task
stringclasses
9 values
state
stringlengths
195
695
instructions
stringclasses
33 values
criteria_keys
listlengths
2
5
criteria_values
listlengths
2
5
label_alias
stringclasses
5 values
label_key
stringclasses
22 values
meta
stringlengths
80
133
debate.winner
{"rubric": "Prefer hard evidence over vibes; break ties toward the side with more hard evidence.", "question": "We should raise prices 10%", "side_a": {"label": "support", "claims": [{"text": "Competitors already priced 12% higher", "hard_evidence": true}, {"text": "Gross margin is below target", "hard_evidence": true}...
Who wins under the stated evidence policy? One Criteria letter.
[ "A", "B" ]
[ "oppose: Side B — oppose the proposal", "support: Side A — support the proposal" ]
B
support
{"gym": "debate_judge", "pro_ev": 2, "con_ev": 1, "schema": "choice", "hard": false, "n_options": 2}
debate.winner
{"side_b": {"label": "oppose", "claims": [{"text": "Competitors will steal our vibes", "hard_evidence": false}]}, "side_a": {"label": "support", "claims": [{"text": "Design partners ask for source escrow", "hard_evidence": true}, {"text": "Inbound PRs already fix bugs we lack time for", "hard_evidence": true}, {"text":...
Given the claims and evidence flags, who should win the debate? One letter.
[ "A", "B" ]
[ "oppose: Side B — oppose the proposal", "support: Side A — support the proposal" ]
B
support
{"gym": "debate_judge", "pro_ev": 2, "con_ev": 0, "schema": "choice", "hard": false, "n_options": 2}
debate.confidence
{"rubric": "Prefer hard evidence over vibes; break ties toward the side with more hard evidence.", "side_a": {"label": "support", "claims": [{"text": "Rust is trendy on HN", "hard_evidence": false}, {"text": "Rust prototype cut CPU 35% in bench", "hard_evidence": true}, {"text": "p99 latency is 4x SLO on hot path", "ha...
How confident should we be in the judgment? Reply with one level letter.
[ "A", "B", "C" ]
[ "2: high confidence", "0: low confidence", "1: medium confidence" ]
C
1
{"gym": "debate_judge", "schema_hint": "score", "schema": "choice", "hard": false, "n_options": 3}
debate.winner
{"question": "We should open-source the core library", "side_a": {"label": "support", "claims": [{"text": "Inbound PRs already fix bugs we lack time for", "hard_evidence": true}, {"text": "Open source is always good actually", "hard_evidence": false}, {"text": "Design partners ask for source escrow", "hard_evidence": t...
Who wins under the stated evidence policy? One Criteria letter.
[ "A", "B" ]
[ "oppose: Side B — oppose the proposal", "support: Side A — support the proposal" ]
A
oppose
{"gym": "debate_judge", "pro_ev": 2, "con_ev": 2, "schema": "choice", "hard": false, "n_options": 2}
alloc.can_fund_all
{"requests": [{"id": "office_snacks", "title": "Office snacks", "cost": 6, "urgency": 0}, {"id": "hiring_bonus", "title": "Hiring referral bonuses", "cost": 56, "urgency": 1}, {"id": "cdn_upgrade", "title": "CDN capacity upgrade", "cost": 50, "urgency": 2}, {"id": "customer_refund_pool", "title": "Customer refund reser...
Would summing all costs fit under budget_remaining? Answer with one letter.
[ "A", "B" ]
[ "yes: Budget covers the sum of all request costs", "no: Budget is insufficient for all requests" ]
B
no
{"gym": "resource_allocator", "schema_hint": "noul", "schema": "choice", "hard": false, "n_options": 2}
debate.enough_evidence
{"question": "We should hire more junior engineers", "side_b": {"label": "oppose", "claims": [{"text": "Ramp cost exceeds budget this half", "hard_evidence": true}, {"text": "Critical path needs senior ownership now", "hard_evidence": true}, {"text": "Juniors ask too many questions", "hard_evidence": false}]}, "rubric"...
Evidence sufficiency check — yes or no via option letter.
[ "A", "B" ]
[ "no: Too thin or tied — needs more evidence", "yes: Enough hard evidence to decide" ]
B
yes
{"gym": "debate_judge", "schema_hint": "noul", "schema": "choice", "hard": false, "n_options": 2}
debate.confidence
{"side_a": {"label": "support", "claims": [{"text": "My friend likes remote", "hard_evidence": false}, {"text": "Attrition dropped 20% after remote policy", "hard_evidence": true}]}, "question": "Remote-first improves productivity", "side_b": {"label": "oppose", "claims": [{"text": "Some customers require on-site works...
Rate decision confidence from the evidence balance. One letter.
[ "A", "B", "C" ]
[ "2: high confidence", "1: medium confidence", "0: low confidence" ]
B
1
{"gym": "debate_judge", "schema_hint": "score", "schema": "choice", "hard": false, "n_options": 3}
ticket.needs_human
{"channel": "email", "ticket": {"subject": "Courier says package damaged", "body": "Thread summary — Package marked delivered but nothing arrived | wait_days=10 | tier=enterprise | priority=p1", "tier": "enterprise", "priority": "p1", "wait_days": 10}}
Does this ticket need a human agent (not a bot)? Reply with one option letter.
[ "A", "B" ]
[ "yes: Needs a human (money, legal, anger, long wait, outage)", "no: Safe for bot / routine" ]
A
yes
{"gym": "ticket_dungeon", "schema_hint": "noul", "schema": "choice", "hard": false, "n_options": 2}
debate.confidence
{"side_b": {"label": "oppose", "claims": [{"text": "Last Friday deploy caused a Sev-1", "hard_evidence": true}, {"text": "On-call coverage is thin on weekends", "hard_evidence": true}, {"text": "I just don't like Fridays", "hard_evidence": false}]}, "question": "We should deploy on Friday", "rubric": "Prefer hard evide...
Rate decision confidence from the evidence balance. One letter.
[ "A", "B", "C" ]
[ "1: medium confidence", "0: low confidence", "2: high confidence" ]
B
0
{"gym": "debate_judge", "schema_hint": "score", "schema": "choice", "hard": false, "n_options": 3}
debate.enough_evidence
{"side_b": {"label": "oppose", "claims": [{"text": "Competitors will steal our vibes", "hard_evidence": false}, {"text": "Core contains proprietary scoring IP", "hard_evidence": true}, {"text": "Support load estimates exceed current headcount", "hard_evidence": true}]}, "rubric": "Prefer hard evidence over vibes; break...
Evidence sufficiency check — yes or no via option letter.
[ "A", "B" ]
[ "no: Too thin or tied — needs more evidence", "yes: Enough hard evidence to decide" ]
B
yes
{"gym": "debate_judge", "schema_hint": "noul", "schema": "choice", "hard": false, "n_options": 2}
alloc.pressure
{"requests": [{"id": "compliance_audit", "title": "Compliance audit support", "cost": 79, "urgency": 3}, {"id": "docs_rewrite", "title": "Docs rewrite contractor", "cost": 26, "urgency": 0}, {"id": "marketing_campaign", "title": "Q3 marketing campaign", "cost": 129, "urgency": 1}, {"id": "security_patch", "title": "Cri...
How constrained is the budget relative to demand? Pick one letter.
[ "A", "B", "C" ]
[ "2: high — demand exceeds budget", "0: low — budget comfortably covers demand", "1: medium — close to capacity" ]
A
2
{"gym": "resource_allocator", "schema_hint": "score", "ratio": 1.4979757085020242, "schema": "choice", "hard": false, "n_options": 3}
alloc.can_fund_all
{"currency": "USD", "budget_remaining": 204, "requests": [{"id": "oncall_stipend", "title": "On-call stipend top-up", "cost": 53, "urgency": 2}, {"id": "ml_experiment", "title": "GPU hours for ML experiment", "cost": 97, "urgency": 1}, {"id": "compliance_audit", "title": "Compliance audit support", "cost": 61, "urgency...
Is the budget enough to approve every request? Reply with one option letter.
[ "A", "B" ]
[ "yes: Budget covers the sum of all request costs", "no: Budget is insufficient for all requests" ]
B
no
{"gym": "resource_allocator", "schema_hint": "noul", "schema": "choice", "hard": false, "n_options": 2}
ticket.needs_human
{"ticket": {"subject": "Invoice shows unexpected billing amount", "body": "From customer (free): Billing portal shows an incorrect charge. SLA wait=10d priority=p0.", "tier": "free", "priority": "p0", "wait_days": 10}, "channel": "phone"}
Can a bot handle this safely, or must a person intervene? One letter.
[ "A", "B" ]
[ "no: Safe for bot / routine", "yes: Needs a human (money, legal, anger, long wait, outage)" ]
B
yes
{"gym": "ticket_dungeon", "schema_hint": "noul", "schema": "choice", "hard": false, "n_options": 2}
debate.winner
{"side_b": {"label": "oppose", "claims": [{"text": "Offices look empty and sad", "hard_evidence": false}, {"text": "New-grad ramp surveys show slower onboarding", "hard_evidence": true}, {"text": "Some customers require on-site workshops", "hard_evidence": true}]}, "side_a": {"label": "support", "claims": [{"text": "At...
Given the claims and evidence flags, who should win the debate? One letter.
[ "A", "B" ]
[ "support: Side A — support the proposal", "oppose: Side B — oppose the proposal" ]
B
oppose
{"gym": "debate_judge", "pro_ev": 1, "con_ev": 2, "schema": "choice", "hard": false, "n_options": 2}
ticket.needs_human
{"ticket": {"subject": "Duplicate card charge on order A-100", "body": "Customer says: Invoice shows wrong amount on my card. Wait time so far: 5 days. Tier: pro.", "tier": "pro", "priority": "p2", "wait_days": 5}, "channel": "phone"}
Can a bot handle this safely, or must a person intervene? One letter.
[ "A", "B" ]
[ "no: Safe for bot / routine", "yes: Needs a human (money, legal, anger, long wait, outage)" ]
B
yes
{"gym": "ticket_dungeon", "schema_hint": "noul", "schema": "choice", "hard": false, "n_options": 2}
debate.winner
{"question": "We should hire more junior engineers", "side_a": {"label": "support", "claims": [{"text": "Juniors are cheaper vibes", "hard_evidence": false}, {"text": "Senior mentors have spare capacity", "hard_evidence": true}, {"text": "Internship NPS was high and conversion strong", "hard_evidence": true}]}, "rubric...
Which side is better supported by evidence? Reply with one option letter.
[ "A", "B" ]
[ "oppose: Side B — oppose the proposal", "support: Side A — support the proposal" ]
B
support
{"gym": "debate_judge", "pro_ev": 2, "con_ev": 2, "schema": "choice", "hard": false, "n_options": 2}
debate.enough_evidence
{"side_a": {"label": "support", "claims": [{"text": "Fridays feel lucky", "hard_evidence": false}]}, "question": "We should deploy on Friday", "rubric": "Prefer hard evidence over vibes; break ties toward the side with more hard evidence.", "side_b": {"label": "oppose", "claims": [{"text": "I just don't like Fridays", ...
Is there enough hard evidence to decide confidently? Reply with one letter.
[ "A", "B" ]
[ "yes: Enough hard evidence to decide", "no: Too thin or tied — needs more evidence" ]
A
yes
{"gym": "debate_judge", "schema_hint": "noul", "schema": "choice", "hard": false, "n_options": 2}
ticket.urgency
{"channel": "chat", "ticket": {"subject": "Charged twice for the same invoice A-678", "body": "Message: Billing portal shows an incorrect charge. Priority flag: p1. Account tier: enterprise.", "tier": "enterprise", "priority": "p1", "wait_days": 10}}
How urgent is this ticket? Pick one level letter from Criteria.
[ "A", "B", "C" ]
[ "1: medium — same day", "0: low — can wait", "2: high — immediate" ]
C
2
{"gym": "ticket_dungeon", "schema_hint": "score", "urgency": 2, "schema": "choice", "hard": false, "n_options": 3}
ticket.route
{"ticket": {"subject": "Compare Business vs Enterprise plans", "body": "Thread summary — What plan is best for a 200 person team? | wait_days=6 | tier=pro | priority=p2", "tier": "pro", "priority": "p2", "wait_days": 6}, "channel": "chat"}
Best team for this case? Answer with a single option letter.
[ "A", "B", "C", "D", "E" ]
[ "other: unclear or general questions", "sales: pricing, upgrades, new accounts", "technical: bugs, crashes, outages, integrations", "shipping: delivery, packages, tracking", "billing: payment, charges, invoices, refunds" ]
B
sales
{"gym": "ticket_dungeon", "domain_seed": "sales", "schema": "choice", "hard": false, "n_options": 5}
debate.winner
{"question": "Remote-first improves productivity", "side_a": {"label": "support", "claims": [{"text": "My friend likes remote", "hard_evidence": false}, {"text": "Hiring pipeline widened internationally", "hard_evidence": true}, {"text": "Attrition dropped 20% after remote policy", "hard_evidence": true}]}, "rubric": "...
Which side is better supported by evidence? Reply with one option letter.
[ "A", "B" ]
[ "support: Side A — support the proposal", "oppose: Side B — oppose the proposal" ]
A
support
{"gym": "debate_judge", "pro_ev": 2, "con_ev": 2, "schema": "choice", "hard": false, "n_options": 2}
alloc.fund_next
{"requests": [{"id": "oncall_stipend", "title": "On-call stipend top-up", "cost": 46, "urgency": 2}, {"id": "customer_refund_pool", "title": "Customer refund reserve", "cost": 69, "urgency": 2}, {"id": "laptop_refresh", "title": "Engineer laptop refresh", "cost": 74, "urgency": 1}], "budget_remaining": 209, "currency":...
Which single request should be funded next given the budget? Reply with one letter.
[ "A", "B", "C" ]
[ "laptop_refresh: Engineer laptop refresh (cost=74, urgency=1)", "oncall_stipend: On-call stipend top-up (cost=46, urgency=2)", "customer_refund_pool: Customer refund reserve (cost=69, urgency=2)" ]
B
oncall_stipend
{"gym": "resource_allocator", "schema": "choice", "hard": false, "n_options": 3}
ticket.route
{"channel": "email", "ticket": {"subject": "How do I change my display name?", "body": "From customer (enterprise): Where can I find documentation?. SLA wait=1d priority=p0.", "tier": "enterprise", "priority": "p0", "wait_days": 1}}
Route this support ticket to the best team. Reply with one option letter.
[ "A", "B", "C", "D", "E" ]
[ "billing: payment, charges, invoices, refunds", "shipping: delivery, packages, tracking", "sales: pricing, upgrades, new accounts", "other: unclear or general questions", "technical: bugs, crashes, outages, integrations" ]
D
other
{"gym": "ticket_dungeon", "domain_seed": "other", "schema": "choice", "hard": false, "n_options": 5}
alloc.pressure
{"budget_remaining": 267, "requests": [{"id": "laptop_refresh", "title": "Engineer laptop refresh", "cost": 84, "urgency": 1}, {"id": "marketing_campaign", "title": "Q3 marketing campaign", "cost": 110, "urgency": 1}, {"id": "office_snacks", "title": "Office snacks", "cost": 24, "urgency": 0}, {"id": "oncall_stipend", ...
Rate overall budget pressure. Reply with one level letter.
[ "A", "B", "C" ]
[ "1: medium — close to capacity", "2: high — demand exceeds budget", "0: low — budget comfortably covers demand" ]
B
2
{"gym": "resource_allocator", "schema_hint": "score", "ratio": 1.205992509363296, "schema": "choice", "hard": false, "n_options": 3}
ticket.urgency
{"ticket": {"subject": "General question about account", "body": "Thread summary — How do I change my display name? | wait_days=1 | tier=free | priority=p2", "tier": "free", "priority": "p2", "wait_days": 1}, "channel": "phone"}
How urgent is this ticket? Pick one level letter from Criteria.
[ "A", "B", "C" ]
[ "2: high — immediate", "1: medium — same day", "0: low — can wait" ]
C
0
{"gym": "ticket_dungeon", "schema_hint": "score", "urgency": 0, "schema": "choice", "hard": false, "n_options": 3}
debate.enough_evidence
{"rubric": "Prefer hard evidence over vibes; break ties toward the side with more hard evidence.", "side_a": {"label": "support", "claims": [{"text": "Fridays feel lucky", "hard_evidence": false}, {"text": "Customers asked for the fix this week", "hard_evidence": true}]}, "side_b": {"label": "oppose", "claims": [{"text...
Evidence sufficiency check — yes or no via option letter.
[ "A", "B" ]
[ "no: Too thin or tied — needs more evidence", "yes: Enough hard evidence to decide" ]
B
yes
{"gym": "debate_judge", "schema_hint": "noul", "schema": "choice", "hard": false, "n_options": 2}
alloc.pressure
{"budget_remaining": 180, "currency": "USD", "requests": [{"id": "office_snacks", "title": "Office snacks", "cost": 15, "urgency": 0}, {"id": "cdn_upgrade", "title": "CDN capacity upgrade", "cost": 51, "urgency": 2}, {"id": "design_sprint", "title": "Design sprint facilitator", "cost": 18, "urgency": 0}, {"id": "laptop...
Budget tightness score — choose one Criteria letter.
[ "A", "B", "C" ]
[ "1: medium — close to capacity", "0: low — budget comfortably covers demand", "2: high — demand exceeds budget" ]
A
1
{"gym": "resource_allocator", "schema_hint": "score", "ratio": 1.1, "schema": "choice", "hard": false, "n_options": 3}
debate.winner
{"question": "We should raise prices 10%", "rubric": "Prefer hard evidence over vibes; break ties toward the side with more hard evidence.", "side_a": {"label": "support", "claims": [{"text": "Prices going up sounds cool", "hard_evidence": false}]}, "side_b": {"label": "oppose", "claims": [{"text": "Enterprise deals ar...
Who wins under the stated evidence policy? One Criteria letter.
[ "A", "B" ]
[ "support: Side A — support the proposal", "oppose: Side B — oppose the proposal" ]
B
oppose
{"gym": "debate_judge", "pro_ev": 0, "con_ev": 1, "schema": "choice", "hard": false, "n_options": 2}
ticket.urgency
{"channel": "chat", "ticket": {"subject": "Package stuck in transit for 9 days", "body": "Thread summary — Delivery ETA keeps slipping | wait_days=9 | tier=enterprise | priority=p0", "tier": "enterprise", "priority": "p0", "wait_days": 9}}
How urgent is this ticket? Pick one level letter from Criteria.
[ "A", "B", "C" ]
[ "2: high — immediate", "1: medium — same day", "0: low — can wait" ]
A
2
{"gym": "ticket_dungeon", "schema_hint": "score", "urgency": 2, "schema": "choice", "hard": false, "n_options": 3}
debate.winner
{"question": "We should deploy on Friday", "side_b": {"label": "oppose", "claims": [{"text": "On-call coverage is thin on weekends", "hard_evidence": true}, {"text": "I just don't like Fridays", "hard_evidence": false}]}, "rubric": "Prefer hard evidence over vibes; break ties toward the side with more hard evidence.", ...
Which side is better supported by evidence? Reply with one option letter.
[ "A", "B" ]
[ "support: Side A — support the proposal", "oppose: Side B — oppose the proposal" ]
A
support
{"gym": "debate_judge", "pro_ev": 2, "con_ev": 1, "schema": "choice", "hard": false, "n_options": 2}
alloc.can_fund_all
{"requests": [{"id": "oncall_stipend", "title": "On-call stipend top-up", "cost": 38, "urgency": 2}, {"id": "design_sprint", "title": "Design sprint facilitator", "cost": 34, "urgency": 0}, {"id": "marketing_campaign", "title": "Q3 marketing campaign", "cost": 121, "urgency": 1}, {"id": "compliance_audit", "title": "Co...
Would summing all costs fit under budget_remaining? Answer with one letter.
[ "A", "B" ]
[ "yes: Budget covers the sum of all request costs", "no: Budget is insufficient for all requests" ]
A
yes
{"gym": "resource_allocator", "schema_hint": "noul", "schema": "choice", "hard": false, "n_options": 2}
debate.confidence
{"question": "We should open-source the core library", "side_b": {"label": "oppose", "claims": [{"text": "Core contains proprietary scoring IP", "hard_evidence": true}, {"text": "Competitors will steal our vibes", "hard_evidence": false}]}, "rubric": "Prefer hard evidence over vibes; break ties toward the side with mor...
Rate decision confidence from the evidence balance. One letter.
[ "A", "B", "C" ]
[ "1: medium confidence", "2: high confidence", "0: low confidence" ]
C
0
{"gym": "debate_judge", "schema_hint": "score", "schema": "choice", "hard": false, "n_options": 3}
debate.enough_evidence
{"question": "Remote-first improves productivity", "side_b": {"label": "oppose", "claims": [{"text": "Offices look empty and sad", "hard_evidence": false}, {"text": "New-grad ramp surveys show slower onboarding", "hard_evidence": true}, {"text": "Some customers require on-site workshops", "hard_evidence": true}]}, "sid...
Is there enough hard evidence to decide confidently? Reply with one letter.
[ "A", "B" ]
[ "no: Too thin or tied — needs more evidence", "yes: Enough hard evidence to decide" ]
B
yes
{"gym": "debate_judge", "schema_hint": "noul", "schema": "choice", "hard": false, "n_options": 2}
debate.enough_evidence
{"side_b": {"label": "oppose", "claims": [{"text": "Team has 1 Rust-experienced engineer", "hard_evidence": true}, {"text": "I dream in garbage collectors", "hard_evidence": false}]}, "rubric": "Prefer hard evidence over vibes; break ties toward the side with more hard evidence.", "side_a": {"label": "support", "claims...
Evidence sufficiency check — yes or no via option letter.
[ "A", "B" ]
[ "no: Too thin or tied — needs more evidence", "yes: Enough hard evidence to decide" ]
B
yes
{"gym": "debate_judge", "schema_hint": "noul", "schema": "choice", "hard": false, "n_options": 2}
ticket.needs_human
{"channel": "email", "ticket": {"subject": "Need enterprise pricing for 200 seats", "body": "Thread summary — Interested in enterprise pricing and a demo | wait_days=9 | tier=free | priority=p0", "tier": "free", "priority": "p0", "wait_days": 9}}
Human escalation required? Pick yes or no via the option letter.
[ "A", "B" ]
[ "no: Safe for bot / routine", "yes: Needs a human (money, legal, anger, long wait, outage)" ]
B
yes
{"gym": "ticket_dungeon", "schema_hint": "noul", "schema": "choice", "hard": false, "n_options": 2}
ticket.urgency
{"channel": "phone", "ticket": {"subject": "API returns 500 on /v1/invoices", "body": "From customer (pro): Stack trace on checkout — looks like a bug. SLA wait=9d priority=p1.", "tier": "pro", "priority": "p1", "wait_days": 9}}
How urgent is this ticket? Pick one level letter from Criteria.
[ "A", "B", "C" ]
[ "0: low — can wait", "1: medium — same day", "2: high — immediate" ]
C
2
{"gym": "ticket_dungeon", "schema_hint": "score", "urgency": 2, "schema": "choice", "hard": false, "n_options": 3}
debate.enough_evidence
{"side_b": {"label": "oppose", "claims": [{"text": "Customers will be mad forever", "hard_evidence": false}]}, "rubric": "Prefer hard evidence over vibes; break ties toward the side with more hard evidence.", "side_a": {"label": "support", "claims": [{"text": "Prices going up sounds cool", "hard_evidence": false}, {"te...
Evidence sufficiency check — yes or no via option letter.
[ "A", "B" ]
[ "no: Too thin or tied — needs more evidence", "yes: Enough hard evidence to decide" ]
B
yes
{"gym": "debate_judge", "schema_hint": "noul", "schema": "choice", "hard": false, "n_options": 2}
debate.confidence
{"rubric": "Prefer hard evidence over vibes; break ties toward the side with more hard evidence.", "question": "Remote-first improves productivity", "side_b": {"label": "oppose", "claims": [{"text": "Some customers require on-site workshops", "hard_evidence": true}]}, "side_a": {"label": "support", "claims": [{"text": ...
Rate decision confidence from the evidence balance. One letter.
[ "A", "B", "C" ]
[ "2: high confidence", "0: low confidence", "1: medium confidence" ]
B
0
{"gym": "debate_judge", "schema_hint": "score", "schema": "choice", "hard": false, "n_options": 3}
debate.confidence
{"question": "We should raise prices 10%", "rubric": "Prefer hard evidence over vibes; break ties toward the side with more hard evidence.", "side_a": {"label": "support", "claims": [{"text": "Competitors already priced 12% higher", "hard_evidence": true}, {"text": "Gross margin is below target", "hard_evidence": true}...
How confident should we be in the judgment? Reply with one level letter.
[ "A", "B", "C" ]
[ "1: medium confidence", "0: low confidence", "2: high confidence" ]
A
1
{"gym": "debate_judge", "schema_hint": "score", "schema": "choice", "hard": false, "n_options": 3}
ticket.urgency
{"ticket": {"subject": "Courier says package damaged", "body": "Thread summary — Delivery ETA keeps slipping | wait_days=7 | tier=free | priority=p0", "tier": "free", "priority": "p0", "wait_days": 7}, "channel": "phone"}
Rate urgency for this ticket on the given scale. Reply with one option letter.
[ "A", "B", "C" ]
[ "0: low — can wait", "1: medium — same day", "2: high — immediate" ]
C
2
{"gym": "ticket_dungeon", "schema_hint": "score", "urgency": 2, "schema": "choice", "hard": false, "n_options": 3}
ticket.route
{"ticket": {"subject": "Interested in upgrading to Pro", "body": "We need a quote for upgrading seats They have waited 6 days. Segment=free.", "tier": "free", "priority": "p0", "wait_days": 6}, "channel": "chat"}
Route this support ticket to the best team. Reply with one option letter.
[ "A", "B", "C", "D" ]
[ "sales: pricing, upgrades, new accounts", "shipping: delivery, packages, tracking", "other: unclear or general questions", "billing: payment, charges, invoices, refunds" ]
A
sales
{"gym": "ticket_dungeon", "domain_seed": "sales", "schema": "choice", "hard": false, "n_options": 4}
ticket.needs_human
{"channel": "email", "ticket": {"subject": "Compare Business vs Enterprise plans", "body": "Thread summary — Interested in enterprise pricing and a demo | wait_days=0 | tier=enterprise | priority=p1", "tier": "enterprise", "priority": "p1", "wait_days": 0}}
Should a human review this before automated handling? Reply with one letter.
[ "A", "B" ]
[ "yes: Needs a human (money, legal, anger, long wait, outage)", "no: Safe for bot / routine" ]
B
no
{"gym": "ticket_dungeon", "schema_hint": "noul", "schema": "choice", "hard": false, "n_options": 2}
alloc.fund_next
{"requests": [{"id": "oncall_stipend", "title": "On-call stipend top-up", "cost": 45, "urgency": 2}, {"id": "office_snacks", "title": "Office snacks", "cost": 8, "urgency": 0}, {"id": "ml_experiment", "title": "GPU hours for ML experiment", "cost": 91, "urgency": 1}], "currency": "USD", "policy": "Fund at most one requ...
Select the next funding target under the policy. One option letter.
[ "A", "B", "C" ]
[ "ml_experiment: GPU hours for ML experiment (cost=91, urgency=1)", "oncall_stipend: On-call stipend top-up (cost=45, urgency=2)", "office_snacks: Office snacks (cost=8, urgency=0)" ]
B
oncall_stipend
{"gym": "resource_allocator", "schema": "choice", "hard": false, "n_options": 3}
debate.enough_evidence
{"question": "We should raise prices 10%", "side_b": {"label": "oppose", "claims": [{"text": "Customers will be mad forever", "hard_evidence": false}]}, "side_a": {"label": "support", "claims": [{"text": "Competitors already priced 12% higher", "hard_evidence": true}]}, "rubric": "Prefer hard evidence over vibes; break...
Is there enough hard evidence to decide confidently? Reply with one letter.
[ "A", "B" ]
[ "yes: Enough hard evidence to decide", "no: Too thin or tied — needs more evidence" ]
B
no
{"gym": "debate_judge", "schema_hint": "noul", "schema": "choice", "hard": false, "n_options": 2}
alloc.pressure
{"requests": [{"id": "ml_experiment", "title": "GPU hours for ML experiment", "cost": 98, "urgency": 1}, {"id": "hiring_bonus", "title": "Hiring referral bonuses", "cost": 42, "urgency": 1}, {"id": "docs_rewrite", "title": "Docs rewrite contractor", "cost": 33, "urgency": 0}], "currency": "USD", "policy": "Fund at most...
Budget tightness score — choose one Criteria letter.
[ "A", "B", "C" ]
[ "2: high — demand exceeds budget", "1: medium — close to capacity", "0: low — budget comfortably covers demand" ]
A
2
{"gym": "resource_allocator", "schema_hint": "score", "ratio": 1.4786324786324787, "schema": "choice", "hard": false, "n_options": 3}
alloc.fund_next
{"policy": "Fund at most one request per step; prefer higher urgency among affordable.", "currency": "USD", "requests": [{"id": "design_sprint", "title": "Design sprint facilitator", "cost": 18, "urgency": 0}, {"id": "security_patch", "title": "Critical security patch", "cost": 50, "urgency": 3}, {"id": "marketing_camp...
Select the next funding target under the policy. One option letter.
[ "A", "B", "C", "D", "E" ]
[ "hiring_bonus: Hiring referral bonuses (cost=59, urgency=1)", "design_sprint: Design sprint facilitator (cost=18, urgency=0)", "security_patch: Critical security patch (cost=50, urgency=3)", "oncall_stipend: On-call stipend top-up (cost=46, urgency=2)", "marketing_campaign: Q3 marketing campaign (cost=122, ...
C
security_patch
{"gym": "resource_allocator", "schema": "choice", "hard": false, "n_options": 5}
ticket.route
{"ticket": {"subject": "Interested in upgrading to Pro", "body": "From customer (pro): Interested in enterprise pricing and a demo. SLA wait=7d priority=p0.", "tier": "pro", "priority": "p0", "wait_days": 7}, "channel": "chat"}
Select the correct routing destination for the ticket state.
[ "A", "B", "C", "D", "E" ]
[ "shipping: delivery, packages, tracking", "other: unclear or general questions", "sales: pricing, upgrades, new accounts", "technical: bugs, crashes, outages, integrations", "billing: payment, charges, invoices, refunds" ]
C
sales
{"gym": "ticket_dungeon", "domain_seed": "sales", "schema": "choice", "hard": false, "n_options": 5}
ticket.needs_human
{"channel": "chat", "ticket": {"subject": "Cannot log in after password reset", "body": "From customer (free): Stack trace on checkout — looks like a bug. SLA wait=1d priority=p1.", "tier": "free", "priority": "p1", "wait_days": 1}}
Should a human review this before automated handling? Reply with one letter.
[ "A", "B" ]
[ "no: Safe for bot / routine", "yes: Needs a human (money, legal, anger, long wait, outage)" ]
A
no
{"gym": "ticket_dungeon", "schema_hint": "noul", "schema": "choice", "hard": false, "n_options": 2}
alloc.fund_next
{"budget_remaining": 282, "policy": "Fund at most one request per step; prefer higher urgency among affordable.", "requests": [{"id": "marketing_campaign", "title": "Q3 marketing campaign", "cost": 123, "urgency": 1}, {"id": "laptop_refresh", "title": "Engineer laptop refresh", "cost": 73, "urgency": 1}, {"id": "ml_exp...
Given budget_remaining and urgencies, which request wins? One letter.
[ "A", "B", "C", "D", "E" ]
[ "oncall_stipend: On-call stipend top-up (cost=36, urgency=2)", "design_sprint: Design sprint facilitator (cost=19, urgency=0)", "laptop_refresh: Engineer laptop refresh (cost=73, urgency=1)", "marketing_campaign: Q3 marketing campaign (cost=123, urgency=1)", "ml_experiment: GPU hours for ML experiment (cost...
A
oncall_stipend
{"gym": "resource_allocator", "schema": "choice", "hard": false, "n_options": 5}
alloc.pressure
{"currency": "USD", "budget_remaining": 133, "requests": [{"id": "marketing_campaign", "title": "Q3 marketing campaign", "cost": 123, "urgency": 1}, {"id": "security_patch", "title": "Critical security patch", "cost": 36, "urgency": 3}, {"id": "docs_rewrite", "title": "Docs rewrite contractor", "cost": 29, "urgency": 0...
Rate overall budget pressure. Reply with one level letter.
[ "A", "B", "C" ]
[ "2: high — demand exceeds budget", "0: low — budget comfortably covers demand", "1: medium — close to capacity" ]
A
2
{"gym": "resource_allocator", "schema_hint": "score", "ratio": 1.8045112781954886, "schema": "choice", "hard": false, "n_options": 3}
ticket.needs_human
{"ticket": {"subject": "Invoice shows unexpected billing amount", "body": "Please reverse the duplicate payment They have waited 5 days. Segment=free.", "tier": "free", "priority": "p2", "wait_days": 5}, "channel": "chat"}
Can a bot handle this safely, or must a person intervene? One letter.
[ "A", "B" ]
[ "no: Safe for bot / routine", "yes: Needs a human (money, legal, anger, long wait, outage)" ]
B
yes
{"gym": "ticket_dungeon", "schema_hint": "noul", "schema": "choice", "hard": false, "n_options": 2}
alloc.pressure
{"requests": [{"id": "ml_experiment", "title": "GPU hours for ML experiment", "cost": 81, "urgency": 1}, {"id": "marketing_campaign", "title": "Q3 marketing campaign", "cost": 119, "urgency": 1}, {"id": "design_sprint", "title": "Design sprint facilitator", "cost": 21, "urgency": 0}, {"id": "office_snacks", "title": "O...
Budget tightness score — choose one Criteria letter.
[ "A", "B", "C" ]
[ "0: low — budget comfortably covers demand", "1: medium — close to capacity", "2: high — demand exceeds budget" ]
A
0
{"gym": "resource_allocator", "schema_hint": "score", "ratio": 0.7635135135135135, "schema": "choice", "hard": false, "n_options": 3}
alloc.can_fund_all
{"policy": "Fund at most one request per step; prefer higher urgency among affordable.", "budget_remaining": 133, "requests": [{"id": "docs_rewrite", "title": "Docs rewrite contractor", "cost": 35, "urgency": 0}, {"id": "security_patch", "title": "Critical security patch", "cost": 37, "urgency": 3}, {"id": "compliance_...
Would summing all costs fit under budget_remaining? Answer with one letter.
[ "A", "B" ]
[ "yes: Budget covers the sum of all request costs", "no: Budget is insufficient for all requests" ]
B
no
{"gym": "resource_allocator", "schema_hint": "noul", "schema": "choice", "hard": false, "n_options": 2}
ticket.needs_human
{"ticket": {"subject": "Charged twice for the same invoice A-340", "body": "Message: I was charged twice and want a refund immediately. Priority flag: p1. Account tier: enterprise.", "tier": "enterprise", "priority": "p1", "wait_days": 0}, "channel": "chat"}
Can a bot handle this safely, or must a person intervene? One letter.
[ "A", "B" ]
[ "no: Safe for bot / routine", "yes: Needs a human (money, legal, anger, long wait, outage)" ]
A
no
{"gym": "ticket_dungeon", "schema_hint": "noul", "schema": "choice", "hard": false, "n_options": 2}
alloc.pressure
{"currency": "USD", "budget_remaining": 112, "requests": [{"id": "ml_experiment", "title": "GPU hours for ML experiment", "cost": 96, "urgency": 1}, {"id": "customer_refund_pool", "title": "Customer refund reserve", "cost": 67, "urgency": 2}, {"id": "oncall_stipend", "title": "On-call stipend top-up", "cost": 39, "urge...
Rate overall budget pressure. Reply with one level letter.
[ "A", "B", "C" ]
[ "0: low — budget comfortably covers demand", "2: high — demand exceeds budget", "1: medium — close to capacity" ]
B
2
{"gym": "resource_allocator", "schema_hint": "score", "ratio": 1.8035714285714286, "schema": "choice", "hard": false, "n_options": 3}
debate.winner
{"side_a": {"label": "support", "claims": [{"text": "Attrition dropped 20% after remote policy", "hard_evidence": true}, {"text": "My friend likes remote", "hard_evidence": false}, {"text": "Hiring pipeline widened internationally", "hard_evidence": true}]}, "side_b": {"label": "oppose", "claims": [{"text": "New-grad r...
Which side is better supported by evidence? Reply with one option letter.
[ "A", "B" ]
[ "oppose: Side B — oppose the proposal", "support: Side A — support the proposal" ]
B
support
{"gym": "debate_judge", "pro_ev": 2, "con_ev": 1, "schema": "choice", "hard": false, "n_options": 2}
ticket.route
{"channel": "chat", "ticket": {"subject": "Compare Business vs Enterprise plans", "body": "Customer says: Can sales walk us through annual pricing?. Wait time so far: 1 days. Tier: enterprise.", "tier": "enterprise", "priority": "p1", "wait_days": 1}}
Select the correct routing destination for the ticket state.
[ "A", "B", "C", "D", "E" ]
[ "billing: payment, charges, invoices, refunds", "shipping: delivery, packages, tracking", "technical: bugs, crashes, outages, integrations", "other: unclear or general questions", "sales: pricing, upgrades, new accounts" ]
E
sales
{"gym": "ticket_dungeon", "domain_seed": "sales", "schema": "choice", "hard": false, "n_options": 5}
debate.confidence
{"side_b": {"label": "oppose", "claims": [{"text": "Rewrite estimated at 2 quarters with feature freeze", "hard_evidence": true}]}, "side_a": {"label": "support", "claims": [{"text": "Rust is trendy on HN", "hard_evidence": false}]}, "rubric": "Prefer hard evidence over vibes; break ties toward the side with more hard ...
Confidence score for this adjudication — pick one Criteria letter.
[ "A", "B", "C" ]
[ "0: low confidence", "1: medium confidence", "2: high confidence" ]
A
0
{"gym": "debate_judge", "schema_hint": "score", "schema": "choice", "hard": false, "n_options": 3}
alloc.fund_next
{"currency": "USD", "policy": "Fund at most one request per step; prefer higher urgency among affordable.", "requests": [{"id": "compliance_audit", "title": "Compliance audit support", "cost": 79, "urgency": 3}, {"id": "ml_experiment", "title": "GPU hours for ML experiment", "cost": 82, "urgency": 1}, {"id": "hiring_bo...
Given budget_remaining and urgencies, which request wins? One letter.
[ "A", "B", "C" ]
[ "hiring_bonus: Hiring referral bonuses (cost=55, urgency=1)", "compliance_audit: Compliance audit support (cost=79, urgency=3)", "ml_experiment: GPU hours for ML experiment (cost=82, urgency=1)" ]
B
compliance_audit
{"gym": "resource_allocator", "schema": "choice", "hard": false, "n_options": 3}
debate.confidence
{"side_b": {"label": "oppose", "claims": [{"text": "Juniors ask too many questions", "hard_evidence": false}, {"text": "Ramp cost exceeds budget this half", "hard_evidence": true}, {"text": "Critical path needs senior ownership now", "hard_evidence": true}]}, "side_a": {"label": "support", "claims": [{"text": "Juniors ...
Rate decision confidence from the evidence balance. One letter.
[ "A", "B", "C" ]
[ "0: low confidence", "2: high confidence", "1: medium confidence" ]
A
0
{"gym": "debate_judge", "schema_hint": "score", "schema": "choice", "hard": false, "n_options": 3}
alloc.fund_next
{"policy": "Fund at most one request per step; prefer higher urgency among affordable.", "currency": "USD", "requests": [{"id": "oncall_stipend", "title": "On-call stipend top-up", "cost": 35, "urgency": 2}, {"id": "hiring_bonus", "title": "Hiring referral bonuses", "cost": 49, "urgency": 1}, {"id": "cdn_upgrade", "tit...
Given budget_remaining and urgencies, which request wins? One letter.
[ "A", "B", "C", "D", "E" ]
[ "hiring_bonus: Hiring referral bonuses (cost=49, urgency=1)", "marketing_campaign: Q3 marketing campaign (cost=112, urgency=1)", "oncall_stipend: On-call stipend top-up (cost=35, urgency=2)", "cdn_upgrade: CDN capacity upgrade (cost=64, urgency=2)", "ml_experiment: GPU hours for ML experiment (cost=86, urge...
C
oncall_stipend
{"gym": "resource_allocator", "schema": "choice", "hard": false, "n_options": 5}
alloc.can_fund_all
{"requests": [{"id": "customer_refund_pool", "title": "Customer refund reserve", "cost": 51, "urgency": 2}, {"id": "laptop_refresh", "title": "Engineer laptop refresh", "cost": 77, "urgency": 1}, {"id": "design_sprint", "title": "Design sprint facilitator", "cost": 26, "urgency": 0}, {"id": "docs_rewrite", "title": "Do...
Would summing all costs fit under budget_remaining? Answer with one letter.
[ "A", "B" ]
[ "yes: Budget covers the sum of all request costs", "no: Budget is insufficient for all requests" ]
A
yes
{"gym": "resource_allocator", "schema_hint": "noul", "schema": "choice", "hard": false, "n_options": 2}
debate.confidence
{"rubric": "Prefer hard evidence over vibes; break ties toward the side with more hard evidence.", "question": "We should deploy on Friday", "side_b": {"label": "oppose", "claims": [{"text": "Last Friday deploy caused a Sev-1", "hard_evidence": true}]}, "side_a": {"label": "support", "claims": [{"text": "Customers aske...
Confidence score for this adjudication — pick one Criteria letter.
[ "A", "B", "C" ]
[ "0: low confidence", "1: medium confidence", "2: high confidence" ]
A
0
{"gym": "debate_judge", "schema_hint": "score", "schema": "choice", "hard": false, "n_options": 3}
debate.winner
{"rubric": "Prefer hard evidence over vibes; break ties toward the side with more hard evidence.", "question": "Remote-first improves productivity", "side_b": {"label": "oppose", "claims": [{"text": "Offices look empty and sad", "hard_evidence": false}, {"text": "New-grad ramp surveys show slower onboarding", "hard_evi...
Who wins under the stated evidence policy? One Criteria letter.
[ "A", "B" ]
[ "support: Side A — support the proposal", "oppose: Side B — oppose the proposal" ]
B
oppose
{"gym": "debate_judge", "pro_ev": 1, "con_ev": 2, "schema": "choice", "hard": false, "n_options": 2}
ticket.needs_human
{"channel": "chat", "ticket": {"subject": "Request a demo and pricing quote", "body": "We need a quote for upgrading seats They have waited 8 days. Segment=enterprise.", "tier": "enterprise", "priority": "p0", "wait_days": 8}}
Human escalation required? Pick yes or no via the option letter.
[ "A", "B" ]
[ "no: Safe for bot / routine", "yes: Needs a human (money, legal, anger, long wait, outage)" ]
B
yes
{"gym": "ticket_dungeon", "schema_hint": "noul", "schema": "choice", "hard": false, "n_options": 2}
alloc.fund_next
{"requests": [{"id": "marketing_campaign", "title": "Q3 marketing campaign", "cost": 121, "urgency": 1}, {"id": "laptop_refresh", "title": "Engineer laptop refresh", "cost": 75, "urgency": 1}, {"id": "security_patch", "title": "Critical security patch", "cost": 42, "urgency": 3}], "policy": "Fund at most one request pe...
Select the next funding target under the policy. One option letter.
[ "A", "B", "C" ]
[ "laptop_refresh: Engineer laptop refresh (cost=75, urgency=1)", "marketing_campaign: Q3 marketing campaign (cost=121, urgency=1)", "security_patch: Critical security patch (cost=42, urgency=3)" ]
C
security_patch
{"gym": "resource_allocator", "schema": "choice", "hard": false, "n_options": 3}
alloc.can_fund_all
{"requests": [{"id": "marketing_campaign", "title": "Q3 marketing campaign", "cost": 125, "urgency": 1}, {"id": "laptop_refresh", "title": "Engineer laptop refresh", "cost": 85, "urgency": 1}, {"id": "compliance_audit", "title": "Compliance audit support", "cost": 67, "urgency": 3}, {"id": "hiring_bonus", "title": "Hir...
Would summing all costs fit under budget_remaining? Answer with one letter.
[ "A", "B" ]
[ "no: Budget is insufficient for all requests", "yes: Budget covers the sum of all request costs" ]
A
no
{"gym": "resource_allocator", "schema_hint": "noul", "schema": "choice", "hard": false, "n_options": 2}
ticket.urgency
{"ticket": {"subject": "Where is the documentation?", "body": "Message: Thanks for the help last week. Priority flag: p2. Account tier: pro.", "tier": "pro", "priority": "p2", "wait_days": 7}, "channel": "email"}
How urgent is this ticket? Pick one level letter from Criteria.
[ "A", "B", "C" ]
[ "1: medium — same day", "0: low — can wait", "2: high — immediate" ]
C
2
{"gym": "ticket_dungeon", "schema_hint": "score", "urgency": 2, "schema": "choice", "hard": false, "n_options": 3}
alloc.pressure
{"requests": [{"id": "docs_rewrite", "title": "Docs rewrite contractor", "cost": 39, "urgency": 0}, {"id": "office_snacks", "title": "Office snacks", "cost": 13, "urgency": 0}, {"id": "marketing_campaign", "title": "Q3 marketing campaign", "cost": 110, "urgency": 1}, {"id": "hiring_bonus", "title": "Hiring referral bon...
How constrained is the budget relative to demand? Pick one letter.
[ "A", "B", "C" ]
[ "0: low — budget comfortably covers demand", "1: medium — close to capacity", "2: high — demand exceeds budget" ]
B
1
{"gym": "resource_allocator", "schema_hint": "score", "ratio": 1.1044776119402986, "schema": "choice", "hard": false, "n_options": 3}
ticket.route
{"ticket": {"subject": "Where is my shipment A-525?", "body": "Customer says: Package marked delivered but nothing arrived. Wait time so far: 2 days. Tier: free.", "tier": "free", "priority": "p3", "wait_days": 2}, "channel": "email"}
Best team for this case? Answer with a single option letter.
[ "A", "B", "C", "D", "E" ]
[ "shipping: delivery, packages, tracking", "sales: pricing, upgrades, new accounts", "other: unclear or general questions", "technical: bugs, crashes, outages, integrations", "billing: payment, charges, invoices, refunds" ]
A
shipping
{"gym": "ticket_dungeon", "domain_seed": "shipping", "schema": "choice", "hard": false, "n_options": 5}
alloc.can_fund_all
{"currency": "USD", "budget_remaining": 241, "requests": [{"id": "design_sprint", "title": "Design sprint facilitator", "cost": 20, "urgency": 0}, {"id": "compliance_audit", "title": "Compliance audit support", "cost": 61, "urgency": 3}, {"id": "security_patch", "title": "Critical security patch", "cost": 49, "urgency"...
Would summing all costs fit under budget_remaining? Answer with one letter.
[ "A", "B" ]
[ "no: Budget is insufficient for all requests", "yes: Budget covers the sum of all request costs" ]
B
yes
{"gym": "resource_allocator", "schema_hint": "noul", "schema": "choice", "hard": false, "n_options": 2}
ticket.route
{"channel": "chat", "ticket": {"subject": "Compare Business vs Enterprise plans", "body": "Thread summary — We need a quote for upgrading seats | wait_days=2 | tier=enterprise | priority=p3", "tier": "enterprise", "priority": "p3", "wait_days": 2}}
Choose the owning queue from Criteria. One letter only.
[ "A", "B", "C", "D", "E" ]
[ "other: unclear or general questions", "sales: pricing, upgrades, new accounts", "billing: payment, charges, invoices, refunds", "shipping: delivery, packages, tracking", "technical: bugs, crashes, outages, integrations" ]
B
sales
{"gym": "ticket_dungeon", "domain_seed": "sales", "schema": "choice", "hard": false, "n_options": 5}
debate.winner
{"side_a": {"label": "support", "claims": [{"text": "Competitors already priced 12% higher", "hard_evidence": true}, {"text": "Gross margin is below target", "hard_evidence": true}, {"text": "Prices going up sounds cool", "hard_evidence": false}]}, "side_b": {"label": "oppose", "claims": [{"text": "Customers will be ma...
Given the claims and evidence flags, who should win the debate? One letter.
[ "A", "B" ]
[ "support: Side A — support the proposal", "oppose: Side B — oppose the proposal" ]
A
support
{"gym": "debate_judge", "pro_ev": 2, "con_ev": 1, "schema": "choice", "hard": false, "n_options": 2}
ticket.route
{"channel": "phone", "ticket": {"subject": "Compare Business vs Enterprise plans", "body": "Thread summary — Interested in enterprise pricing and a demo | wait_days=2 | tier=pro | priority=p1", "tier": "pro", "priority": "p1", "wait_days": 2}}
Select the correct routing destination for the ticket state.
[ "A", "B", "C", "D", "E" ]
[ "shipping: delivery, packages, tracking", "sales: pricing, upgrades, new accounts", "other: unclear or general questions", "technical: bugs, crashes, outages, integrations", "billing: payment, charges, invoices, refunds" ]
B
sales
{"gym": "ticket_dungeon", "domain_seed": "sales", "schema": "choice", "hard": false, "n_options": 5}
ticket.needs_human
{"channel": "email", "ticket": {"subject": "Cannot log in after password reset", "body": "Customer says: The app crashes every time I open settings. Wait time so far: 0 days. Tier: pro.", "tier": "pro", "priority": "p3", "wait_days": 0}}
Should a human review this before automated handling? Reply with one letter.
[ "A", "B" ]
[ "no: Safe for bot / routine", "yes: Needs a human (money, legal, anger, long wait, outage)" ]
A
no
{"gym": "ticket_dungeon", "schema_hint": "noul", "schema": "choice", "hard": false, "n_options": 2}
debate.enough_evidence
{"side_a": {"label": "support", "claims": [{"text": "Inbound PRs already fix bugs we lack time for", "hard_evidence": true}, {"text": "Open source is always good actually", "hard_evidence": false}]}, "rubric": "Prefer hard evidence over vibes; break ties toward the side with more hard evidence.", "side_b": {"label": "o...
Do we have sufficient evidence-backed claims (not vibes)? Reply with one letter.
[ "A", "B" ]
[ "yes: Enough hard evidence to decide", "no: Too thin or tied — needs more evidence" ]
A
yes
{"gym": "debate_judge", "schema_hint": "noul", "schema": "choice", "hard": false, "n_options": 2}
alloc.can_fund_all
{"requests": [{"id": "marketing_campaign", "title": "Q3 marketing campaign", "cost": 126, "urgency": 1}, {"id": "design_sprint", "title": "Design sprint facilitator", "cost": 19, "urgency": 0}, {"id": "hiring_bonus", "title": "Hiring referral bonuses", "cost": 44, "urgency": 1}], "currency": "USD", "budget_remaining": ...
Can we fund ALL remaining requests within the budget? Reply with one letter.
[ "A", "B" ]
[ "yes: Budget covers the sum of all request costs", "no: Budget is insufficient for all requests" ]
A
yes
{"gym": "resource_allocator", "schema_hint": "noul", "schema": "choice", "hard": false, "n_options": 2}
alloc.fund_next
{"budget_remaining": 96, "currency": "USD", "policy": "Fund at most one request per step; prefer higher urgency among affordable.", "requests": [{"id": "oncall_stipend", "title": "On-call stipend top-up", "cost": 38, "urgency": 2}, {"id": "compliance_audit", "title": "Compliance audit support", "cost": 70, "urgency": 3...
Pick the best request to fund now. Prefer high urgency that fits the remaining budget.
[ "A", "B", "C", "D", "E" ]
[ "compliance_audit: Compliance audit support (cost=70, urgency=3)", "oncall_stipend: On-call stipend top-up (cost=38, urgency=2)", "security_patch: Critical security patch (cost=45, urgency=3)", "ml_experiment: GPU hours for ML experiment (cost=94, urgency=1)", "laptop_refresh: Engineer laptop refresh (cost=...
C
security_patch
{"gym": "resource_allocator", "schema": "choice", "hard": false, "n_options": 5}
debate.enough_evidence
{"side_a": {"label": "support", "claims": [{"text": "Prices going up sounds cool", "hard_evidence": false}]}, "question": "We should raise prices 10%", "rubric": "Prefer hard evidence over vibes; break ties toward the side with more hard evidence.", "side_b": {"label": "oppose", "claims": [{"text": "Customers will be m...
Evidence sufficiency check — yes or no via option letter.
[ "A", "B" ]
[ "yes: Enough hard evidence to decide", "no: Too thin or tied — needs more evidence" ]
B
no
{"gym": "debate_judge", "schema_hint": "noul", "schema": "choice", "hard": false, "n_options": 2}
debate.winner
{"side_b": {"label": "oppose", "claims": [{"text": "Enterprise deals are price-sensitive this quarter", "hard_evidence": true}, {"text": "Customers will be mad forever", "hard_evidence": false}, {"text": "Churn spiked after last increase", "hard_evidence": true}]}, "rubric": "Prefer hard evidence over vibes; break ties...
Apply the rubric: hard evidence beats vibes. Pick the winning side letter.
[ "A", "B" ]
[ "oppose: Side B — oppose the proposal", "support: Side A — support the proposal" ]
A
oppose
{"gym": "debate_judge", "pro_ev": 2, "con_ev": 2, "schema": "choice", "hard": false, "n_options": 2}
debate.winner
{"rubric": "Prefer hard evidence over vibes; break ties toward the side with more hard evidence.", "side_b": {"label": "oppose", "claims": [{"text": "Churn spiked after last increase", "hard_evidence": true}]}, "side_a": {"label": "support", "claims": [{"text": "Gross margin is below target", "hard_evidence": true}, {"...
Given the claims and evidence flags, who should win the debate? One letter.
[ "A", "B" ]
[ "oppose: Side B — oppose the proposal", "support: Side A — support the proposal" ]
B
support
{"gym": "debate_judge", "pro_ev": 1, "con_ev": 1, "schema": "choice", "hard": false, "n_options": 2}
debate.confidence
{"question": "We should open-source the core library", "rubric": "Prefer hard evidence over vibes; break ties toward the side with more hard evidence.", "side_b": {"label": "oppose", "claims": [{"text": "Core contains proprietary scoring IP", "hard_evidence": true}]}, "side_a": {"label": "support", "claims": [{"text": ...
How confident should we be in the judgment? Reply with one level letter.
[ "A", "B" ]
[ "2: high confidence", "0: low confidence" ]
B
0
{"gym": "debate_judge", "schema_hint": "score", "schema": "choice", "hard": false, "n_options": 2}
alloc.pressure
{"budget_remaining": 238, "requests": [{"id": "marketing_campaign", "title": "Q3 marketing campaign", "cost": 127, "urgency": 1}, {"id": "security_patch", "title": "Critical security patch", "cost": 47, "urgency": 3}, {"id": "office_snacks", "title": "Office snacks", "cost": 20, "urgency": 0}, {"id": "laptop_refresh", ...
Budget tightness score — choose one Criteria letter.
[ "A", "B", "C" ]
[ "2: high — demand exceeds budget", "1: medium — close to capacity", "0: low — budget comfortably covers demand" ]
B
1
{"gym": "resource_allocator", "schema_hint": "score", "ratio": 1.138655462184874, "schema": "choice", "hard": false, "n_options": 3}
debate.enough_evidence
{"side_a": {"label": "support", "claims": [{"text": "p99 latency is 4x SLO on hot path", "hard_evidence": true}]}, "question": "We should rewrite the backend in Rust", "side_b": {"label": "oppose", "claims": [{"text": "I dream in garbage collectors", "hard_evidence": false}]}, "rubric": "Prefer hard evidence over vibes...
Do we have sufficient evidence-backed claims (not vibes)? Reply with one letter.
[ "A", "B" ]
[ "no: Too thin or tied — needs more evidence", "yes: Enough hard evidence to decide" ]
A
no
{"gym": "debate_judge", "schema_hint": "noul", "schema": "choice", "hard": false, "n_options": 2}
ticket.route
{"channel": "email", "ticket": {"subject": "Charged twice for the same invoice A-632", "body": "Customer says: Invoice shows wrong amount on my card. Wait time so far: 10 days. Tier: pro.", "tier": "pro", "priority": "p0", "wait_days": 10}}
Select the correct routing destination for the ticket state.
[ "A", "B", "C", "D", "E" ]
[ "billing: payment, charges, invoices, refunds", "other: unclear or general questions", "sales: pricing, upgrades, new accounts", "shipping: delivery, packages, tracking", "technical: bugs, crashes, outages, integrations" ]
A
billing
{"gym": "ticket_dungeon", "domain_seed": "billing", "schema": "choice", "hard": false, "n_options": 5}
alloc.pressure
{"budget_remaining": 309, "requests": [{"id": "security_patch", "title": "Critical security patch", "cost": 44, "urgency": 3}, {"id": "marketing_campaign", "title": "Q3 marketing campaign", "cost": 126, "urgency": 1}, {"id": "design_sprint", "title": "Design sprint facilitator", "cost": 15, "urgency": 0}, {"id": "compl...
How constrained is the budget relative to demand? Pick one letter.
[ "A", "B", "C" ]
[ "0: low — budget comfortably covers demand", "2: high — demand exceeds budget", "1: medium — close to capacity" ]
C
1
{"gym": "resource_allocator", "schema_hint": "score", "ratio": 0.8284789644012945, "schema": "choice", "hard": false, "n_options": 3}
alloc.fund_next
{"policy": "Fund at most one request per step; prefer higher urgency among affordable.", "currency": "USD", "budget_remaining": 161, "requests": [{"id": "hiring_bonus", "title": "Hiring referral bonuses", "cost": 51, "urgency": 1}, {"id": "security_patch", "title": "Critical security patch", "cost": 43, "urgency": 3}, ...
Pick the best request to fund now. Prefer high urgency that fits the remaining budget.
[ "A", "B", "C" ]
[ "oncall_stipend: On-call stipend top-up (cost=52, urgency=2)", "hiring_bonus: Hiring referral bonuses (cost=51, urgency=1)", "security_patch: Critical security patch (cost=43, urgency=3)" ]
C
security_patch
{"gym": "resource_allocator", "schema": "choice", "hard": false, "n_options": 3}
debate.enough_evidence
{"question": "We should hire more junior engineers", "side_a": {"label": "support", "claims": [{"text": "Senior mentors have spare capacity", "hard_evidence": true}, {"text": "Juniors are cheaper vibes", "hard_evidence": false}]}, "side_b": {"label": "oppose", "claims": [{"text": "Critical path needs senior ownership n...
Evidence sufficiency check — yes or no via option letter.
[ "A", "B" ]
[ "no: Too thin or tied — needs more evidence", "yes: Enough hard evidence to decide" ]
A
no
{"gym": "debate_judge", "schema_hint": "noul", "schema": "choice", "hard": false, "n_options": 2}
alloc.pressure
{"policy": "Fund at most one request per step; prefer higher urgency among affordable.", "currency": "USD", "budget_remaining": 241, "requests": [{"id": "ml_experiment", "title": "GPU hours for ML experiment", "cost": 89, "urgency": 1}, {"id": "compliance_audit", "title": "Compliance audit support", "cost": 69, "urgenc...
Rate overall budget pressure. Reply with one level letter.
[ "A", "B", "C" ]
[ "0: low — budget comfortably covers demand", "2: high — demand exceeds budget", "1: medium — close to capacity" ]
C
1
{"gym": "resource_allocator", "schema_hint": "score", "ratio": 1.0, "schema": "choice", "hard": false, "n_options": 3}
alloc.pressure
{"requests": [{"id": "cdn_upgrade", "title": "CDN capacity upgrade", "cost": 45, "urgency": 2}, {"id": "hiring_bonus", "title": "Hiring referral bonuses", "cost": 55, "urgency": 1}, {"id": "customer_refund_pool", "title": "Customer refund reserve", "cost": 61, "urgency": 2}, {"id": "ml_experiment", "title": "GPU hours ...
How constrained is the budget relative to demand? Pick one letter.
[ "A", "B", "C" ]
[ "1: medium — close to capacity", "0: low — budget comfortably covers demand", "2: high — demand exceeds budget" ]
C
2
{"gym": "resource_allocator", "schema_hint": "score", "ratio": 1.2757009345794392, "schema": "choice", "hard": false, "n_options": 3}
debate.confidence
{"side_a": {"label": "support", "claims": [{"text": "Design partners ask for source escrow", "hard_evidence": true}, {"text": "Open source is always good actually", "hard_evidence": false}]}, "question": "We should open-source the core library", "side_b": {"label": "oppose", "claims": [{"text": "Core contains proprieta...
Rate decision confidence from the evidence balance. One letter.
[ "A", "B" ]
[ "0: low confidence", "1: medium confidence" ]
A
0
{"gym": "debate_judge", "schema_hint": "score", "schema": "choice", "hard": false, "n_options": 2}
ticket.route
{"channel": "chat", "ticket": {"subject": "Where is my shipment A-735?", "body": "Thread summary — Courier lost my shipment | wait_days=2 | tier=free | priority=p1", "tier": "free", "priority": "p1", "wait_days": 2}}
Choose the owning queue from Criteria. One letter only.
[ "A", "B", "C", "D", "E" ]
[ "technical: bugs, crashes, outages, integrations", "other: unclear or general questions", "billing: payment, charges, invoices, refunds", "sales: pricing, upgrades, new accounts", "shipping: delivery, packages, tracking" ]
E
shipping
{"gym": "ticket_dungeon", "domain_seed": "shipping", "schema": "choice", "hard": false, "n_options": 5}
ticket.urgency
{"ticket": {"subject": "Delivery marked complete but missing", "body": "Message: Delivery ETA keeps slipping. Priority flag: p3. Account tier: pro.", "tier": "pro", "priority": "p3", "wait_days": 6}, "channel": "phone"}
Urgency classification — choose exactly one Criteria letter.
[ "A", "B", "C" ]
[ "2: high — immediate", "0: low — can wait", "1: medium — same day" ]
A
2
{"gym": "ticket_dungeon", "schema_hint": "score", "urgency": 2, "schema": "choice", "hard": false, "n_options": 3}
alloc.fund_next
{"budget_remaining": 292, "requests": [{"id": "oncall_stipend", "title": "On-call stipend top-up", "cost": 42, "urgency": 2}, {"id": "design_sprint", "title": "Design sprint facilitator", "cost": 26, "urgency": 0}, {"id": "ml_experiment", "title": "GPU hours for ML experiment", "cost": 99, "urgency": 1}, {"id": "cdn_up...
Which single request should be funded next given the budget? Reply with one letter.
[ "A", "B", "C", "D", "E" ]
[ "ml_experiment: GPU hours for ML experiment (cost=99, urgency=1)", "oncall_stipend: On-call stipend top-up (cost=42, urgency=2)", "cdn_upgrade: CDN capacity upgrade (cost=61, urgency=2)", "laptop_refresh: Engineer laptop refresh (cost=73, urgency=1)", "design_sprint: Design sprint facilitator (cost=26, urge...
B
oncall_stipend
{"gym": "resource_allocator", "schema": "choice", "hard": false, "n_options": 5}
ticket.urgency
{"ticket": {"subject": "Refund not received for cancelled sub", "body": "Thread summary — Billing portal shows an incorrect charge | wait_days=1 | tier=enterprise | priority=p0", "tier": "enterprise", "priority": "p0", "wait_days": 1}, "channel": "phone"}
How urgent is this ticket? Pick one level letter from Criteria.
[ "A", "B", "C" ]
[ "2: high — immediate", "0: low — can wait", "1: medium — same day" ]
A
2
{"gym": "ticket_dungeon", "schema_hint": "score", "urgency": 2, "schema": "choice", "hard": false, "n_options": 3}
alloc.pressure
{"requests": [{"id": "office_snacks", "title": "Office snacks", "cost": 22, "urgency": 0}, {"id": "design_sprint", "title": "Design sprint facilitator", "cost": 35, "urgency": 0}, {"id": "docs_rewrite", "title": "Docs rewrite contractor", "cost": 36, "urgency": 0}, {"id": "cdn_upgrade", "title": "CDN capacity upgrade",...
Budget tightness score — choose one Criteria letter.
[ "A", "B", "C" ]
[ "0: low — budget comfortably covers demand", "2: high — demand exceeds budget", "1: medium — close to capacity" ]
C
1
{"gym": "resource_allocator", "schema_hint": "score", "ratio": 0.9407407407407408, "schema": "choice", "hard": false, "n_options": 3}
ticket.route
{"ticket": {"subject": "Delivery marked complete but missing", "body": "Thread summary — Package marked delivered but nothing arrived | wait_days=4 | tier=free | priority=p2", "tier": "free", "priority": "p2", "wait_days": 4}, "channel": "phone"}
Choose the owning queue from Criteria. One letter only.
[ "A", "B", "C", "D", "E" ]
[ "shipping: delivery, packages, tracking", "other: unclear or general questions", "sales: pricing, upgrades, new accounts", "technical: bugs, crashes, outages, integrations", "billing: payment, charges, invoices, refunds" ]
A
shipping
{"gym": "ticket_dungeon", "domain_seed": "shipping", "schema": "choice", "hard": false, "n_options": 5}
alloc.fund_next
{"currency": "USD", "policy": "Fund at most one request per step; prefer higher urgency among affordable.", "budget_remaining": 206, "requests": [{"id": "docs_rewrite", "title": "Docs rewrite contractor", "cost": 33, "urgency": 0}, {"id": "ml_experiment", "title": "GPU hours for ML experiment", "cost": 97, "urgency": 1...
Given budget_remaining and urgencies, which request wins? One letter.
[ "A", "B", "C", "D" ]
[ "ml_experiment: GPU hours for ML experiment (cost=97, urgency=1)", "compliance_audit: Compliance audit support (cost=76, urgency=3)", "cdn_upgrade: CDN capacity upgrade (cost=45, urgency=2)", "docs_rewrite: Docs rewrite contractor (cost=33, urgency=0)" ]
B
compliance_audit
{"gym": "resource_allocator", "schema": "choice", "hard": false, "n_options": 4}
debate.enough_evidence
{"side_a": {"label": "support", "claims": [{"text": "Design partners ask for source escrow", "hard_evidence": true}]}, "side_b": {"label": "oppose", "claims": [{"text": "Core contains proprietary scoring IP", "hard_evidence": true}]}, "question": "We should open-source the core library", "rubric": "Prefer hard evidence...
Is there enough hard evidence to decide confidently? Reply with one letter.
[ "A", "B" ]
[ "no: Too thin or tied — needs more evidence", "yes: Enough hard evidence to decide" ]
A
no
{"gym": "debate_judge", "schema_hint": "noul", "schema": "choice", "hard": false, "n_options": 2}
alloc.can_fund_all
{"policy": "Fund at most one request per step; prefer higher urgency among affordable.", "requests": [{"id": "marketing_campaign", "title": "Q3 marketing campaign", "cost": 110, "urgency": 1}, {"id": "ml_experiment", "title": "GPU hours for ML experiment", "cost": 89, "urgency": 1}, {"id": "laptop_refresh", "title": "E...
Is the budget enough to approve every request? Reply with one option letter.
[ "A", "B" ]
[ "no: Budget is insufficient for all requests", "yes: Budget covers the sum of all request costs" ]
B
yes
{"gym": "resource_allocator", "schema_hint": "noul", "schema": "choice", "hard": false, "n_options": 2}
debate.enough_evidence
{"rubric": "Prefer hard evidence over vibes; break ties toward the side with more hard evidence.", "question": "We should rewrite the backend in Rust", "side_a": {"label": "support", "claims": [{"text": "Rust prototype cut CPU 35% in bench", "hard_evidence": true}, {"text": "p99 latency is 4x SLO on hot path", "hard_ev...
Do we have sufficient evidence-backed claims (not vibes)? Reply with one letter.
[ "A", "B" ]
[ "yes: Enough hard evidence to decide", "no: Too thin or tied — needs more evidence" ]
A
yes
{"gym": "debate_judge", "schema_hint": "noul", "schema": "choice", "hard": false, "n_options": 2}