QT/sure
1
0
Fork 0
mirror of https://github.com/we-promise/sure.git synced 2026-04-09 15:24:48 +00:00
Code Issues Packages Projects Releases Wiki Activity
Files
d4ab29ae1bc03f6bae073d9669aa32f434cc94de
sure/db/eval_data/merchant_detection_golden_v1.yml
soky srm 88952e4714 Small llms improvements (#400) 
* Initial implementation

* FIX keys

* Add langfuse evals support

* FIX trace upload

* Delete .claude/settings.local.json

Signed-off-by: soky srm <sokysrm@gmail.com>

* Update client.rb

* Small LLMs improvements

* Keep batch size normal

* Update categorizer

* FIX json mode

* Add reasonable alternative to matching

* FIX thinking blocks for llms

* Implement json mode support with AUTO mode

* Make auto default for everyone

* FIX linter

* Address review

* Allow export manual categories

* FIX user export

* FIX oneshot example pollution

* Update categorization_golden_v1.yml

* Update categorization_golden_v1.yml

* Trim to 100 items

* Update auto_categorizer.rb

* FIX for auto retry in auto mode

* Separate the Eval Logic from the Auto-Categorizer

The expected_null_count parameter conflates eval-specific logic with production categorization logic.

* Force json mode on evals

* Introduce a more mixed dataset

150 items, performance from a local model:

By Difficulty:
  easy: 93.22% accuracy (55/59)
  medium: 93.33% accuracy (42/45)
  hard: 92.86% accuracy (26/28)
  edge_case: 100.0% accuracy (18/18)

* Improve datasets

Remove Data leakage from prompts

* Create eval runs as "pending"

---------

Signed-off-by: soky srm <sokysrm@gmail.com>
Signed-off-by: Juan José Mata <juanjo.mata@gmail.com>
Co-authored-by: Juan José Mata <juanjo.mata@gmail.com>
2025-12-07 18:11:34 +01:00

1118 lines
25 KiB
YAML
Raw Blame History

---
name: merchant_detection_golden_v1
description: Golden dataset for merchant name and URL detection evaluation
eval_type: merchant_detection
version: "1.0"
metadata:
created_at: "2024-12-01"
source: manual_curation
context:
merchants:
- id: "mcdonalds"
name: "McDonald's"
- id: "starbucks"
name: "Starbucks"
- id: "amazon"
name: "Amazon"
- id: "netflix"
name: "Netflix"
- id: "uber"
name: "Uber"
- id: "spotify"
name: "Spotify"
- id: "target"
name: "Target"
- id: "costco"
name: "Costco"
- id: "apple"
name: "Apple"
- id: "google"
name: "Google"
samples:
# ===== EASY - Clear global brands =====
- id: merch_easy_001
difficulty: easy
tags: [fast_food, global_brand]
input:
id: txn_001
amount: 12.99
classification: expense
description: "MCDONALD'S #12345 SPRINGFIELD IL"
expected:
business_name: "McDonald's"
business_url: "mcdonalds.com"
- id: merch_easy_002
difficulty: easy
tags: [coffee, global_brand]
input:
id: txn_002
amount: 5.75
classification: expense
description: "STARBUCKS STORE #9876"
expected:
business_name: "Starbucks"
business_url: "starbucks.com"
- id: merch_easy_003
difficulty: easy
tags: [ecommerce, global_brand]
input:
id: txn_003
amount: 89.99
classification: expense
description: "AMAZON.COM*1A2B3C4D AMZN.COM/BILL"
expected:
business_name: "Amazon"
business_url: "amazon.com"
- id: merch_easy_004
difficulty: easy
tags: [streaming, global_brand]
input:
id: txn_004
amount: 15.99
classification: expense
description: "NETFLIX.COM"
expected:
business_name: "Netflix"
business_url: "netflix.com"
- id: merch_easy_005
difficulty: easy
tags: [rideshare, global_brand]
input:
id: txn_005
amount: 23.50
classification: expense
description: "UBER *TRIP HELP.UBER.COM"
expected:
business_name: "Uber"
business_url: "uber.com"
- id: merch_easy_006
difficulty: easy
tags: [streaming, global_brand]
input:
id: txn_006
amount: 14.99
classification: expense
description: "SPOTIFY USA"
expected:
business_name: "Spotify"
business_url: "spotify.com"
- id: merch_easy_007
difficulty: easy
tags: [retail, global_brand]
input:
id: txn_007
amount: 156.78
classification: expense
description: "TARGET #1234"
expected:
business_name: "Target"
business_url: "target.com"
- id: merch_easy_008
difficulty: easy
tags: [retail, global_brand]
input:
id: txn_008
amount: 234.56
classification: expense
description: "COSTCO WHSE #5678"
expected:
business_name: "Costco"
business_url: "costco.com"
- id: merch_easy_009
difficulty: easy
tags: [tech, global_brand]
input:
id: txn_009
amount: 9.99
classification: expense
description: "APPLE.COM/BILL"
expected:
business_name: "Apple"
business_url: "apple.com"
- id: merch_easy_010
difficulty: easy
tags: [tech, global_brand]
input:
id: txn_010
amount: 12.99
classification: expense
description: "GOOGLE *STORAGE"
expected:
business_name: "Google"
business_url: "google.com"
- id: merch_easy_011
difficulty: easy
tags: [fast_food, global_brand]
input:
id: txn_011
amount: 8.50
classification: expense
description: "BURGER KING #456"
expected:
business_name: "Burger King"
business_url: "bk.com"
- id: merch_easy_012
difficulty: easy
tags: [fast_food, global_brand]
input:
id: txn_012
amount: 9.99
classification: expense
description: "TACO BELL #789"
expected:
business_name: "Taco Bell"
business_url: "tacobell.com"
- id: merch_easy_013
difficulty: easy
tags: [fast_food, global_brand]
input:
id: txn_013
amount: 11.50
classification: expense
description: "WENDYS #456"
expected:
business_name: "Wendy's"
business_url: "wendys.com"
- id: merch_easy_014
difficulty: easy
tags: [coffee, global_brand]
input:
id: txn_014
amount: 4.25
classification: expense
description: "DUNKIN #12345"
expected:
business_name: "Dunkin'"
business_url: "dunkindonuts.com"
- id: merch_easy_015
difficulty: easy
tags: [grocery, global_brand]
input:
id: txn_015
amount: 156.32
classification: expense
description: "WHOLE FOODS MKT #10234"
expected:
business_name: "Whole Foods Market"
business_url: "wholefoodsmarket.com"
- id: merch_easy_016
difficulty: easy
tags: [grocery, global_brand]
input:
id: txn_016
amount: 87.45
classification: expense
description: "TRADER JOE'S #567"
expected:
business_name: "Trader Joe's"
business_url: "traderjoes.com"
- id: merch_easy_017
difficulty: easy
tags: [gas, global_brand]
input:
id: txn_017
amount: 45.00
classification: expense
description: "SHELL OIL 573849234"
expected:
business_name: "Shell"
business_url: "shell.com"
- id: merch_easy_018
difficulty: easy
tags: [gas, global_brand]
input:
id: txn_018
amount: 52.30
classification: expense
description: "CHEVRON STATION #1234"
expected:
business_name: "Chevron"
business_url: "chevron.com"
- id: merch_easy_019
difficulty: easy
tags: [rideshare, global_brand]
input:
id: txn_019
amount: 18.75
classification: expense
description: "LYFT *RIDE SAT 7PM"
expected:
business_name: "Lyft"
business_url: "lyft.com"
- id: merch_easy_020
difficulty: easy
tags: [electronics, global_brand]
input:
id: txn_020
amount: 299.99
classification: expense
description: "BEST BUY 00000456"
expected:
business_name: "Best Buy"
business_url: "bestbuy.com"
# ===== MEDIUM - Less obvious merchants =====
- id: merch_medium_001
difficulty: medium
tags: [restaurant, chain]
input:
id: txn_021
amount: 67.50
classification: expense
description: "OLIVE GARDEN #456"
expected:
business_name: "Olive Garden"
business_url: "olivegarden.com"
- id: merch_medium_002
difficulty: medium
tags: [restaurant, chain]
input:
id: txn_022
amount: 45.00
classification: expense
description: "CHEESECAKE FACTORY"
expected:
business_name: "The Cheesecake Factory"
business_url: "thecheesecakefactory.com"
- id: merch_medium_003
difficulty: medium
tags: [pharmacy, chain]
input:
id: txn_023
amount: 24.99
classification: expense
description: "CVS/PHARMACY #4567"
expected:
business_name: "CVS Pharmacy"
business_url: "cvs.com"
- id: merch_medium_004
difficulty: medium
tags: [pharmacy, chain]
input:
id: txn_024
amount: 35.50
classification: expense
description: "WALGREENS #12345"
expected:
business_name: "Walgreens"
business_url: "walgreens.com"
- id: merch_medium_005
difficulty: medium
tags: [fitness, chain]
input:
id: txn_025
amount: 39.99
classification: expense
description: "PLANET FITNESS MONTHLY"
expected:
business_name: "Planet Fitness"
business_url: "planetfitness.com"
- id: merch_medium_006
difficulty: medium
tags: [airline, brand]
input:
id: txn_026
amount: 345.00
classification: expense
description: "UNITED AIRLINES 0162345678"
expected:
business_name: "United Airlines"
business_url: "united.com"
- id: merch_medium_007
difficulty: medium
tags: [hotel, brand]
input:
id: txn_027
amount: 189.00
classification: expense
description: "MARRIOTT HOTELS NYC"
expected:
business_name: "Marriott"
business_url: "marriott.com"
- id: merch_medium_008
difficulty: medium
tags: [retail, chain]
input:
id: txn_028
amount: 234.56
classification: expense
description: "WALMART SUPERCENTER #1234"
expected:
business_name: "Walmart"
business_url: "walmart.com"
- id: merch_medium_009
difficulty: medium
tags: [streaming]
input:
id: txn_029
amount: 6.99
classification: expense
description: "HULU LLC"
expected:
business_name: "Hulu"
business_url: "hulu.com"
- id: merch_medium_010
difficulty: medium
tags: [streaming]
input:
id: txn_030
amount: 8.99
classification: expense
description: "DISNEY PLUS"
expected:
business_name: "Disney+"
business_url: "disneyplus.com"
- id: merch_medium_011
difficulty: medium
tags: [clothing, brand]
input:
id: txn_031
amount: 89.99
classification: expense
description: "ZARA USA INC"
expected:
business_name: "Zara"
business_url: "zara.com"
- id: merch_medium_012
difficulty: medium
tags: [clothing, brand]
input:
id: txn_032
amount: 65.00
classification: expense
description: "H&M HENNES MAURITZ"
expected:
business_name: "H&M"
business_url: "hm.com"
- id: merch_medium_013
difficulty: medium
tags: [utility]
input:
id: txn_033
amount: 145.00
classification: expense
description: "XFINITY INTERNET"
expected:
business_name: "Xfinity"
business_url: "xfinity.com"
- id: merch_medium_014
difficulty: medium
tags: [telecom]
input:
id: txn_034
amount: 89.00
classification: expense
description: "AT&T WIRELESS"
expected:
business_name: "AT&T"
business_url: "att.com"
- id: merch_medium_015
difficulty: medium
tags: [telecom]
input:
id: txn_035
amount: 112.00
classification: expense
description: "VERIZON WIRELESS"
expected:
business_name: "Verizon"
business_url: "verizon.com"
- id: merch_medium_016
difficulty: medium
tags: [ecommerce]
input:
id: txn_036
amount: 45.00
classification: expense
description: "ETSY.COM"
expected:
business_name: "Etsy"
business_url: "etsy.com"
- id: merch_medium_017
difficulty: medium
tags: [ecommerce]
input:
id: txn_037
amount: 89.00
classification: expense
description: "WAYFAIR*PURCHASE"
expected:
business_name: "Wayfair"
business_url: "wayfair.com"
- id: merch_medium_018
difficulty: medium
tags: [home, brand]
input:
id: txn_038
amount: 123.00
classification: expense
description: "IKEA US EAST LLC"
expected:
business_name: "IKEA"
business_url: "ikea.com"
- id: merch_medium_019
difficulty: medium
tags: [hotel]
input:
id: txn_039
amount: 234.00
classification: expense
description: "AIRBNB *HMQT5J6QQJ"
expected:
business_name: "Airbnb"
business_url: "airbnb.com"
- id: merch_medium_020
difficulty: medium
tags: [entertainment]
input:
id: txn_040
amount: 45.00
classification: expense
description: "AMC THEATRES #1234"
expected:
business_name: "AMC Theatres"
business_url: "amctheatres.com"
# ===== HARD - Delivery services, subsidiaries =====
- id: merch_hard_001
difficulty: hard
tags: [delivery, aggregator]
input:
id: txn_041
amount: 45.00
classification: expense
description: "DOORDASH*CHIPOTLE"
expected:
business_name: "DoorDash"
business_url: "doordash.com"
- id: merch_hard_002
difficulty: hard
tags: [delivery, aggregator]
input:
id: txn_042
amount: 67.00
classification: expense
description: "GRUBHUB*THAI KITCHEN"
expected:
business_name: "Grubhub"
business_url: "grubhub.com"
- id: merch_hard_003
difficulty: hard
tags: [delivery, aggregator]
input:
id: txn_043
amount: 89.00
classification: expense
description: "UBEREATS *UBER EATS"
expected:
business_name: "Uber Eats"
business_url: "ubereats.com"
- id: merch_hard_004
difficulty: hard
tags: [delivery, grocery]
input:
id: txn_044
amount: 234.00
classification: expense
description: "INSTACART*SAFEWAY"
expected:
business_name: "Instacart"
business_url: "instacart.com"
- id: merch_hard_005
difficulty: hard
tags: [subscription, variant]
input:
id: txn_045
amount: 156.00
classification: expense
description: "AMAZON PRIME*1A2B3C"
expected:
business_name: "Amazon"
business_url: "amazon.com"
- id: merch_hard_006
difficulty: hard
tags: [payment_processor]
input:
id: txn_046
amount: 45.00
classification: expense
description: "SQ *DOWNTOWN CAFE"
expected:
business_name: null
business_url: null
- id: merch_hard_007
difficulty: hard
tags: [streaming, variant]
input:
id: txn_047
amount: 19.99
classification: expense
description: "HBO MAX"
expected:
business_name: "Max"
business_url: "max.com"
- id: merch_hard_008
difficulty: hard
tags: [tech, variant]
input:
id: txn_048
amount: 1299.00
classification: expense
description: "APPLE STORE #R123"
expected:
business_name: "Apple"
business_url: "apple.com"
- id: merch_hard_009
difficulty: hard
tags: [food, variant]
input:
id: txn_049
amount: 34.50
classification: expense
description: "PANERA BREAD #567"
expected:
business_name: "Panera Bread"
business_url: "panerabread.com"
- id: merch_hard_010
difficulty: hard
tags: [convenience]
input:
id: txn_050
amount: 12.50
classification: expense
description: "7-ELEVEN #34567"
expected:
business_name: "7-Eleven"
business_url: "7-eleven.com"
# ===== EDGE CASES - Should return null =====
- id: merch_edge_001
difficulty: edge_case
tags: [should_be_null, generic]
input:
id: txn_051
amount: 15.00
classification: expense
description: "POS DEBIT 12345"
expected:
business_name: null
business_url: null
- id: merch_edge_002
difficulty: edge_case
tags: [should_be_null, generic]
input:
id: txn_052
amount: 100.00
classification: expense
description: "ACH WITHDRAWAL"
expected:
business_name: null
business_url: null
- id: merch_edge_003
difficulty: edge_case
tags: [should_be_null, payment]
input:
id: txn_053
amount: 78.00
classification: expense
description: "PAYPAL *JOHNSMITH"
expected:
business_name: null
business_url: null
- id: merch_edge_004
difficulty: edge_case
tags: [should_be_null, transfer]
input:
id: txn_054
amount: 500.00
classification: expense
description: "ONLINE TRANSFER TO CHK 1234"
expected:
business_name: null
business_url: null
- id: merch_edge_005
difficulty: edge_case
tags: [should_be_null, atm]
input:
id: txn_055
amount: 200.00
classification: expense
description: "ATM WITHDRAWAL 12345"
expected:
business_name: null
business_url: null
- id: merch_edge_006
difficulty: edge_case
tags: [should_be_null, generic]
input:
id: txn_056
amount: 75.00
classification: expense
description: "MISC SERVICES LLC"
expected:
business_name: null
business_url: null
- id: merch_edge_007
difficulty: edge_case
tags: [should_be_null, wire]
input:
id: txn_057
amount: 1500.00
classification: expense
description: "WIRE TRANSFER OUT"
expected:
business_name: null
business_url: null
- id: merch_edge_008
difficulty: edge_case
tags: [should_be_null, check]
input:
id: txn_058
amount: 350.00
classification: expense
description: "CHECK #1234"
expected:
business_name: null
business_url: null
- id: merch_edge_009
difficulty: edge_case
tags: [should_be_null, fee]
input:
id: txn_059
amount: 35.00
classification: expense
description: "SERVICE CHARGE"
expected:
business_name: null
business_url: null
- id: merch_edge_010
difficulty: edge_case
tags: [should_be_null, p2p]
input:
id: txn_060
amount: 250.00
classification: expense
description: "ZELLE TO JOHN DOE"
expected:
business_name: null
business_url: null
# Additional samples
- id: merch_easy_021
difficulty: easy
tags: [fast_food]
input:
id: txn_061
amount: 14.99
classification: expense
description: "CHIPOTLE ONLINE"
expected:
business_name: "Chipotle"
business_url: "chipotle.com"
- id: merch_easy_022
difficulty: easy
tags: [fast_food]
input:
id: txn_062
amount: 8.99
classification: expense
description: "SUBWAY #12345"
expected:
business_name: "Subway"
business_url: "subway.com"
- id: merch_easy_023
difficulty: easy
tags: [coffee]
input:
id: txn_063
amount: 6.50
classification: expense
description: "PEETS COFFEE #456"
expected:
business_name: "Peet's Coffee"
business_url: "peets.com"
- id: merch_easy_024
difficulty: easy
tags: [grocery]
input:
id: txn_064
amount: 145.67
classification: expense
description: "KROGER #456"
expected:
business_name: "Kroger"
business_url: "kroger.com"
- id: merch_easy_025
difficulty: easy
tags: [grocery]
input:
id: txn_065
amount: 98.34
classification: expense
description: "PUBLIX SUPER MARKET"
expected:
business_name: "Publix"
business_url: "publix.com"
- id: merch_medium_021
difficulty: medium
tags: [restaurant]
input:
id: txn_066
amount: 123.45
classification: expense
description: "RUTH'S CHRIS STEAK"
expected:
business_name: "Ruth's Chris Steak House"
business_url: "ruthschris.com"
- id: merch_medium_022
difficulty: medium
tags: [restaurant]
input:
id: txn_067
amount: 89.00
classification: expense
description: "P.F. CHANGS #234"
expected:
business_name: "P.F. Chang's"
business_url: "pfchangs.com"
- id: merch_medium_023
difficulty: medium
tags: [gas]
input:
id: txn_068
amount: 48.50
classification: expense
description: "EXXONMOBIL 12345"
expected:
business_name: "ExxonMobil"
business_url: "exxon.com"
- id: merch_medium_024
difficulty: medium
tags: [gas]
input:
id: txn_069
amount: 55.00
classification: expense
description: "BP#1234567"
expected:
business_name: "BP"
business_url: "bp.com"
- id: merch_medium_025
difficulty: medium
tags: [hotel]
input:
id: txn_070
amount: 245.00
classification: expense
description: "HILTON HOTELS"
expected:
business_name: "Hilton"
business_url: "hilton.com"
- id: merch_hard_011
difficulty: hard
tags: [payment_processor]
input:
id: txn_071
amount: 34.00
classification: expense
description: "VENMO *PIZZA PLACE"
expected:
business_name: null
business_url: null
- id: merch_hard_012
difficulty: hard
tags: [subscription]
input:
id: txn_072
amount: 14.99
classification: expense
description: "APPLE.COM/BILL ONE"
expected:
business_name: "Apple"
business_url: "apple.com"
- id: merch_hard_013
difficulty: hard
tags: [tech]
input:
id: txn_073
amount: 99.99
classification: expense
description: "MICROSOFT*OFFICE 365"
expected:
business_name: "Microsoft"
business_url: "microsoft.com"
- id: merch_hard_014
difficulty: hard
tags: [gaming]
input:
id: txn_074
amount: 59.99
classification: expense
description: "STEAMPOWERED.COM"
expected:
business_name: "Steam"
business_url: "steampowered.com"
- id: merch_hard_015
difficulty: hard
tags: [subscription]
input:
id: txn_075
amount: 10.99
classification: expense
description: "YOUTUBE PREMIUM"
expected:
business_name: "YouTube"
business_url: "youtube.com"
- id: merch_easy_026
difficulty: easy
tags: [airline]
input:
id: txn_076
amount: 456.00
classification: expense
description: "DELTA AIR LINES"
expected:
business_name: "Delta Air Lines"
business_url: "delta.com"
- id: merch_easy_027
difficulty: easy
tags: [airline]
input:
id: txn_077
amount: 389.00
classification: expense
description: "SOUTHWEST AIRLINES"
expected:
business_name: "Southwest Airlines"
business_url: "southwest.com"
- id: merch_easy_028
difficulty: easy
tags: [hotel]
input:
id: txn_078
amount: 178.00
classification: expense
description: "HYATT REGENCY"
expected:
business_name: "Hyatt"
business_url: "hyatt.com"
- id: merch_medium_026
difficulty: medium
tags: [fitness]
input:
id: txn_079
amount: 29.99
classification: expense
description: "LA FITNESS CLUB"
expected:
business_name: "LA Fitness"
business_url: "lafitness.com"
- id: merch_medium_027
difficulty: medium
tags: [fitness]
input:
id: txn_080
amount: 49.99
classification: expense
description: "ORANGETHEORY FITNESS"
expected:
business_name: "Orangetheory Fitness"
business_url: "orangetheory.com"
- id: merch_edge_011
difficulty: edge_case
tags: [should_be_null]
input:
id: txn_081
amount: 1.00
classification: expense
description: "PENDING AUTHORIZATION"
expected:
business_name: null
business_url: null
- id: merch_edge_012
difficulty: edge_case
tags: [should_be_null]
input:
id: txn_082
amount: 0.00
classification: expense
description: "VOID TRANSACTION"
expected:
business_name: null
business_url: null
- id: merch_medium_028
difficulty: medium
tags: [entertainment]
input:
id: txn_083
amount: 89.00
classification: expense
description: "TICKETMASTER *EVENT"
expected:
business_name: "Ticketmaster"
business_url: "ticketmaster.com"
- id: merch_medium_029
difficulty: medium
tags: [entertainment]
input:
id: txn_084
amount: 150.00
classification: expense
description: "STUBHUB INC"
expected:
business_name: "StubHub"
business_url: "stubhub.com"
- id: merch_medium_030
difficulty: medium
tags: [car_rental]
input:
id: txn_085
amount: 56.00
classification: expense
description: "HERTZ RENT-A-CAR"
expected:
business_name: "Hertz"
business_url: "hertz.com"
- id: merch_easy_029
difficulty: easy
tags: [grocery]
input:
id: txn_086
amount: 178.90
classification: expense
description: "SAM'S CLUB #8765"
expected:
business_name: "Sam's Club"
business_url: "samsclub.com"
- id: merch_easy_030
difficulty: easy
tags: [pharmacy]
input:
id: txn_087
amount: 67.89
classification: expense
description: "RITE AID #1234"
expected:
business_name: "Rite Aid"
business_url: "riteaid.com"
- id: merch_hard_016
difficulty: hard
tags: [aggregator]
input:
id: txn_088
amount: 234.00
classification: expense
description: "COSTCO.COM"
expected:
business_name: "Costco"
business_url: "costco.com"
- id: merch_hard_017
difficulty: hard
tags: [fitness]
input:
id: txn_089
amount: 150.00
classification: expense
description: "EQUINOX MEMBERSHIP"
expected:
business_name: "Equinox"
business_url: "equinox.com"
- id: merch_hard_018
difficulty: hard
tags: [subscription]
input:
id: txn_090
amount: 4.99
classification: expense
description: "PARAMOUNT+ ESSENTIAL"
expected:
business_name: "Paramount+"
business_url: "paramountplus.com"
Reference in New Issue View Git Blame Copy Permalink