id stringlengths 5 5 | now timestamp[s]date 2023-03-12 18:00:00 2028-11-17 18:00:00 | source stringclasses 1
value | group stringclasses 8
values | family stringclasses 20
values | difficulty stringclasses 3
values | query stringlengths 4 72 | context listlengths 0 4 | no_tool bool 2
classes | mode stringclasses 1
value | held_out_tool bool 1
class | has_cents bool 2
classes | persona_uid stringclasses 45
values | persona_axes dict | accept listlengths 0 16 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
s0000 | 2023-08-25T18:00:00 | synth | followup | followup | medium | which categories? | [
{
"role": "user",
"content": "how much did i spend this month"
},
{
"role": "assistant",
"content": "",
"tool_calls": [
{
"id": "call_1",
"type": "function",
"function": {
"name": "aggregate",
"arguments": {
"metric": "sum",
... | false | null | false | false | e04 | {
"income": "low",
"fuliza": true,
"profile": "personal",
"intensity": "heavy"
} | [
{
"call": {
"name": "aggregate",
"args": {
"metric": "sum",
"filter": {
"class": "spending",
"period": "thisMonth"
},
"group_by": "category"
}
},
"result": {
"metric": "sum",
"group_by": "category",
"groups": [
... |
s0001 | 2027-12-14T18:00:00 | synth | followup | followup | medium | and yesterday? | [
{
"role": "user",
"content": "how much did i spend today"
},
{
"role": "assistant",
"content": "",
"tool_calls": [
{
"id": "call_1",
"type": "function",
"function": {
"name": "aggregate",
"arguments": {
"metric": "sum",
... | false | null | false | false | e10 | {
"income": "mid",
"fuliza": true,
"profile": "mixed",
"intensity": "light"
} | [
{
"call": {
"name": "aggregate",
"args": {
"metric": "sum",
"filter": {
"class": "spending",
"period": "custom",
"start": "2027-12-13",
"end": "2027-12-14"
}
}
},
"result": {
"metric": "sum",
"value": 1199
... |
s0002 | 2026-05-16T18:00:00 | synth | list_merchant | merchant | hard | how much have i sent to Beatrice | [] | false | null | false | false | e27 | {
"income": "low",
"fuliza": true,
"profile": "mixed",
"intensity": "light"
} | [
{
"call": {
"name": "query_transactions",
"args": {
"filter": {
"merchant": [
"BEATRICE ATIENO 0702526852",
"254702526852 BEATRICE ATIENO",
"BEATRICE ATIENO 0702***852"
]
}
}
},
"result": {
"count": 56,
... |
s0003 | 2026-03-23T18:00:00 | synth | followup | followup | medium | what about yesterday | [
{
"role": "user",
"content": "how much did i spend today"
},
{
"role": "assistant",
"content": "",
"tool_calls": [
{
"id": "call_1",
"type": "function",
"function": {
"name": "aggregate",
"arguments": {
"metric": "sum",
... | false | null | false | true | e32 | {
"income": "mid",
"fuliza": true,
"profile": "personal",
"intensity": "heavy"
} | [
{
"call": {
"name": "aggregate",
"args": {
"metric": "sum",
"filter": {
"class": "spending",
"period": "custom",
"start": "2026-03-22",
"end": "2026-03-23"
}
}
},
"result": {
"metric": "sum",
"value": 3790.05
... |
s0004 | 2027-07-18T18:00:00 | synth | no_tool | no_tool | easy | hi there | [] | true | null | false | false | e24 | {
"income": "mid",
"fuliza": true,
"profile": "mixed",
"intensity": "heavy"
} | [] |
s0005 | 2027-06-04T18:00:00 | synth | no_tool | no_tool | easy | hi thee | [] | true | null | false | false | e25 | {
"income": "mid",
"fuliza": true,
"profile": "personal",
"intensity": "heavy"
} | [] |
s0006 | 2027-11-17T18:00:00 | synth | balance_class | transfers | medium | How much did i move between my own accounts so far this year | [] | false | null | false | true | e33 | {
"income": "mid",
"fuliza": false,
"profile": "mixed",
"intensity": "light"
} | [
{
"call": {
"name": "aggregate",
"args": {
"metric": "sum",
"filter": {
"class": "transfers",
"period": "thisYear"
}
}
},
"result": {
"metric": "sum",
"value": 142959.16
}
}
] |
s0007 | 2028-01-08T18:00:00 | synth | trend_compare | matrix | hard | which day of the wk do i spend the most the previous mnth | [] | false | null | false | false | e10 | {
"income": "mid",
"fuliza": true,
"profile": "mixed",
"intensity": "light"
} | [
{
"call": {
"name": "matrix",
"args": {
"metric": "sum",
"rowKey": "weekday",
"colKey": "class",
"filter": {
"class": "spending",
"period": "lastMonth"
}
}
},
"result": {
"metric": "sum",
"rowKey": "weekday",
... |
s0008 | 2027-05-06T18:00:00 | synth | trend_compare | matrix | hard | wich day of the week do i spnd the most previous month | [] | false | null | false | false | e24 | {
"income": "mid",
"fuliza": true,
"profile": "mixed",
"intensity": "heavy"
} | [
{
"call": {
"name": "matrix",
"args": {
"metric": "sum",
"rowKey": "weekday",
"colKey": "week",
"filter": {
"class": "spending",
"period": "lastMonth"
}
}
},
"result": {
"metric": "sum",
"rowKey": "weekday",
... |
s0009 | 2026-06-08T18:00:00 | synth | balance_class | income | medium | hw much did i receive so far this week | [] | false | null | false | false | e43 | {
"income": "mid",
"fuliza": true,
"profile": "mixed",
"intensity": "heavy"
} | [
{
"call": {
"name": "aggregate",
"args": {
"metric": "sum",
"filter": {
"class": "income",
"period": "thisWeek"
}
}
},
"result": {
"metric": "sum",
"value": 1044
}
}
] |
s0010 | 2025-06-18T18:00:00 | synth | balance_class | balance | easy | mpesa balance? | [] | false | null | false | false | e03 | {
"income": "high",
"fuliza": false,
"profile": "personal",
"intensity": "heavy"
} | [
{
"call": {
"name": "get_balance",
"args": {
"source": "mpesa"
}
},
"result": {
"source": "mpesa",
"available": true,
"balance": 64790.24,
"as_of": "2025-06-18T20:18:00.000"
}
}
] |
s0011 | 2028-01-22T18:00:00 | synth | balance_class | balance | easy | what's my mpesa balance | [] | false | null | false | false | e33 | {
"income": "mid",
"fuliza": false,
"profile": "mixed",
"intensity": "light"
} | [
{
"call": {
"name": "get_balance",
"args": {
"source": "mpesa"
}
},
"result": {
"source": "mpesa",
"available": true,
"balance": 0,
"as_of": "2028-01-22T20:28:00.000"
}
}
] |
s0012 | 2027-06-01T18:00:00 | synth | amount | spend_custom | medium | total spending the previous week? | [] | false | null | false | true | e24 | {
"income": "mid",
"fuliza": true,
"profile": "mixed",
"intensity": "heavy"
} | [
{
"call": {
"name": "aggregate",
"args": {
"metric": "sum",
"filter": {
"class": "spending",
"period": "custom",
"start": "2027-05-24",
"end": "2027-05-31"
}
}
},
"result": {
"metric": "sum",
"value": 28697.55
... |
s0013 | 2024-12-15T18:00:00 | synth | amount | spend_custom | hard | spending last weekend | [] | false | null | false | true | e34 | {
"income": "low",
"fuliza": true,
"profile": "personal",
"intensity": "heavy"
} | [
{
"call": {
"name": "aggregate",
"args": {
"metric": "sum",
"filter": {
"class": "spending",
"period": "custom",
"start": "2024-12-07",
"end": "2024-12-09"
}
}
},
"result": {
"metric": "sum",
"value": 5295.82
... |
s0014 | 2027-01-31T18:00:00 | synth | amount | spend_total | easy | what did i spend today so far | [] | false | null | false | false | e20 | {
"income": "high",
"fuliza": false,
"profile": "personal",
"intensity": "light"
} | [
{
"call": {
"name": "aggregate",
"args": {
"metric": "sum",
"filter": {
"class": "spending",
"period": "today"
}
}
},
"result": {
"metric": "sum",
"value": 1846
}
}
] |
s0015 | 2028-06-20T18:00:00 | synth | list_merchant | merchant | hard | ttal sent to ann nyambura | [] | false | null | false | false | e41 | {
"income": "mid",
"fuliza": false,
"profile": "personal",
"intensity": "heavy"
} | [
{
"call": {
"name": "query_transactions",
"args": {
"filter": {
"merchant": [
"ANN NYAMBURA 0769004754",
"254769004754 ANN NYAMBURA",
"ANN NYAMBURA 0769***754"
]
}
}
},
"result": {
"count": 9,
"returned... |
s0016 | 2026-09-13T18:00:00 | synth | advice | advice | medium | give me advice to reduce my spending | [] | false | null | false | false | e20 | {
"income": "high",
"fuliza": false,
"profile": "personal",
"intensity": "light"
} | [
{
"call": {
"name": "aggregate",
"args": {
"metric": "sum",
"filter": {
"class": "spending",
"period": "lastMonth"
},
"group_by": "category"
}
},
"result": {
"metric": "sum",
"group_by": "category",
"groups": [
... |
s0017 | 2023-08-01T18:00:00 | synth | advice | advice | medium | how do i spend less | [] | false | null | false | false | e04 | {
"income": "low",
"fuliza": true,
"profile": "personal",
"intensity": "heavy"
} | [
{
"call": {
"name": "aggregate",
"args": {
"metric": "sum",
"filter": {
"class": "spending",
"period": "lastMonth"
},
"group_by": "category"
}
},
"result": {
"metric": "sum",
"group_by": "category",
"groups": [
... |
s0018 | 2025-02-16T18:00:00 | synth | amount | spend_total | easy | what did i spend last month | [] | false | null | false | true | e03 | {
"income": "high",
"fuliza": false,
"profile": "personal",
"intensity": "heavy"
} | [
{
"call": {
"name": "aggregate",
"args": {
"metric": "sum",
"filter": {
"class": "spending",
"period": "lastMonth"
}
}
},
"result": {
"metric": "sum",
"value": 405500.1
}
}
] |
s0019 | 2027-04-29T18:00:00 | synth | advice | advice | medium | were should i cut back | [] | false | null | false | false | e35 | {
"income": "high",
"fuliza": true,
"profile": "mixed",
"intensity": "heavy"
} | [
{
"call": {
"name": "aggregate",
"args": {
"metric": "sum",
"filter": {
"class": "spending",
"period": "lastMonth"
},
"group_by": "category"
}
},
"result": {
"metric": "sum",
"group_by": "category",
"groups": [
... |
s0020 | 2024-12-29T18:00:00 | synth | list_merchant | merchant | hard | How much did i receive from kevin mutua | [] | false | null | false | false | e06 | {
"income": "mid",
"fuliza": false,
"profile": "personal",
"intensity": "heavy"
} | [
{
"call": {
"name": "query_transactions",
"args": {
"filter": {
"merchant": [
"Kevin Mutua 0759665448",
"254759665448 Kevin Mutua",
"Kevin Mutua 0759***448"
]
}
}
},
"result": {
"count": 2,
"returned": ... |
s0021 | 2025-03-12T18:00:00 | synth | followup | followup | medium | Break it down by category | [
{
"role": "user",
"content": "how much did i spend this month"
},
{
"role": "assistant",
"content": "",
"tool_calls": [
{
"id": "call_1",
"type": "function",
"function": {
"name": "aggregate",
"arguments": {
"metric": "sum",
... | false | null | false | false | e01 | {
"income": "high",
"fuliza": false,
"profile": "personal",
"intensity": "heavy"
} | [
{
"call": {
"name": "aggregate",
"args": {
"metric": "sum",
"filter": {
"class": "spending",
"period": "thisMonth"
},
"group_by": "category"
}
},
"result": {
"metric": "sum",
"group_by": "category",
"groups": [
... |
s0022 | 2026-08-22T18:00:00 | synth | amount | spend_category | medium | what did i spend on fare so far today | [] | false | null | false | false | e45 | {
"income": "low",
"fuliza": true,
"profile": "personal",
"intensity": "light"
} | [
{
"call": {
"name": "aggregate",
"args": {
"metric": "sum",
"filter": {
"class": "spending",
"category": "Transport & Fuel",
"period": "today"
}
}
},
"result": {
"metric": "sum",
"value": 316
}
}
] |
s0023 | 2024-08-31T18:00:00 | synth | advice | advice | medium | How do i spend less | [] | false | null | false | false | e30 | {
"income": "low",
"fuliza": true,
"profile": "personal",
"intensity": "light"
} | [
{
"call": {
"name": "aggregate",
"args": {
"metric": "sum",
"filter": {
"class": "spending",
"period": "lastMonth"
},
"group_by": "category"
}
},
"result": {
"metric": "sum",
"group_by": "category",
"groups": [
... |
s0024 | 2023-05-03T18:00:00 | synth | trend_compare | trend | medium | chart my spending year to date | [] | false | null | false | false | e36 | {
"income": "low",
"fuliza": false,
"profile": "business",
"intensity": "heavy"
} | [
{
"call": {
"name": "time_series",
"args": {
"metric": "sum",
"bucket": "monthly",
"filter": {
"class": "spending",
"period": "thisYear"
}
}
},
"result": {
"metric": "sum",
"bucket": "monthly",
"buckets": [
{... |
s0025 | 2025-06-02T18:00:00 | synth | balance_class | balance | easy | mpesa balance? | [] | false | null | false | false | e34 | {
"income": "low",
"fuliza": true,
"profile": "personal",
"intensity": "heavy"
} | [
{
"call": {
"name": "get_balance",
"args": {
"source": "mpesa"
}
},
"result": {
"source": "mpesa",
"available": true,
"balance": 69602.42,
"as_of": "2025-06-02T14:46:00.000"
}
}
] |
s0026 | 2027-10-17T18:00:00 | synth | breakdown_count | breakdown | medium | spnding by categoy last weekend | [] | false | null | false | false | e24 | {
"income": "mid",
"fuliza": true,
"profile": "mixed",
"intensity": "heavy"
} | [
{
"call": {
"name": "aggregate",
"args": {
"metric": "sum",
"filter": {
"class": "spending",
"period": "custom",
"start": "2027-10-09",
"end": "2027-10-11"
},
"group_by": "category"
}
},
"result": {
"metric":... |
s0027 | 2027-12-26T18:00:00 | synth | breakdown_count | breakdown | medium | spnding by category last wekend | [] | false | null | false | false | e29 | {
"income": "mid",
"fuliza": true,
"profile": "personal",
"intensity": "heavy"
} | [
{
"call": {
"name": "aggregate",
"args": {
"metric": "sum",
"filter": {
"class": "spending",
"period": "custom",
"start": "2027-12-18",
"end": "2027-12-20"
},
"group_by": "category"
}
},
"result": {
"metric":... |
s0028 | 2026-09-23T18:00:00 | synth | balance_class | income | medium | how much money came in this wk | [] | false | null | false | false | e43 | {
"income": "mid",
"fuliza": true,
"profile": "mixed",
"intensity": "heavy"
} | [
{
"call": {
"name": "aggregate",
"args": {
"metric": "sum",
"filter": {
"class": "income",
"period": "thisWeek"
}
}
},
"result": {
"metric": "sum",
"value": 307
}
}
] |
s0029 | 2026-12-25T18:00:00 | synth | advice | advice | medium | where should i cut back | [] | false | null | false | false | e17 | {
"income": "mid",
"fuliza": true,
"profile": "personal",
"intensity": "light"
} | [
{
"call": {
"name": "aggregate",
"args": {
"metric": "sum",
"filter": {
"class": "spending",
"period": "lastMonth"
},
"group_by": "category"
}
},
"result": {
"metric": "sum",
"group_by": "category",
"groups": [
... |
s0030 | 2026-06-21T18:00:00 | synth | followup | followup | medium | and yesterday | [
{
"role": "user",
"content": "how much did i spend today"
},
{
"role": "assistant",
"content": "",
"tool_calls": [
{
"id": "call_1",
"type": "function",
"function": {
"name": "aggregate",
"arguments": {
"metric": "sum",
... | false | null | false | true | e31 | {
"income": "low",
"fuliza": true,
"profile": "personal",
"intensity": "heavy"
} | [
{
"call": {
"name": "aggregate",
"args": {
"metric": "sum",
"filter": {
"class": "spending",
"period": "custom",
"start": "2026-06-20",
"end": "2026-06-21"
}
}
},
"result": {
"metric": "sum",
"value": 593.59
... |
s0031 | 2027-05-08T18:00:00 | synth | list_merchant | list | medium | transactions above 1000 last weekend | [] | false | null | false | false | e22 | {
"income": "mid",
"fuliza": false,
"profile": "personal",
"intensity": "light"
} | [
{
"call": {
"name": "query_transactions",
"args": {
"filter": {
"class": "spending",
"period": "custom",
"start": "2027-05-01",
"end": "2027-05-03",
"amount": {
"min": 100000
}
},
"sort": "amountDesc"
... |
s0032 | 2028-11-12T18:00:00 | synth | list_merchant | merchant | hard | show my transactions with naoi waweru | [] | false | null | false | false | e37 | {
"income": "mid",
"fuliza": true,
"profile": "mixed",
"intensity": "heavy"
} | [
{
"call": {
"name": "query_transactions",
"args": {
"filter": {
"merchant": [
"Naomi Waweru 0722275800",
"254722275800 Naomi Waweru",
"Naomi Waweru 0722***800"
]
}
}
},
"result": {
"count": 11,
"returne... |
s0033 | 2027-04-01T18:00:00 | synth | amount | spend_custom | medium | what did i sped year to date | [] | false | null | false | true | e19 | {
"income": "high",
"fuliza": false,
"profile": "personal",
"intensity": "heavy"
} | [
{
"call": {
"name": "aggregate",
"args": {
"metric": "sum",
"filter": {
"class": "spending",
"period": "thisYear"
}
}
},
"result": {
"metric": "sum",
"value": 746947.59
}
}
] |
s0034 | 2028-01-22T18:00:00 | synth | breakdown_count | count | medium | number of payments this wk | [] | false | null | false | false | e29 | {
"income": "mid",
"fuliza": true,
"profile": "personal",
"intensity": "heavy"
} | [
{
"call": {
"name": "aggregate",
"args": {
"metric": "count",
"filter": {
"class": "spending",
"period": "thisWeek"
}
}
},
"result": {
"metric": "count",
"value": 22
}
},
{
"call": {
"name": "query_transactions",... |
s0035 | 2024-10-05T18:00:00 | synth | trend_compare | trend | medium | show me a graph of my spending this mnth | [] | false | null | false | false | e13 | {
"income": "mid",
"fuliza": true,
"profile": "mixed",
"intensity": "light"
} | [
{
"call": {
"name": "time_series",
"args": {
"metric": "sum",
"bucket": "daily",
"filter": {
"class": "spending",
"period": "thisMonth"
}
}
},
"result": {
"metric": "sum",
"bucket": "daily",
"buckets": [
{
... |
s0036 | 2028-05-14T18:00:00 | synth | breakdown_count | breakdown | easy | break down my spending the previous month by category | [] | false | null | false | false | e26 | {
"income": "mid",
"fuliza": false,
"profile": "mixed",
"intensity": "light"
} | [
{
"call": {
"name": "aggregate",
"args": {
"metric": "sum",
"filter": {
"class": "spending",
"period": "lastMonth"
},
"group_by": "category"
}
},
"result": {
"metric": "sum",
"group_by": "category",
"groups": [
... |
s0037 | 2026-09-07T18:00:00 | synth | advice | advice | medium | how can i save more mney | [] | false | null | false | false | e21 | {
"income": "mid",
"fuliza": true,
"profile": "personal",
"intensity": "heavy"
} | [
{
"call": {
"name": "aggregate",
"args": {
"metric": "sum",
"filter": {
"class": "spending",
"period": "lastMonth"
},
"group_by": "category"
}
},
"result": {
"metric": "sum",
"group_by": "category",
"groups": [
... |
s0038 | 2028-08-26T18:00:00 | synth | followup | followup | medium | how about last month | [
{
"role": "user",
"content": "how much did i spend on netflix & spotify this month"
},
{
"role": "assistant",
"content": "",
"tool_calls": [
{
"id": "call_1",
"type": "function",
"function": {
"name": "aggregate",
"arguments": {
"... | false | null | false | true | e37 | {
"income": "mid",
"fuliza": true,
"profile": "mixed",
"intensity": "heavy"
} | [
{
"call": {
"name": "aggregate",
"args": {
"metric": "sum",
"filter": {
"class": "spending",
"category": "Netflix & Spotify",
"period": "lastMonth"
}
}
},
"result": {
"metric": "sum",
"value": 2154.39
}
}
] |
s0039 | 2023-10-17T18:00:00 | synth | trend_compare | matrix | hard | spnding heatmap by weekday and week this mnth | [] | false | null | false | false | e04 | {
"income": "low",
"fuliza": true,
"profile": "personal",
"intensity": "heavy"
} | [
{
"call": {
"name": "matrix",
"args": {
"metric": "sum",
"rowKey": "weekday",
"colKey": "week",
"filter": {
"class": "spending",
"period": "thisMonth"
}
}
},
"result": {
"metric": "sum",
"rowKey": "weekday",
... |
s0040 | 2025-04-22T18:00:00 | synth | trend_compare | compare | hard | am i spending more than last month? | [] | false | null | false | false | e23 | {
"income": "low",
"fuliza": true,
"profile": "mixed",
"intensity": "heavy"
} | [
{
"call": {
"name": "compare_periods",
"args": {
"metric": "sum",
"filter": {
"class": "spending"
},
"periodA": {
"period": "lastMonth"
},
"periodB": {
"period": "thisMonth"
}
}
},
"result": {
"... |
s0041 | 2027-02-25T18:00:00 | synth | amount | spend_total | easy | How much did i spend so far this year | [] | false | null | false | true | e35 | {
"income": "high",
"fuliza": true,
"profile": "mixed",
"intensity": "heavy"
} | [
{
"call": {
"name": "aggregate",
"args": {
"metric": "sum",
"filter": {
"class": "spending",
"period": "thisYear"
}
}
},
"result": {
"metric": "sum",
"value": 598871.23
}
}
] |
s0042 | 2026-10-28T18:00:00 | synth | followup | followup | medium | Break it down by category | [
{
"role": "user",
"content": "how much did i spend this month"
},
{
"role": "assistant",
"content": "",
"tool_calls": [
{
"id": "call_1",
"type": "function",
"function": {
"name": "aggregate",
"arguments": {
"metric": "sum",
... | false | null | false | false | e11 | {
"income": "high",
"fuliza": true,
"profile": "personal",
"intensity": "light"
} | [
{
"call": {
"name": "aggregate",
"args": {
"metric": "sum",
"filter": {
"class": "spending",
"period": "thisMonth"
},
"group_by": "category"
}
},
"result": {
"metric": "sum",
"group_by": "category",
"groups": [
... |
s0043 | 2028-04-26T18:00:00 | synth | list_merchant | list | medium | transactions above 5000 year to date | [] | false | null | false | false | e37 | {
"income": "mid",
"fuliza": true,
"profile": "mixed",
"intensity": "heavy"
} | [
{
"call": {
"name": "query_transactions",
"args": {
"filter": {
"class": "spending",
"period": "thisYear",
"amount": {
"min": 500000
}
},
"sort": "amountDesc"
}
},
"result": {
"count": 17,
"returned... |
s0044 | 2026-12-06T18:00:00 | synth | list_merchant | list | medium | largest payments this month | [] | false | top1 | false | false | e09 | {
"income": "low",
"fuliza": false,
"profile": "business",
"intensity": "heavy"
} | [
{
"call": {
"name": "query_transactions",
"args": {
"filter": {
"class": "spending",
"period": "thisMonth"
},
"sort": "amountDesc",
"limit": 5
}
},
"result": {
"count": 25,
"returned": 5,
"truncated": true,
"ro... |
s0045 | 2028-04-01T18:00:00 | synth | followup | followup | medium | how about last month | [
{
"role": "user",
"content": "how much did i spend on getting around this month"
},
{
"role": "assistant",
"content": "",
"tool_calls": [
{
"id": "call_1",
"type": "function",
"function": {
"name": "aggregate",
"arguments": {
"met... | false | null | false | true | e29 | {
"income": "mid",
"fuliza": true,
"profile": "personal",
"intensity": "heavy"
} | [
{
"call": {
"name": "aggregate",
"args": {
"metric": "sum",
"filter": {
"class": "spending",
"category": "Transport & Fuel",
"period": "lastMonth"
}
}
},
"result": {
"metric": "sum",
"value": 2887.45
}
}
] |
s0046 | 2028-01-30T18:00:00 | synth | trend_compare | compare | hard | am i spending more on medical cover than last month | [] | false | null | false | false | e10 | {
"income": "mid",
"fuliza": true,
"profile": "mixed",
"intensity": "light"
} | [
{
"call": {
"name": "compare_periods",
"args": {
"metric": "sum",
"filter": {
"class": "spending",
"category": "Medical Cover"
},
"periodA": {
"period": "lastMonth"
},
"periodB": {
"period": "thisMonth"
}... |
s0047 | 2026-04-27T18:00:00 | synth | amount | spend_total | easy | total spending this week? | [] | false | null | false | false | e21 | {
"income": "mid",
"fuliza": true,
"profile": "personal",
"intensity": "heavy"
} | [
{
"call": {
"name": "aggregate",
"args": {
"metric": "sum",
"filter": {
"class": "spending",
"period": "thisWeek"
}
}
},
"result": {
"metric": "sum",
"value": 320
}
}
] |
s0048 | 2028-06-07T18:00:00 | synth | advice | advice | medium | give me advice to reduce my spnding | [] | false | null | false | false | e41 | {
"income": "mid",
"fuliza": false,
"profile": "personal",
"intensity": "heavy"
} | [
{
"call": {
"name": "aggregate",
"args": {
"metric": "sum",
"filter": {
"class": "spending",
"period": "thisMonth"
},
"group_by": "category"
}
},
"result": {
"metric": "sum",
"group_by": "category",
"groups": [
... |
s0049 | 2027-06-29T18:00:00 | synth | amount | spend_custom | hard | Spending the previous week | [] | false | null | false | true | e17 | {
"income": "mid",
"fuliza": true,
"profile": "personal",
"intensity": "light"
} | [
{
"call": {
"name": "aggregate",
"args": {
"metric": "sum",
"filter": {
"class": "spending",
"period": "custom",
"start": "2027-06-21",
"end": "2027-06-28"
}
}
},
"result": {
"metric": "sum",
"value": 11061.68
... |
s0050 | 2025-10-25T18:00:00 | synth | breakdown_count | count | medium | How many times did i pay for stuff this week | [] | false | null | false | false | e16 | {
"income": "mid",
"fuliza": false,
"profile": "personal",
"intensity": "light"
} | [
{
"call": {
"name": "aggregate",
"args": {
"metric": "count",
"filter": {
"class": "spending",
"period": "thisWeek"
}
}
},
"result": {
"metric": "count",
"value": 12
}
},
{
"call": {
"name": "query_transactions",... |
s0051 | 2026-07-23T18:00:00 | synth | balance_class | withdrawals | medium | how much cash did i withdraw the previous month | [] | false | null | false | false | e20 | {
"income": "high",
"fuliza": false,
"profile": "personal",
"intensity": "light"
} | [
{
"call": {
"name": "aggregate",
"args": {
"metric": "sum",
"filter": {
"class": "withdrawals",
"period": "lastMonth"
}
}
},
"result": {
"metric": "sum",
"value": 50300
}
}
] |
s0052 | 2025-02-20T18:00:00 | synth | followup | followup | medium | which categories? | [
{
"role": "user",
"content": "how much did i spend this month"
},
{
"role": "assistant",
"content": "",
"tool_calls": [
{
"id": "call_1",
"type": "function",
"function": {
"name": "aggregate",
"arguments": {
"metric": "sum",
... | false | null | false | false | e42 | {
"income": "low",
"fuliza": true,
"profile": "mixed",
"intensity": "heavy"
} | [
{
"call": {
"name": "aggregate",
"args": {
"metric": "sum",
"filter": {
"class": "spending",
"period": "thisMonth"
},
"group_by": "category"
}
},
"result": {
"metric": "sum",
"group_by": "category",
"groups": [
... |
s0053 | 2027-04-01T18:00:00 | synth | breakdown_count | count | medium | how many tmes did i pay for stuff so far this year | [] | false | null | false | false | e24 | {
"income": "mid",
"fuliza": true,
"profile": "mixed",
"intensity": "heavy"
} | [
{
"call": {
"name": "aggregate",
"args": {
"metric": "count",
"filter": {
"class": "spending",
"period": "thisYear"
}
}
},
"result": {
"metric": "count",
"value": 283
}
},
{
"call": {
"name": "query_transactions"... |
s0054 | 2025-08-26T18:00:00 | synth | amount | spend_source | medium | mpesa spending so far this year | [] | false | null | false | true | e16 | {
"income": "mid",
"fuliza": false,
"profile": "personal",
"intensity": "light"
} | [
{
"call": {
"name": "aggregate",
"args": {
"metric": "sum",
"filter": {
"class": "spending",
"source": "mpesa",
"period": "thisYear"
}
}
},
"result": {
"metric": "sum",
"value": 603938.95
}
}
] |
s0055 | 2023-12-02T18:00:00 | synth | followup | followup | medium | break it down by category | [
{
"role": "user",
"content": "how much did i spend this month"
},
{
"role": "assistant",
"content": "",
"tool_calls": [
{
"id": "call_1",
"type": "function",
"function": {
"name": "aggregate",
"arguments": {
"metric": "sum",
... | false | null | false | false | e38 | {
"income": "mid",
"fuliza": true,
"profile": "personal",
"intensity": "light"
} | [
{
"call": {
"name": "aggregate",
"args": {
"metric": "sum",
"filter": {
"class": "spending",
"period": "thisMonth"
},
"group_by": "category"
}
},
"result": {
"metric": "sum",
"group_by": "category",
"groups": [
... |
s0056 | 2028-07-20T18:00:00 | synth | no_tool | no_tool | easy | delee my fuliza transactions | [] | true | null | false | false | e41 | {
"income": "mid",
"fuliza": false,
"profile": "personal",
"intensity": "heavy"
} | [] |
s0057 | 2025-07-01T18:00:00 | synth | breakdown_count | count | medium | How many txns still need a category | [] | false | null | false | false | e07 | {
"income": "low",
"fuliza": false,
"profile": "personal",
"intensity": "light"
} | [
{
"call": {
"name": "aggregate",
"args": {
"metric": "count",
"filter": {
"categorySource": "none"
}
}
},
"result": {
"metric": "count",
"value": 77
}
},
{
"call": {
"name": "aggregate",
"args": {
"metric": "... |
s0058 | 2026-04-10T18:00:00 | synth | advice | advice | medium | how can i save more money | [] | false | null | false | false | e43 | {
"income": "mid",
"fuliza": true,
"profile": "mixed",
"intensity": "heavy"
} | [
{
"call": {
"name": "aggregate",
"args": {
"metric": "sum",
"filter": {
"class": "spending",
"period": "lastMonth"
},
"group_by": "category"
}
},
"result": {
"metric": "sum",
"group_by": "category",
"groups": [
... |
s0059 | 2026-08-11T18:00:00 | synth | trend_compare | trend | medium | chart my shopping spending previous month | [] | false | null | false | false | e32 | {
"income": "mid",
"fuliza": true,
"profile": "personal",
"intensity": "heavy"
} | [
{
"call": {
"name": "time_series",
"args": {
"metric": "sum",
"bucket": "daily",
"filter": {
"class": "spending",
"category": "Shopping",
"period": "lastMonth"
}
}
},
"result": {
"metric": "sum",
"bucket": "daily... |
s0060 | 2025-01-01T18:00:00 | synth | advice | advice | medium | how do i spnd less | [] | false | null | false | false | e42 | {
"income": "low",
"fuliza": true,
"profile": "mixed",
"intensity": "heavy"
} | [
{
"call": {
"name": "aggregate",
"args": {
"metric": "sum",
"filter": {
"class": "spending",
"period": "lastMonth"
},
"group_by": "category"
}
},
"result": {
"metric": "sum",
"group_by": "category",
"groups": [
... |
s0061 | 2027-02-25T18:00:00 | synth | followup | followup | medium | and last month? | [
{
"role": "user",
"content": "how much did i spend on kid this month"
},
{
"role": "assistant",
"content": "",
"tool_calls": [
{
"id": "call_1",
"type": "function",
"function": {
"name": "aggregate",
"arguments": {
"metric": "sum"... | false | null | false | true | e35 | {
"income": "high",
"fuliza": true,
"profile": "mixed",
"intensity": "heavy"
} | [
{
"call": {
"name": "aggregate",
"args": {
"metric": "sum",
"filter": {
"class": "spending",
"category": "Kids",
"period": "lastMonth"
}
}
},
"result": {
"metric": "sum",
"value": 21498.93
}
}
] |
s0062 | 2023-12-18T18:00:00 | synth | followup | followup | medium | break it down by category | [
{
"role": "user",
"content": "how much did i spend this month"
},
{
"role": "assistant",
"content": "",
"tool_calls": [
{
"id": "call_1",
"type": "function",
"function": {
"name": "aggregate",
"arguments": {
"metric": "sum",
... | false | null | false | false | e44 | {
"income": "mid",
"fuliza": false,
"profile": "personal",
"intensity": "light"
} | [
{
"call": {
"name": "aggregate",
"args": {
"metric": "sum",
"filter": {
"class": "spending",
"period": "thisMonth"
},
"group_by": "category"
}
},
"result": {
"metric": "sum",
"group_by": "category",
"groups": [
... |
s0063 | 2028-04-13T18:00:00 | synth | advice | advice | medium | give me advice to reduce my spndig | [] | false | null | false | false | e29 | {
"income": "mid",
"fuliza": true,
"profile": "personal",
"intensity": "heavy"
} | [
{
"call": {
"name": "aggregate",
"args": {
"metric": "sum",
"filter": {
"class": "spending",
"period": "thisMonth"
},
"group_by": "category"
}
},
"result": {
"metric": "sum",
"group_by": "category",
"groups": [
... |
s0064 | 2025-09-25T18:00:00 | synth | amount | spend_source | medium | m-pesa spending this mnth | [] | false | null | false | true | e07 | {
"income": "low",
"fuliza": false,
"profile": "personal",
"intensity": "light"
} | [
{
"call": {
"name": "aggregate",
"args": {
"metric": "sum",
"filter": {
"class": "spending",
"source": "mpesa",
"period": "thisMonth"
}
}
},
"result": {
"metric": "sum",
"value": 65932.7
}
}
] |
s0065 | 2024-12-22T18:00:00 | synth | trend_compare | matrix | hard | spnding heatmap by weekday and week in the current mnth | [] | false | null | false | false | e42 | {
"income": "low",
"fuliza": true,
"profile": "mixed",
"intensity": "heavy"
} | [
{
"call": {
"name": "matrix",
"args": {
"metric": "sum",
"rowKey": "weekday",
"colKey": "class",
"filter": {
"class": "spending",
"period": "thisMonth"
}
}
},
"result": {
"metric": "sum",
"rowKey": "weekday",
... |
s0066 | 2027-08-10T18:00:00 | synth | amount | spend_source | medium | m-pesa spending yesterday | [] | false | null | false | false | e40 | {
"income": "mid",
"fuliza": true,
"profile": "business",
"intensity": "light"
} | [
{
"call": {
"name": "aggregate",
"args": {
"metric": "sum",
"filter": {
"class": "spending",
"source": "mpesa",
"period": "custom",
"start": "2027-08-09",
"end": "2027-08-10"
}
}
},
"result": {
"metric": "s... |
s0067 | 2027-10-16T18:00:00 | synth | followup | followup | medium | Which categories? | [
{
"role": "user",
"content": "how much did i spend this month"
},
{
"role": "assistant",
"content": "",
"tool_calls": [
{
"id": "call_1",
"type": "function",
"function": {
"name": "aggregate",
"arguments": {
"metric": "sum",
... | false | null | false | false | e18 | {
"income": "mid",
"fuliza": false,
"profile": "personal",
"intensity": "heavy"
} | [
{
"call": {
"name": "aggregate",
"args": {
"metric": "sum",
"filter": {
"class": "spending",
"period": "thisMonth"
},
"group_by": "category"
}
},
"result": {
"metric": "sum",
"group_by": "category",
"groups": [
... |
s0068 | 2024-09-24T18:00:00 | synth | followup | followup | medium | how about yesterday | [
{
"role": "user",
"content": "how much did i spend today"
},
{
"role": "assistant",
"content": "",
"tool_calls": [
{
"id": "call_1",
"type": "function",
"function": {
"name": "aggregate",
"arguments": {
"metric": "sum",
... | false | null | false | false | e06 | {
"income": "mid",
"fuliza": false,
"profile": "personal",
"intensity": "heavy"
} | [
{
"call": {
"name": "aggregate",
"args": {
"metric": "sum",
"filter": {
"class": "spending",
"period": "custom",
"start": "2024-09-23",
"end": "2024-09-24"
}
}
},
"result": {
"metric": "sum",
"value": 6754
... |
s0069 | 2027-03-21T18:00:00 | synth | balance_class | withdrawals | medium | How much cash did i withdraw this month | [] | false | null | false | false | e09 | {
"income": "low",
"fuliza": false,
"profile": "business",
"intensity": "heavy"
} | [
{
"call": {
"name": "aggregate",
"args": {
"metric": "sum",
"filter": {
"class": "withdrawals",
"period": "thisMonth"
}
}
},
"result": {
"metric": "sum",
"value": 56500
}
}
] |
s0070 | 2028-03-23T18:00:00 | synth | trend_compare | compare | hard | this month vs last month | [] | false | null | false | false | e29 | {
"income": "mid",
"fuliza": true,
"profile": "personal",
"intensity": "heavy"
} | [
{
"call": {
"name": "compare_periods",
"args": {
"metric": "sum",
"filter": {
"class": "spending"
},
"periodA": {
"period": "lastMonth"
},
"periodB": {
"period": "thisMonth"
}
}
},
"result": {
"... |
s0071 | 2026-09-18T18:00:00 | synth | advice | advice | medium | how can i save more money | [] | false | null | false | false | e43 | {
"income": "mid",
"fuliza": true,
"profile": "mixed",
"intensity": "heavy"
} | [
{
"call": {
"name": "aggregate",
"args": {
"metric": "sum",
"filter": {
"class": "spending",
"period": "thisMonth"
},
"group_by": "category"
}
},
"result": {
"metric": "sum",
"group_by": "category",
"groups": [
... |
s0072 | 2028-03-09T18:00:00 | synth | trend_compare | compare | hard | Compare my saving spending this month and last month | [] | false | null | false | false | e15 | {
"income": "mid",
"fuliza": false,
"profile": "mixed",
"intensity": "heavy"
} | [
{
"call": {
"name": "compare_periods",
"args": {
"metric": "sum",
"filter": {
"class": "spending",
"category": "Savings"
},
"periodA": {
"period": "lastMonth"
},
"periodB": {
"period": "thisMonth"
}
... |
s0073 | 2026-01-08T18:00:00 | synth | no_tool | no_tool | easy | Transfer 500 to my savings | [] | true | null | false | false | e27 | {
"income": "low",
"fuliza": true,
"profile": "mixed",
"intensity": "light"
} | [] |
s0074 | 2025-08-23T18:00:00 | synth | followup | followup | medium | breakdown by category | [
{
"role": "user",
"content": "how much did i spend this month"
},
{
"role": "assistant",
"content": "",
"tool_calls": [
{
"id": "call_1",
"type": "function",
"function": {
"name": "aggregate",
"arguments": {
"metric": "sum",
... | false | null | false | false | e16 | {
"income": "mid",
"fuliza": false,
"profile": "personal",
"intensity": "light"
} | [
{
"call": {
"name": "aggregate",
"args": {
"metric": "sum",
"filter": {
"class": "spending",
"period": "thisMonth"
},
"group_by": "category"
}
},
"result": {
"metric": "sum",
"group_by": "category",
"groups": [
... |
s0075 | 2027-09-23T18:00:00 | synth | advice | advice | medium | give me advice to reduce ma spending | [] | false | null | false | false | e25 | {
"income": "mid",
"fuliza": true,
"profile": "personal",
"intensity": "heavy"
} | [
{
"call": {
"name": "aggregate",
"args": {
"metric": "sum",
"filter": {
"class": "spending",
"period": "lastMonth"
},
"group_by": "category"
}
},
"result": {
"metric": "sum",
"group_by": "category",
"groups": [
... |
s0076 | 2028-02-10T18:00:00 | synth | no_tool | no_tool | easy | thanks! | [] | true | null | false | false | e41 | {
"income": "mid",
"fuliza": false,
"profile": "personal",
"intensity": "heavy"
} | [] |
s0077 | 2026-06-20T18:00:00 | synth | followup | followup | medium | and yesterday? | [
{
"role": "user",
"content": "how much did i spend today"
},
{
"role": "assistant",
"content": "",
"tool_calls": [
{
"id": "call_1",
"type": "function",
"function": {
"name": "aggregate",
"arguments": {
"metric": "sum",
... | false | null | false | false | e20 | {
"income": "high",
"fuliza": false,
"profile": "personal",
"intensity": "light"
} | [
{
"call": {
"name": "aggregate",
"args": {
"metric": "sum",
"filter": {
"class": "spending",
"period": "custom",
"start": "2026-06-19",
"end": "2026-06-20"
}
}
},
"result": {
"metric": "sum",
"value": 8019
... |
s0078 | 2027-10-01T18:00:00 | synth | no_tool | no_tool | easy | Transfer 500 to my savings | [] | true | null | false | false | e10 | {
"income": "mid",
"fuliza": true,
"profile": "mixed",
"intensity": "light"
} | [] |
s0079 | 2025-09-05T18:00:00 | synth | advice | advice | medium | give me advice to reduce my spending | [] | false | null | false | false | e12 | {
"income": "mid",
"fuliza": true,
"profile": "mixed",
"intensity": "heavy"
} | [
{
"call": {
"name": "aggregate",
"args": {
"metric": "sum",
"filter": {
"class": "spending",
"period": "thisMonth"
},
"group_by": "category"
}
},
"result": {
"metric": "sum",
"group_by": "category",
"groups": [
... |
s0080 | 2025-04-07T18:00:00 | synth | breakdown_count | breakdown | easy | where did my money go last month | [] | false | null | false | false | e13 | {
"income": "mid",
"fuliza": true,
"profile": "mixed",
"intensity": "light"
} | [
{
"call": {
"name": "aggregate",
"args": {
"metric": "sum",
"filter": {
"class": "spending",
"period": "lastMonth"
},
"group_by": "category"
}
},
"result": {
"metric": "sum",
"group_by": "category",
"groups": [
... |
s0081 | 2026-10-01T18:00:00 | synth | no_tool | no_tool | easy | send 1000 to kevin | [] | true | null | false | false | e09 | {
"income": "low",
"fuliza": false,
"profile": "business",
"intensity": "heavy"
} | [] |
s0082 | 2027-12-06T18:00:00 | synth | breakdown_count | breakdown | medium | Where did my money go last weekend | [] | false | null | false | false | e26 | {
"income": "mid",
"fuliza": false,
"profile": "mixed",
"intensity": "light"
} | [
{
"call": {
"name": "aggregate",
"args": {
"metric": "sum",
"filter": {
"class": "spending",
"period": "custom",
"start": "2027-12-04",
"end": "2027-12-06"
},
"group_by": "category"
}
},
"result": {
"metric":... |
s0083 | 2025-05-30T18:00:00 | synth | amount | spend_source | medium | how much did i spend through mpesa last month | [] | false | null | false | true | e34 | {
"income": "low",
"fuliza": true,
"profile": "personal",
"intensity": "heavy"
} | [
{
"call": {
"name": "aggregate",
"args": {
"metric": "sum",
"filter": {
"class": "spending",
"source": "mpesa",
"period": "lastMonth"
}
}
},
"result": {
"metric": "sum",
"value": 109690.54
}
}
] |
s0084 | 2023-11-03T18:00:00 | synth | breakdown_count | breakdown | medium | what are ma top spending categories the last 30 days | [] | false | null | false | false | e14 | {
"income": "low",
"fuliza": true,
"profile": "personal",
"intensity": "heavy"
} | [
{
"call": {
"name": "aggregate",
"args": {
"metric": "sum",
"filter": {
"class": "spending",
"period": "custom",
"start": "2023-10-05",
"end": "2023-11-04"
},
"group_by": "category"
}
},
"result": {
"metric":... |
s0085 | 2027-07-25T18:00:00 | synth | balance_class | balance | easy | how much is on my m-pesa | [] | false | null | false | false | e19 | {
"income": "high",
"fuliza": false,
"profile": "personal",
"intensity": "heavy"
} | [
{
"call": {
"name": "get_balance",
"args": {
"source": "mpesa"
}
},
"result": {
"source": "mpesa",
"available": true,
"balance": 0,
"as_of": "2027-07-25T15:23:00.000"
}
}
] |
Code, scorer and the answer key (the app's tool code) live on GitHub: https://github.com/wilfredeveloper/day31-bench. The scored model is at https://huggingface.co/wilfredeveloper/day31-lfm2.5-1.2b-v5-GGUF.
Day31-Bench
A benchmark for small on-device language models that answer personal-finance questions by calling tools, set in Kenyan mobile money (M-Pesa, Fuliza, paybills, bank transfers).
A model gets a question such as "how much did I spend on transport on the 29th last month?" and a set of tools that query the user's transactions. It must pick the right tool with the right arguments. The benchmark checks the number the tool returned, not the model's prose. The tools are the real ones from the day31 budgeting app, so a pass means the phone would have shown the right number.
- 930 cases (
data/eval_v3.jsonl) over 45 synthetic people, each with their own generated M-Pesa/bank ledger. - 70 held-out-tool cases (
data/eval_v3_heldout_tool.jsonl): the model is given a tool it never saw in training (get_fuliza_summary) and must use it from its description alone. - No real financial data. Every ledger is synthetic. The app developer's own-ledger eval set is kept private and is not part of this release.
- No LLM judge. Expected answers are computed by running the app's own Dart tools (
answer_key/) on each synthetic ledger.
Results
VALUE = share of cases where a tool call the model made returned the expected answer. Brackets are 95% Wilson intervals.
All models are 4-bit (Q4_K_M) GGUFs run through Ollama (llama.cpp) at temperature 0.
| model | prompt | first 325 cases | all 930 cases |
|---|---|---|---|
| LFM2.5-1.2B-Instruct, stock | v2 | 10.5% [7.6–14.3] | not run |
| day31 finetune v3 | v3 | 82.2% [77.6–85.9] | not run |
| day31 finetune v4 | v3 | 82.5% [78.0–86.2] | not run |
| day31 finetune v5 (released, 0.73 GB) | v3 | 81.5% [77.0–85.4] | 84.7% [82.3–86.9] |
The first 325 cases are a fixed, nested subset of the 930. Use it for cheap comparisons, and the full set to detect smaller gaps. On the 325-case subset, v3, v4 and v5 are statistically tied (McNemar exact test, p > 0.6).
Where the finetuned 1.2B model still fails (v5, all 930): advice 46/106, matrix 8/31, merchant 36/61, hard difficulty 108/160.
On an earlier 25-case version of the held-out-tool split, finetunes v3 and v4 scored 0%: they answered confidently with a familiar tool instead.
What's in a case
{"id": "s0000", "now": "2023-08-25T21:00:00+03:00", "persona_uid": "e01",
"family": "followup", "difficulty": "medium", "has_cents": false,
"query": "which categories?",
"context": [ ...earlier turns, with real tool results... ],
"accept": [ {"call": {"name": "aggregate", "args": {...}}, "result": {...}}, ... ]}
now: the date the question is asked. Relative dates ("last weekend") are resolved against it, in East Africa Time.accept: every acceptable call, with the result the real tools return. Any model call whose result matches one of these passes.- 20 families, from simple totals to trends, comparisons, pivots, follow-ups, advice and refusals. 200 cases involve cents.
- Persona axes: income, Fuliza use, personal/business profile, merchant intensity.
Run it
# 1. Build the answer key (Dart ≥ 3.5)
cd answer_key && dart pub get && dart compile exe bin/ai_eval.dart -o ai_eval && cd ..
# 2. Serve your model (llama.cpp shown; Ollama also works) and run the benchmark
pip install transformers # only for the model's chat template
llama-server -m your-model.gguf --port 8080 &
TZ=Africa/Nairobi python3 harness/run_model.py --server llamacpp --url http://127.0.0.1:8080 \
--model your-model --tokenizer <hf-tokenizer-dir> --tag my-run --prompt-version v3
# 3. Score (writes runs/my-run/score.json; pass several tags to compare)
python3 harness/score.py my-run ft-1.2b-v5-q4km
Use --format lfm3 for models that emit Qwen-style XML tool calls. For the held-out-tool split, add
--eval data/eval_v3_heldout_tool.jsonl --spec data/ai_spec_h3.json to the run, and --eval data/eval_v3_heldout_tool.jsonl to scoring.
runs/ holds the reference predictions behind the table above. python3 harness/score.py ft-1.2b-v5-q4km reproduces 788/930.
Layout
| path | what |
|---|---|
data/eval_v3.jsonl, data/eval_v3_heldout_tool.jsonl |
the cases |
data/ledgers/ |
the 45 synthetic ledgers (transactions + account links) |
data/ai_spec.json, data/ai_spec_h3.json |
the system prompt and tool schemas, exported from the app |
answer_key/ |
the app's tool code, extracted verbatim, plus a small JSON-lines server (bin/ai_eval.dart) |
harness/ |
model driver, chat-format parsers, scorer |
runs/ |
reference predictions and scores |
Limitations
- Synthetic data. Questions and ledgers are generated. Real users phrase things differently, and the gap between synthetic and real scores is not yet published.
- English only. Swahili and Sheng are not covered yet.
- One app's tools. The benchmark measures tool use against day31's tool design. Other designs would need their own answer key.
- Prompt matters. Finetunes should be scored with the prompt they were trained on. The prompt version is recorded in each run's
meta.json.
Licence
Code (answer_key/, harness/): Apache-2.0. Data (data/, runs/): CC BY 4.0.
The released model weights are a derivative of LFM2.5-1.2B and are distributed under Liquid AI's LFM Open License v1.0.
- Downloads last month
- 60