Task gallery / Games / Version 1
Task evidence / console

tower-siege

saturatedCurrent calibratedChosen benchmark version
Reference solution (oracle)
Pass
No-op resistance check
Pass
QA verdict
Not recorded
Aggregate eligibility
Eligible version

The task

You're entered as the engineer in a bot tournament for Tower Siege, a small tower-defence game. You are the defender: attackers march down a set of parallel lanes toward your base, and every one that reaches it takes a bite out of your base's health. Build and upgrade towers to hold. Survive the whole assault — until the attacker has spent its budget and the field is clear, or the tick clock runs out with your base …

Excerpt only; not a complete executable task specification. Task ID task_0aaf7fc7cc7b4cf0 · version 1. Selection status: not fully recorded.

Results by model and effort

This exact task version only. Every rate is passes / valid runs with a Wilson 95% interval; n≤2 remains counts only. Macro aggregation does not apply within a single task. Baselines are summarized in the quality gates, not ranked. A passed no-op resistance check means that doing nothing did not solve the task.

Current evidence

Model / effortVerifier result [95% Wilson]Diagnostic labels / validHARNESS_ERROR · excludedOutcome classes
Claude Opus 5 · default (unverified)5/5 · 100.0% [56.6%, 100.0%]5 / 50
GOOD_FAILURE: 0
BAD_FAILURE: 0
GOOD_SUCCESS: 5
BAD_SUCCESS: 0

Legacy evidence

Model / effortVerifier result [95% Wilson]Diagnostic labels / validHARNESS_ERROR · excludedOutcome classes

Difficulty, with uncertainty

Selected-best difficulty: 0.0% [0.0%, 43.4%]. This is one minus the best measured current-row pass rate among rows with at least five valid runs.

The maximum is selected from the observed rows, so the estimate is optimistic. Its Wilson interval is not adjusted for model selection. “Hard” requires qualifying GOOD_FAILURE evidence; the threshold flag additionally requires the interval and recorded quality gates to meet the documented rule. This is not independent validation. Threshold flag: no.

Every recorded run

7 recorded trials, including controls. Diagnostic labels do not replace verifier reward. Labels and quality checks are source declarations, not independently attested here. Internal evidence remains private; a missing public URL is shown explicitly. Records with invalid or absent rewards are not scored.

Run / modelDate / sourceReward / eligibilityDiagnostic class / sourceResourcesEvidence
trial_2109f365c1704f8a
nop@@		Oct 10, 2026
console		0 · validNo diagnostic label
derived		Tokens: —
Turns: —
Cost: —		Not publicly available
trial_948756b921044011
oracle@@		Oct 10, 2026
console		1 · validNo diagnostic label
derived		Tokens: —
Turns: —
Cost: —		Not publicly available
trial_1c97679e51704202
Claude Opus 5 · default (unverified)		Oct 10, 2026
console		1 · validGOOD_SUCCESS
frozen-classifier		Tokens: 112,043
Turns: 75
Cost: $9.8718		Not publicly available
trial_7e68f02a9a03433e
Claude Opus 5 · default (unverified)		Oct 10, 2026
console		1 · validGOOD_SUCCESS
frozen-classifier		Tokens: 173,689
Turns: 142
Cost: $21.9787		Not publicly available
trial_89d680c5cea947b3
Claude Opus 5 · default (unverified)		Oct 10, 2026
console		1 · validGOOD_SUCCESS
frozen-classifier		Tokens: 98,959
Turns: 98
Cost: $10.1477		Not publicly available
trial_d407b618364b438e
Claude Opus 5 · default (unverified)		Oct 10, 2026
console		1 · validGOOD_SUCCESS
frozen-classifier		Tokens: 104,017
Turns: 110
Cost: $12.0019		Not publicly available
trial_fed535b534d04c74
Claude Opus 5 · default (unverified)		Oct 10, 2026
console		1 · validGOOD_SUCCESS
frozen-classifier		Tokens: 164,293
Turns: 162
Cost: $22.1947		Not publicly available