Dataset Viewer
Auto-converted to Parquet Duplicate
crm_customer_id
stringlengths
8
8
client_id
stringlengths
8
8
match_method
stringclasses
3 values
FRF-6340
FRF-6340
primary
TOB-2160
TOB-2160
primary
PAB-6120
PAB-6120
primary
ELM-7580
ELM-7580
primary
BRW-6580
BRW-6580
primary
KAZ-7480
KAZ-7480
primary
YAY-9440
YAY-9440
primary
BOV-6320
BOV-6320
primary
ALD-1320
ALD-1320
primary
NIB-7940
NIB-7940
primary
REH-9240
REH-9240
primary
TAG-6100
TAG-6100
primary
JAC-1560
JAC-1560
primary
SYB-6000
SYB-6000
primary
HER-2080
HER-2080
primary
EMT-9760
EMT-9760
primary
YUV-6700
YUV-6700
primary
PIK-5220
PIK-5220
primary
JEP-2860
JEP-2860
primary
BEG-6860
BEG-6860
primary
JUG-2220
JUG-2220
primary
CHG-3680
CHG-3680
primary
MAD-5980
MAD-5980
primary
CED-5620
CED-5620
primary
JAJ-7200
JAJ-7200
primary
LOP-9360
LOP-9360
primary
ANH-2960
ANH-2960
primary
MAP-9880
MAP-9880
primary
JOD-9840
JOD-9840
primary
PAB-1920
PAB-1920
primary
LEG-9440
LEG-9440
primary
MIW-9680
MIW-9680
primary
NIM-6140
NIM-6140
primary
COF-5460
COF-5460
primary
ALB-3420
ALB-3420
primary
JAM-8780
JAM-8780
primary
JOW-5240
JOW-5240
primary
EVH-2500
EVH-2500
primary
GRR-5260
GRR-5260
primary
MAM-3640
MAM-3640
primary
FAK-5360
FAK-5360
primary
ALK-2920
ALK-2920
primary
LOB-3320
LOB-3320
primary
STD-9400
STD-9400
primary
ALS-4640
ALS-4640
primary
JOP-1680
JOP-1680
primary
JEB-5200
JEB-5200
primary
COC-5700
COC-5700
primary
SEC-4380
SEC-4380
primary
OLM-4100
OLM-4100
primary
CHC-9380
CHC-9380
primary
JAH-7200
JAH-7200
primary
FRA-1340
FRA-1340
primary
LAA-6960
LAA-6960
primary
YEK-4940
YEK-4940
primary
ALM-3600
ALM-3600
primary
KIL-7220
KIL-7220
primary
ISC-7600
ISC-7600
primary
DAN-2760
DAN-2760
primary
REM-6480
REM-6480
primary
JAS-4300
JAS-4300
primary
POS-6220
POS-6220
primary
CHS-9020
CHS-9020
primary
KHA-7080
KHA-7080
primary
SAN-5360
SAN-5360
primary
TEY-6660
TEY-6660
primary
CLK-6160
CLK-6160
primary
ROP-2980
ROP-2980
primary
JAG-9140
JAG-9140
primary
SHD-3760
SHD-3760
primary
HAR-7060
HAR-7060
primary
RAL-2640
RAL-2640
primary
COH-5500
COH-5500
primary
TEM-6380
TEM-6380
primary
SVD-6180
SVD-6180
primary
AMD-9440
AMD-9440
primary
VEM-8980
VEM-8980
primary
MAD-1920
MAD-1920
primary
RAW-5800
RAW-5800
primary
SAK-8400
SAK-8400
primary
ZOL-5960
ZOL-5960
primary
ROL-5580
ROL-5580
primary
DUS-4200
DUS-4200
primary
LAD-6660
LAD-6660
primary
RIJ-4760
RIJ-4760
primary
DAT-6940
DAT-6940
primary
ANS-2740
ANS-2740
primary
HAW-7900
HAW-7900
primary
FRH-4500
FRH-4500
primary
KAL-5760
KAL-5760
primary
GLS-8960
GLS-8960
primary
MID-9940
MID-9940
primary
BIP-1000
BIP-1000
primary
IVD-6820
IVD-6820
primary
BRM-9040
BRM-9040
primary
ANB-9300
ANB-9300
primary
JAF-6380
JAF-6380
primary
BEC-7980
BEC-7980
primary
JOD-7780
JOD-7780
primary
CLS-1800
CLS-1800
primary
End of preview. Expand in Data Studio

Collections Compass: inspectable generated data

Coffeebeans · Team Members: Ayush Pattanayak and Pranshu Porwal

We built Collections Compass for the CIBC Collections Hackathon. This repository contains our current generated publications and quality evidence, kept outside GitHub because of their size. Every person, account, call and source document in the supplied Maple Bank release is synthetic. These are educational prototype artifacts, not CIBC customer data or operating systems.

Start here

Open the dataset viewer for verified_accounts to inspect 1,494,930 account-snapshot rows with verified ownership, original source keys, product/currency separation and lineage. Switch to identity_links to inspect unresolved and conflicting matches. We preserve uncertainty rather than treating every candidate as an owner.

Output Rows Meaning
Verified accounts 1,494,930 Owned account snapshots; balances remain separate by product and currency
Identity links 1,844,500 Verified, review-required, unresolved and conflicting links
CRM aliases 1,020,000 Source CRM IDs and conservatively resolved client IDs
Payment evidence 12,397 Selected promise-matched/reversal evidence from the supplied transaction release

The viewer calls each exported table a train split for browsing compatibility. These are inspection tables, not a model-training claim. The 1,015,228 resolved alias IDs are not a claim of perfectly matching all one million synthetic people.

What we generated

curated/ contains the exact three databases published by our build: Client 360, analytics and matched payments. Their JSON pointers preserve accepted versions and SHA-256 fingerprints. inventory.json records all tables, row counts, schemas and the corresponding code commit.

The C360 publication has 1,494,930 verified account links. We exclude 349,570 other account-snapshot rows from automatic owned totals: 43,720 need review, 293,772 are unresolved and 12,078 conflict. Those are account-snapshot rows, not distinct customer counts. Agreement scores are field-agreement measures, not calibrated probabilities.

Payment evidence comes from scanning all 12,401,924 supplied transaction rows across 220 partitions. Completeness applies to matching and reversals within the sampled release, not the complete bank ledger. Current analytics includes outcome fields for retrospective questions; those fields are not decision-feature inputs.

Inspect with DuckDB

Download a current database from curated/, then open it read-only:

import duckdb

db = duckdb.connect("curated/c360-20261003T072538295000Z.duckdb", read_only=True)
print(db.execute("SELECT status, count(*) FROM identity_links GROUP BY status").fetchall())
print(db.execute("SELECT system, product_type, currency, count(*) FROM accounts GROUP BY ALL").fetchall())

The application expects these database files and pointer JSONs together in its ignored data/curated/ directory. Other flows still need the organiser's released policy/transcript files. This artifact repository is not a standalone hosted application.

Evidence and limitations

contracts/c360.json extends DC-COLL-001. evidence/data-quality.md states affected-record counts and handling. The sample JSON contains actual released-client application snapshots. The final engineering run passed 1,044 checks; the 50-row benchmark was generated with no manual answer edits. Format validation does not establish factual grading. The semantic-review receipt separately records five independently reproduced numeric extras and known reference limitations.

The business snapshot is 28 September 2026. Prototype ingestion time does not reconstruct historical bank knowledge. A deposit balance is not repayment capacity. Hardship and other protective information cannot be used to prioritise recovery. No recovery uplift or preparation-time improvement has been measured.

Source: the organiser-supplied, fully synthetic Maple Bank Collections release. Our derived outputs preserve that provenance; this upload does not grant additional rights in the supplied source material. Raw release archives, obsolete publications, local review history, provider caches, credentials and audio are excluded.

Code: https://github.com/Pranshu00/coffeebeans-collections-compass (competition repository; access may be required). Code revision: 2475c130778d789d6a8b6466773de4052d90cb1d. Check manifest.json for exact sizes and SHA-256 hashes.

Downloads last month
-