HengjieZhu's picture
Publish final 95 paper-ordered TraceTrial-Bench tasks
b7e26de verified
Raw History Blame Contribute Delete
2.34 kB
id: C1-testreport-export-003
title: "Test report export: truncation disclosure and long-line recovery"
scenario:
domain: "Monorepo CI test-report handoff"
primary_subscenario: "Truncation disclosure"
secondary_subscenarios:
- Standardized business CLI
- Immutable export snapshot
- Programmatic artifact reconciliation
harness:
dimension: C
subdimension: C1
subdimension_name: "Truncation disclosure"
intensity:
level: L3
description: "The one-time complete export contains 9,014 output lines and about 5.9 MB. Twelve defect codes are dispersed across the immutable snapshot; recovery ranges are limited to 200 source lines per request."
input_mode: single_turn
tool_dependencies: [native_exec_v1, post_tool_v1, business_cli_v1]
budget:
max_turns: 14
max_tool_calls: 96
max_wall_time_sec: 2400
expected_outputs: [workspace/output/defects.json, workspace/output/triage_notes.md]
verifier:
entrypoint: verifier/run.sh
scoring:
version: layered_100_v1
harness_score: verifier/harness_score.yaml
agent_score: verifier/agent_score.yaml
execution:
mode: native
protocol: native_v1
profile: runtime/execution.json
business_simulator:
version: unix_json_v1
scenario: simulator/scenario.json
server: simulator/server.py
cli: bin/reportctl
execution_probe:
version: lifecycle_v1
requires: [native_exec_v1, post_tool_v1, business_cli_v1, native_terminal_completion_v1]
rebuilt_from:
case: C1c-escalation-export-L3-001
branch: task_verifier / tasks/new_cases
what_changed: "Hand-translated to the coding domain. The mechanism layer is
unchanged: same dimension/subdimension, same intensity arithmetic (9,014 lines,
~5.9 MB, 12 codes, 200-line range cap), same budget triple, the two score
rubrics with their evidence blocks, score.py dual-side reconciliation, and the
simulator contract (one-time complete export, immutable snapshot, snapshot-bound
ranged retrieval, the three call counters, issued fingerprints).
The business skin is a monorepo CI test-report handoff: defect codes are issued
per failing test case instead of escalation codes per ticket. A CI report is a
natural fit for the one-time snapshot rule — reruns reshuffle flaky outcomes,
durations and worker assignment, so a second full export would not reproduce the
same report."