Escape the Dungeon · GAD
escape-the-dungeon/v10
Composite score
0.000
API interrupted · Anthropic server overload
This run was stopped by an HTTP 529 overloaded_error from the Anthropic API — a server-side transient load issue, not a rate limit or agent failure. The data captured here reflects partial progress before the API stopped responding. Excluded from cross-round comparisons per decision gad-64.
Details: Anthropic API returned 529 overloaded_error at tool_uses=55. Not a rate limit (full budget available), transient server overload. Previous attempt (not preserved) crashed at tool_uses=18 with the same error.
Full round 4 findings →Human review note
GAD v10 — api-interrupted per gad-64 (HTTP 529 pattern crashed the planning phase before scene implementation). Main title screen felt novel and visually cool which is where the 0.10 on UI polish comes from, but everything beyond it is absent: no game loop, no mechanics, no ingenuity payoff, no stable play. Preserved as a data point, excluded from cross-round quality comparisons (gad-63 + gad-64). User playtest 2026-04-09.
Reviewed by human · 2026-04-09
Dimension scores
Where the composite came from
Each dimension is scored 0.0 – 1.0 and combined using the weights in evals/escape-the-dungeon/gad.json. Human review dominates on purpose — process metrics alone can't rescue a broken run.
| Dimension | Score | Bar |
|---|---|---|
| Human review | 0.020 | |
| Requirement coverage | 0.083 | |
| Planning quality | 0.900 | |
| Per-task discipline | 0.000 | |
| Time efficiency | 0.700 |
Composite formula
How 0.000 was calculated
The composite score is a weighted sum of the dimensions above. Weights come from evals/escape-the-dungeon/gad.json. Contribution = score × weight; dimensions sorted by contribution so you can see what actually moved the needle.
| Dimension | Weight | Score | Contribution |
|---|---|---|---|
| planning_quality | 0.15 | 0.900 | 0.1350(72%) |
| time_efficiency | 0.05 | 0.700 | 0.0350(19%) |
| requirement_coverage | 0.15 | 0.083 | 0.0125(7%) |
| human_review | 0.30 | 0.020 | 0.0060(3%) |
| per_task_discipline | 0.15 | 0.000 | 0.0000(0%) |
| skill_accuracy | 0.10 | 0.000 | 0.0000(0%) |
| Weighted sum | 0.90 | 0.1885 |
Note: The weighted sum above (0.1885) doesn't exactly match the stored composite (0.0000). The difference is usually the v3 low-score cap (composite < 0.20 → 0.40, composite < 0.10 → 0.25) or a run with an older scoring pass.
Skill accuracy breakdown
Did the agent invoke the right skills at the right moments?
Skill accuracy data isn't relevant for this run (no expected trigger set).
Human review rubric
Where the reviewer scored this run best and worst
Each axis is a rubric dimension from evals/escape-the-dungeon/gad.json → human_review_rubric. The filled polygon shows the reviewer's per-dimension scores, 0.0 at center to 1.0 at the edge. The aggregate score (0.020) is the weighted sum of the dimensions using weights declared in the rubric.
| Dimension | Score | Weight |
|---|---|---|
Playability | 0.00 | 0.30 |
UI polish | 0.10 | 0.20 |
Mechanics implementation | 0.00 | 0.20 |
Ingenuity requirement met | 0.00 | 0.20 |
Stability | 0.00 | 0.10 |
Gate report
Requirement coverage
Reviewer notes on gates
G1 NOT MET — no game loop. Dist is scaffold-only title 'ESCAPE THE DUNGEON v10 — scaffold booted' with no navigation, combat, or forge wired. router.ts is an 11-line stub. G2 NOT MET — runes and spells data authored (221 lines of runes.ts) but no forge UI. G3 PARTIAL — scaffold title has iconify-icon + styled h1 but that is the only rendered surface. G4 NOT MET — pressure mechanics not implemented beyond data declarations.
Process metrics