Brief 19 · Prompt edition 2.0

Last-Light Relay

Build a complete planning-and-execution instrument for restoring a fictional island relay station before nightfall, in one self-contained HTML file: compute an optimal resource-feasible schedule, play through its consequences, then replan around a crane closure without rewriting work already started.

Read the complete prompt
Last-Light Relay benchmark illustration
Test concept · editorial illustration

1 recorded evaluation of this frozen brief · artifacts published unedited

Recorded evaluations

Each artifact was produced against edition 2.0 of this brief and is published exactly as delivered. The model run page links the run record, which states which source-level checks were performed, which checks remain for a human browser review, and how this delivery differs from a hosted-API single-message episode.

What this tests

long-horizon planning compressed into one shot; resource-constrained scheduling, optimality, execution and replanning with immutable history.

Proposed difficulty

harder than Starfall/Glider on proving a complete route to completion, and harder than Watch on keeping several time-dependent views consistent. It is not a claim of harder graphics or larger compute.

These are design hypotheses, not observations of model failure. Empirical discrimination has not been measured.

One message, one HTML response

One complete prompt as one user message; one self-contained HTML response. No tools, external dependencies or follow-up repair turns during authoring. Private reasoning is unrestricted. A human reviewer later checks the artifact in a browser; model self-reported success is not evidence.

Do not append BENCHMARK-PROMPT.md to this brief.

Rubric and pass rules

Each rubric has five binary criteria worth 20 design points each. Award a criterion only when every named check succeeds; otherwise award zero. These are rubric weights, not model scores. Passing requires all five criteria, every embedded checklist item and the shared delivery gate: exactly one self-contained HTML artifact, no external requests or dependencies, no tools during generation and no placeholder mechanisms. Review at 1280×800 and 390×844 with keyboard navigation and reduced motion. Use ledgers, exports, independent arithmetic and source inspection.

Feasible schedules · 20 design points
Checklist 2, 4 and 7; check every tick, predecessor and half-open boundary, plus editable durations.
Exact optimality · 20 design points
Default finish 10 and recovery finish 11 with lower-bound arguments; source search covers legal domain and handles horizon infeasibility without overclaiming.
Frozen-history recovery · 20 design points
Checklist 5–6; A,B,C unchanged, pending tasks rescheduled, execution remains at 4, endpoints behave correctly.
Shared state and lifecycle · 20 design points
Chart, resource strips, execution, export, manual validation and stale-search handling agree; all reset semantics hold.
Artifact and usability · 20 design points
Shared delivery gate and checklist 1; station states legible, responsive search, keyboard and accessibility preferences.

Pass: all criteria/checklist items.

Review beyond the examples

vary durations and closure endpoints inside the public domain; manually introduce invalid overlaps; inspect solver completeness, not just its chosen witness. Fixed witnesses demonstrate satisfiability and derived bounds, not model results. Two equally optimal schedules are both valid; no arbitrary tie-break penalizes a correct planner. Keep domains small enough for exact search in a native browser; do not expand to general industrial scheduling.

Versions and prompt revisions

Superbench 1.x results retain their original protocols. The core/specialist repository-run workflow does not schedule v2. City Scroll’s media-assisted specialist protocol remains separate. Results across versions are not quantitatively comparable, and no historical scores are migrated.

Edition 2.0 is the launch contract. Changes to behavior, fixtures or pass criteria require a new documented edition; future evaluations must retain the exact evaluated prompt revision. Typographic corrections must be identified.

Open the full prompt and validation checklist