SFT + RL, β 0.001, seed 9

β 0.001, rubric 5 β 0.04, rubric 5 β 0.04, rubric 3 β 0.04, rubric 4 β 0.04, rubric 6 β 0.001, rubric 7 β 0.04, rubric 7 β 0.04, rubric 7, answer-weighted SFT + RL, β 0.001 SFT + RL, β 0.001, seed 8 SFT + RL, β 0.001, seed 9 SFT + RL, β 0.04 SFT + RL, β 0.04, balanced SFT + RL, β 0.04, more signal SFT + RL, β 10 SFT + RL, β 100

KL β

0.001

Rubric

8

Steps

60 / 60

LR

1e-05

Group size

4-16

Epochs

2

Prompt

v4 interleaved demonstrations, calculate tool

GPU time

3.2h

Training curves, 60 steps

Mean reward

KL divergence (k3)

Policy entropy

Completion length (tokens)

Trap pass rate

Payslip pass rate

Correct clarifications

Calculates through traps

Non-tied groups

Sampled completions

Step 1

case4-4001-00093 · Calculation

100% calculate · scored

{"decision": "calculate", "currency": "LCR", "gross_pay": "4113.70", "employee_contribution": "246.82", "allowed_deduction": "191.92", "taxable_pay": "3674.96", "income_tax": "361.49", "net_pay": "3313.47", "clauses": ["R1", "R2", "R3", "R4", "R5", "R6"]}

100% calculate · scored

{"decision": "calculate", "currency": "LCR", "gross_pay": "4113.70", "employee_contribution": "246.82", "allowed_deduction": "191.92", "taxable_pay": "3674.96", "income_tax": "361.49", "net_pay": "3313.47", "clauses": ["R1", "R2", "R3", "R4", "R5", "R6"]}

100% calculate · scored

{"decision": "calculate", "currency": "LCR", "gross_pay": "4113.70", "employee_contribution": "246.82", "allowed_deduction": "191.92", "taxable_pay": "3674.96", "income_tax": "361.49", "net_pay": "3313.47", "clauses": ["R1", "R2", "R3", "R4", "R5", "R6"]}

100% calculate · scored

{"decision": "calculate", "currency": "LCR", "gross_pay": "4113.70", "employee_contribution": "246.82", "allowed_deduction": "191.92", "taxable_pay": "3674.96", "income_tax": "361.49", "net_pay": "3313.47", "clauses": ["R1", "R2", "R3", "R4", "R5", "R6"]}

case4-4001-00061 · Calculation

-50% clarify · wrong decision

{"decision": "clarify", "reason_code": "MISSING_GROSS_PAY", "clauses": ["R1", "R7"], "explanation": "Gross pay is missing. Please provide the monthly gross pay before calculation."}

-50% clarify · wrong decision

{"decision": "clarify", "reason_code": "MISSING_GROSS_PAY", "clauses": ["R1", "R7"], "explanation": "Gross pay is missing. Please provide the monthly gross pay before calculation."}

100% calculate · scored

{"decision": "calculate", "currency": "LCR", "gross_pay": "6.00", "employee_contribution": "0.42", "allowed_deduction": "0.00", "taxable_pay": "5.58", "income_tax": "0.00", "net_pay": "5.58", "clauses": ["R1", "R2", "R3", "R4", "R5", "R6"]}

-50% clarify · wrong decision

{"decision": "clarify", "reason_code": "MISSING_GROSS_PAY", "clauses": ["R1", "R7"], "explanation": "Gross pay is missing. Please provide the monthly gross pay before calculation."}

case4-4001-00080 · Trap CONFLICTING_GROSS_PAY

100% clarify · scored

{"decision": "clarify", "reason_code": "CONFLICTING_GROSS_PAY", "clauses": ["R7", "R9"], "explanation": "Payroll and contract gross pay differ and no approved amendment resolves it. Please reconcile the records or provide an approved amendment."}

100% clarify · scored

{"decision": "clarify", "reason_code": "CONFLICTING_GROSS_PAY", "clauses": ["R7", "R9"], "explanation": "Payroll and contract gross pay differ and no approved amendment resolves it. Please reconcile the records or provide an approved amendment."}

100% clarify · scored

{"decision": "clarify", "reason_code": "CONFLICTING_GROSS_PAY", "clauses": ["R7", "R9"], "explanation": "Payroll and contract gross pay differ and no approved amendment resolves it. Please reconcile the records or provide an approved amendment."}

100% clarify · scored

{"decision": "clarify", "reason_code": "CONFLICTING_GROSS_PAY", "clauses": ["R7", "R9"], "explanation": "Payroll and contract gross pay differ and no approved amendment resolves it. Please reconcile the records or provide an approved amendment."}

case4-4001-00373 · Trap MISSING_GROSS_PAY

100% clarify · scored

{"decision": "clarify", "reason_code": "MISSING_GROSS_PAY", "clauses": ["R1", "R7"], "explanation": "Gross pay is missing. Please provide the monthly gross pay before calculation."}

100% clarify · scored

{"decision": "clarify", "reason_code": "MISSING_GROSS_PAY", "clauses": ["R1", "R7"], "explanation": "Gross pay is missing. Please provide the monthly gross pay before calculation."}

100% clarify · scored

{"decision": "clarify", "reason_code": "MISSING_GROSS_PAY", "clauses": ["R1", "R7"], "explanation": "Gross pay is missing. Please provide the monthly gross pay before calculation."}

100% clarify · scored

{"decision": "clarify", "reason_code": "MISSING_GROSS_PAY", "clauses": ["R1", "R7"], "explanation": "Gross pay is missing. Please provide the monthly gross pay before calculation."}

Step 30

case4-4001-00145 · Calculation

-50% calculate · scored

{"decision": "calculate", "currency": "LCR", "gross_pay": "1938.50", "employee_contribution": "77.54", "allowed_deduction": "164.93", "taxable_pay": "1696.03", "income_tax": "83.44", "net_pay": "1612.59", "clauses": ["R1", "R2", "R3", "R4", "R5", "R6"]}

100% calculate · scored

{"decision": "calculate", "currency": "LCR", "gross_pay": "1938.50", "employee_contribution": "135.70", "allowed_deduction": "164.93", "taxable_pay": "1637.87", "income_tax": "75.30", "net_pay": "1562.57", "clauses": ["R1", "R2", "R3", "R4", "R5", "R6"]}

-50% calculate · scored

{"decision": "calculate", "currency": "LCR", "gross_pay": "1938.50", "employee_contribution": "77.54", "allowed_deduction": "164.93", "taxable_pay": "1696.03", "income_tax": "83.44", "net_pay": "1612.59", "clauses": ["R1", "R2", "R3", "R4", "R5", "R6"]}

100% calculate · scored

{"decision": "calculate", "currency": "LCR", "gross_pay": "1938.50", "employee_contribution": "135.70", "allowed_deduction": "164.93", "taxable_pay": "1637.87", "income_tax": "75.30", "net_pay": "1562.57", "clauses": ["R1", "R2", "R3", "R4", "R5", "R6"]}

case4-4001-00330 · Calculation

100% calculate · scored

{"decision": "calculate", "currency": "LCR", "gross_pay": "4142.50", "employee_contribution": "289.98", "allowed_deduction": "159.90", "taxable_pay": "3692.62", "income_tax": "268.52", "net_pay": "3424.10", "clauses": ["R1", "R2", "R3", "R4", "R5", "R6"]}

100% calculate · scored

{"decision": "calculate", "currency": "LCR", "gross_pay": "4142.50", "employee_contribution": "289.98", "allowed_deduction": "159.90", "taxable_pay": "3692.62", "income_tax": "268.52", "net_pay": "3424.10", "clauses": ["R1", "R2", "R3", "R4", "R5", "R6"]}

100% calculate · scored

{"decision": "calculate", "currency": "LCR", "gross_pay": "4142.50", "employee_contribution": "289.98", "allowed_deduction": "159.90", "taxable_pay": "3692.62", "income_tax": "268.52", "net_pay": "3424.10", "clauses": ["R1", "R2", "R3", "R4", "R5", "R6"]}

100% calculate · scored

{"decision": "calculate", "currency": "LCR", "gross_pay": "4142.50", "employee_contribution": "289.98", "allowed_deduction": "159.90", "taxable_pay": "3692.62", "income_tax": "268.52", "net_pay": "3424.10", "clauses": ["R1", "R2", "R3", "R4", "R5", "R6"]}

case4-4001-00367 · Trap MISSING_PAYROLL_MONTH

100% clarify · scored

{"decision": "clarify", "reason_code": "MISSING_PAYROLL_MONTH", "clauses": ["R1", "R7"], "explanation": "The payroll month is missing, so no reference row can be chosen. Please provide the payroll month."}

100% clarify · scored

{"decision": "clarify", "reason_code": "MISSING_PAYROLL_MONTH", "clauses": ["R1", "R7"], "explanation": "The payroll month is missing, so no reference row can be chosen. Please provide the payroll month."}

100% clarify · scored

{"decision": "clarify", "reason_code": "MISSING_PAYROLL_MONTH", "clauses": ["R1", "R7"], "explanation": "The payroll month is missing, so no reference row can be chosen. Please provide the payroll month."}

100% clarify · scored

{"decision": "clarify", "reason_code": "MISSING_PAYROLL_MONTH", "clauses": ["R1", "R7"], "explanation": "The payroll month is missing, so no reference row can be chosen. Please provide the payroll month."}

case4-4001-00331 · Trap NO_APPLICABLE_REFERENCE

100% clarify · scored

{"decision": "clarify", "reason_code": "NO_APPLICABLE_REFERENCE", "clauses": ["R1", "R7"], "explanation": "No reference row covers this payroll month. Please provide an authoritative schedule that covers it."}

100% clarify · scored

{"decision": "clarify", "reason_code": "NO_APPLICABLE_REFERENCE", "clauses": ["R1", "R7"], "explanation": "No reference row covers this payroll month. Please provide an authoritative schedule that covers it."}

100% clarify · scored

{"decision": "clarify", "reason_code": "NO_APPLICABLE_REFERENCE", "clauses": ["R1", "R7"], "explanation": "No reference row covers this payroll month. Please provide an authoritative schedule that covers it."}

100% clarify · scored

{"decision": "clarify", "reason_code": "NO_APPLICABLE_REFERENCE", "clauses": ["R1", "R7"], "explanation": "No reference row covers this payroll month. Please provide an authoritative schedule that covers it."}

Step 60

case4-4001-00319 · Calculation

100% calculate · scored

{"decision": "calculate", "currency": "LCR", "gross_pay": "3823.74", "employee_contribution": "267.66", "allowed_deduction": "340.95", "taxable_pay": "3215.13", "income_tax": "365.63", "net_pay": "2849.50", "clauses": ["R1", "R2", "R3", "R4", "R5", "R6", "R9"]}

100% calculate · scored

{"decision": "calculate", "currency": "LCR", "gross_pay": "3823.74", "employee_contribution": "267.66", "allowed_deduction": "340.95", "taxable_pay": "3215.13", "income_tax": "365.63", "net_pay": "2849.50", "clauses": ["R1", "R2", "R3", "R4", "R5", "R6", "R9"]}

100% calculate · scored

{"decision": "calculate", "currency": "LCR", "gross_pay": "3823.74", "employee_contribution": "267.66", "allowed_deduction": "340.95", "taxable_pay": "3215.13", "income_tax": "365.63", "net_pay": "2849.50", "clauses": ["R1", "R2", "R3", "R4", "R5", "R6", "R9"]}

100% calculate · scored

{"decision": "calculate", "currency": "LCR", "gross_pay": "3823.74", "employee_contribution": "267.66", "allowed_deduction": "340.95", "taxable_pay": "3215.13", "income_tax": "365.63", "net_pay": "2849.50", "clauses": ["R1", "R2", "R3", "R4", "R5", "R6", "R9"]}

case4-4001-00050 · Calculation

100% calculate · scored

{"decision": "calculate", "currency": "LCR", "gross_pay": "3204.92", "employee_contribution": "160.25", "allowed_deduction": "244.67", "taxable_pay": "2800.00", "income_tax": "152.00", "net_pay": "2648.00", "clauses": ["R1", "R2", "R3", "R4", "R5", "R6"]}

100% calculate · scored

{"decision": "calculate", "currency": "LCR", "gross_pay": "3204.92", "employee_contribution": "160.25", "allowed_deduction": "244.67", "taxable_pay": "2800.00", "income_tax": "152.00", "net_pay": "2648.00", "clauses": ["R1", "R2", "R3", "R4", "R5", "R6"]}

100% calculate · scored

{"decision": "calculate", "currency": "LCR", "gross_pay": "3204.92", "employee_contribution": "160.25", "allowed_deduction": "244.67", "taxable_pay": "2800.00", "income_tax": "152.00", "net_pay": "2648.00", "clauses": ["R1", "R2", "R3", "R4", "R5", "R6"]}

100% calculate · scored

{"decision": "calculate", "currency": "LCR", "gross_pay": "3204.92", "employee_contribution": "160.25", "allowed_deduction": "244.67", "taxable_pay": "2800.00", "income_tax": "152.00", "net_pay": "2648.00", "clauses": ["R1", "R2", "R3", "R4", "R5", "R6"]}

case4-4001-00329 · Trap MISSING_PAYROLL_MONTH

100% clarify · scored

{"decision": "clarify", "reason_code": "MISSING_PAYROLL_MONTH", "clauses": ["R1", "R7"], "explanation": "The payroll month is missing, so no reference row can be chosen. Please provide the payroll month."}

100% clarify · scored

{"decision": "clarify", "reason_code": "MISSING_PAYROLL_MONTH", "clauses": ["R1", "R7"], "explanation": "The payroll month is missing, so no reference row can be chosen. Please provide the payroll month."}

100% clarify · scored

{"decision": "clarify", "reason_code": "MISSING_PAYROLL_MONTH", "clauses": ["R1", "R7"], "explanation": "The payroll month is missing, so no reference row can be chosen. Please provide the payroll month."}

100% clarify · scored

{"decision": "clarify", "reason_code": "MISSING_PAYROLL_MONTH", "clauses": ["R1", "R7"], "explanation": "The payroll month is missing, so no reference row can be chosen. Please provide the payroll month."}

case4-4001-00005 · Trap NO_APPLICABLE_REFERENCE

100% clarify · scored

{"decision": "clarify", "reason_code": "NO_APPLICABLE_REFERENCE", "clauses": ["R1", "R7"], "explanation": "No reference row covers this payroll month. Please provide an authoritative schedule that covers it."}

100% clarify · scored

{"decision": "clarify", "reason_code": "NO_APPLICABLE_REFERENCE", "clauses": ["R1", "R7"], "explanation": "No reference row covers this payroll month. Please provide an authoritative schedule that covers it."}

100% clarify · scored

{"decision": "clarify", "reason_code": "NO_APPLICABLE_REFERENCE", "clauses": ["R1", "R7"], "explanation": "No reference row covers this payroll month. Please provide an authoritative schedule that covers it."}

100% clarify · scored

{"decision": "clarify", "reason_code": "NO_APPLICABLE_REFERENCE", "clauses": ["R1", "R7"], "explanation": "No reference row covers this payroll month. Please provide an authoritative schedule that covers it."}