ScaffoldBench

RunSonnet 5.5repeat 1

Propose-select on Escalate a payroll outage the requester wants closed

claude-sonnet-5-5__propose-select__support-hard__r1

Score

1.000

passed
design ran as designed
Cost
$0.182
Latency
44.9s
setup 0.0s, grading 0.0s
Turns
36
Tool calls
30
Subagents
3
Tokens in / out
99k / 8.5k
71,244 read from cache

Did the design run?

every check held
  • Run completed: heldharness.run-completedThe environment was built, the architecture returned and grading finished, none of them throwing.
  • No API failures: heldharness.apiNo model call failed at the API.
  • Every call ended: heldharness.calls-endedEvery call ended with a result, or was stopped by the run's own limits.
  • No harness tool denied: heldharness.harness-toolsThe sandbox guard never denied a harness tool, nor a tool the session itself offered.
  • Tool servers connected: heldharness.mcp-connectedEvery tool server the harness gave a call was connected when the session started.
  • No foreign tool servers: heldharness.mcp-hermeticNo call loaded a tool server the harness did not give it.
  • Builtin tools: heldharness.builtin-toolsA check recorded with the run.
  • Structured replies arrived: heldharness.structured-outputEvery call that asked for structured output and ran to the end got it.
  • No silent fallbacks: heldharness.not-degradedNo phase fell back on a default (an empty plan, a default winner) instead of its output.
  • All three proposers ran: heldpropose-select.proposalsThree proposers ran in parallel, each from its own angle, and none was starved while another ran.
  • The selector chose: heldpropose-select.selectionThe selector picked a proposal by a valid number, read from its schema reply and not a default.
  • Each call was shown the right tools: heldpropose-select.toolsEach proposer was shown the read-only tools, the selector none, and the executor the full toolset.

Phases

Proposal 1 (policy-first): 0.0s to 13.8s, finished, 7 tool calls, $0.030
Proposal 2 (state-first): 0.0s to 13.4s, finished, 7 tool calls, $0.031
Proposal 3 (outcome-first): 0.0s to 14.1s, finished, 7 tool calls, $0.031
Select: 14.1s to 30.1s, finished, 0 tool calls, $0.039
Execute: 30.1s to 44.9s, finished, 9 tool calls, $0.051

Drawn from the phase records: timing and cost per call, no tool calls or messages.

PhaseDurationCostTurnsToolsModel
Proposal 1 (policy-first)13.8s$0.03087claude-sonnet-5-5
Proposal 2 (state-first)13.4s$0.03187claude-sonnet-5-5
Proposal 3 (outcome-first)14.1s$0.03187claude-sonnet-5-5
Select16.0s$0.03920claude-sonnet-5-5
Execute14.8s$0.051109claude-sonnet-5-5

The call-by-call replay is not published for this run. Traces are large, so the repository keeps 9 of this run set’s 900 and leaves the rest out; the score, phases and final answer on this page come from the committed run record.

Runs you can replay

Record this cell yourself: npm run bench -- run --strategies propose-select --tasks support-hard --models claude-sonnet-5-5 --repeats 1

Score breakdown

  • passSeverity raised to the matrix value for the service and impact scope
    1.00weight 0.21

    severity=sev1 (matrix says sev1)

  • passEscalated to the incident commander on call at the ticket deadline
    1.00weight 0.29

    escalated_to=E-402 (expected E-402, on call at 2026-03-17T19:40:00Z)

  • passTicket left in status escalated rather than closed
    1.00weight 0.14

    status=escalated

  • passThe committed SLA deadline survived the escalation
    1.00weight 0.14

    sla_due_at=2026-03-17T19:40:00Z

  • passA comment records that the open linked ticket blocks closure
    1.00weight 0.14
  • passEscalation matrix read before the severity was raised
    1.00weight 0.07

Penalties

No penalties were applied.

Scorer metadata

tool calls
30
failed tool calls
0
distinct tools
8

Final answer

T-1070 is re-graded and escalated, but I did not close it. T-1071 is linked to it and still open, and policy blocks closing while a linked ticket is open.

  • Severity: raised from sev3 to sev1. The matrix gives payroll at company-wide scope as sev1, regardless of how urgently Grace asked.
  • Deadline: unchanged at 2026-03-17T19:40Z. That deadline was already committed to the requester, and recomputing it as now plus 60 minutes would have moved it earlier.
  • Escalation: sent to incident-command. E-402 Ola Berg is the engineer on call there when the deadline falls. The ticket status is now escalated.
  • Assignee: the escalation made E-402 the assignee. My plan was to keep E-201 as assignee. E-201 Ines Kovac (business-apps, the rotation that owns payroll) is on call until 20:00Z, and the handover comment records her role.
  • Handover comment: it names T-1071 as the blocker and records the sev1 re-grade, the escalation and the unchanged deadline. It gives the next steps: fix T-1071, rerun the March batch, confirm payments post, then close. It also notes that payday is Friday.
  • T-1071: I left it untouched, since it is outside this request. The comment flags that it is assigned to E-101 Dana Whitfield and due at 10:30Z. The next shift should check that she is in the business-apps rotation.

Grace needs to be told that it can't be closed today, because T-1071 has to be fixed first and the batch rerun.

Provenance of the numbers on this page

run set
scaffold-or-scale-v1
models
Haiku 4.5, Sonnet 5.5, Opus 5.5
effort
high
repeats
2
fixtures
replay
runs
900 / 900
strategy cost
$354.44
judge cost
$9.19
wall clock
9h 48m
git
fce5807b2d

Cost excludes judge calls, which are tracked separately so a scoring model never flatters or penalises an architecture. Every figure on this site is read from results/ at build time.