piracy were required to change no result: Runnalls’ mixture reduction from Lenticulum to Mycelium (P17) and the sample moments from AdversarialFactors to LenticulumCore (P18), with the old public names still working.

Mistral checked behaviour first. It compared the moved functions with the code before the move, copied verbatim from HEAD: P17 on 40 random mixtures with every budget and several prune values, and P18 on 40 random sample beliefs (matrix, vectors, scalars, weights scaled by 1e300, uniform). Both were identical. Lenticulum.reduce_mixture === Mycelium.reduce_mixture, Aqua passes in the four touched packages, and no example, benchmark or docs page used the old locations.

#findingright?what was donetest
F1The header comment of LenticulumCore’s gaussian_belief.jl still placed reduce_mixture in Lenticulumrightcorrected—
F2The backlog rows P17 and P18 were still “proposed”right for the diff Mistral had; the backlog was updated after the diff was takennothing more—
F3Mycelium’s gaussian.md had no section on the code moved into gaussian.jlright (the convention of a note beside each source file)section “Mixture reduction” added—
F4LenticulumCore’s new _sample_weights and Mycelium’s existing _sample_weights (messages.jl) differ for all-zero weights (an error against uniform weights)right as a hazard, not a defect: each module calls its own, and the SampleBelief constructor refuses all-zero weightsLenticulumCore’s renamed _moment_weights, with a comment—

Leaning / resolution. No code defect. F1, F3 and F4 were fixed in notes and a name.

Checked after the triage: julia --project=docs docs/make.jl passed (API docs and vault), with reduce_mixture on Mycelium’s API page. Not checked: nothing outside this repository calls the old locations (no registered dependants).