ui: live solver dashboard — a run() front-end on the stepped-session seam (fleet UI proposal)

[InauguralPhysicist]

Proposal (from Jon): EigenMiniSat could have UI. The pieces are already in place — minisat.eigs is a pure importable core, and the stepped-session API exposes the full stats struct mid-solve (conflicts, resolutions, learnts, peak_learnts, max_level, decisions, propagations, restarts — the same counters the ladder harness prints every 300s). That's exactly the fleet's core + run() split (eigen-edit pattern): add a front-end that opens a window and drives the solver a step-budget per frame.

Scope sketch:

1. Problem picker — the built-in families (pigeonhole, graph coloring, Tseitin tori by rows×cols) + DIMACS file load.
2. Live trajectory dashboard — conflicts/resolutions/learnts-over-time plots updating as the solver runs, peak_learnts and max_level readouts, restart markers. This is the counters-are-the-result-of-record discipline made visible.
3. Search view — current assignment as a variable grid (colored by value/unassigned), decision-level depth bar, activity heat. Solve/pause/step/reset transport.
4. Fleet UI standard from birth (Tidepool#43 / eigen-edit#3 / eigen-sheet#50): themed toolkit shell, render-decode + mouse oracles alongside the UI, desktop-gui-qa screenshots as acceptance.

Forcing-function contract (the reason to build it, not just the demo):

• lib/ui has no chart/plot widget — this forces that gap upstream from a second consumer simultaneously with eigen-sheet#26 (dynamic charts). Two consumers forcing one gap is the strongest upstream signal we get.
• A step-budget-per-frame compute loop inside a UI event loop is a new pattern for the toolkit (long-running work + responsive chrome) — likely to surface scheduling/repaint gaps no current app hits.
• Observer angle: the solver's bindings under the observer while it runs is the on-brand demo — measurement of a live search, for free.

Gate before build: eigen-portfolio-strategist validates the contract and adds the coverage-map row (hq/PORTFOLIO.md), per the fleet-membership rule. Build owner once approved: eigenscript-app-fleet-engineer.

Non-goal: solver performance work — the UI must not perturb the counters; a headless run and a UI-stepped run of the same case must produce byte-identical counter trajectories (that assert IS the oracle).

[InauguralPhysicist]

Gate cleared: CEO go 2026-08-02, ladder order as filed — hq/proposals/fleet-ui-expansion.md is APPROVED and the coverage map (hq/PORTFOLIO.md, lib/ui lane) now carries the program. Rungs proceed one at a time; this rung starts when the previous one is green (suite + oracles + desktop-gui-qa screenshots).

[InauguralPhysicist]

The x-y plot widget has landed upstream — EigenScript#819 (PR #824), merged 2026-08-02, all 18 CI checks green including asan+ubsan at a zero leak tally.

chart was generalized in place into a data-coordinate x-y plot (no parallel plot type; a series with no x keeps index-x, so nothing regressed). It was designed against all four waiting consumers at once — this issue among them — so the API should already fit: named multi-series each with its own x list, fixed_aspect, labelled vline/hline/point markers, add_point/add_xy/chart_trim for live incremental data, and drag-pan/wheel-zoom owned by the widget (it reads the toolkit's own pointer state, so no consumer has to shadow mousemove — see #822).

Clipped and theme-keyed by construction: the suite asserts zero primitives outside the widget rect under extreme pan and zoom, planted-fault validated, and confirmed in real pixels (the ring around each chart decodes to exactly the panel background). API reference is in docs/STDLIB.md.

This consumer can drop its hand-rolled canvas plotting whenever the next slice touches that code. It arrives on the next EIGS_REF bump.