stripeddonkey-data/lagos-synthetic-telecom-longitudinal-journeys

Lagos Synthetic Longitudinal Telecom Journeys — V1

A fully synthetic event-level telecom dataset for longitudinal reasoning, lifecycle analysis, usage modeling, recharge/bundle behavior, mobility, device/SIM history and AI evaluation.

Primary research question: How do subscriber lifecycle state, network usage, recharge and bundle spending, mobility, SIM/device replacement and tariff changes co-evolve through time?

Generated from the Fidelra City Twin Engine (FCTE) for STRIPED DONKEY.

No row represents a real person, real subscriber, real phone account, real SIM, real device, or real telecom transaction.

At a glance

Metric	Value
Sample subscriptions	500
Synthetic citizens	500
Longitudinal telecom events	14,424
Event domains	7
Event types	15
Earliest event	2016-09-06
Latest event	2026-08-31
Modeled snapshot date	2026-09-01
Geography modeled	Lagos, Nigeria
Currency for monetary events	NGN
License	CC BY-NC 4.0

What makes this dataset different?

This is not a flat subscriber table. It preserves connected synthetic telecom histories:

activation → lifecycle transitions → SIM/device episodes → tariff changes → monthly usage → airtime top-ups → data-bundle purchases → mobility → service events → account closure

Files

telecom_events

Split	Rows
Train	11,473
Validation	1,545
Test	1,406

subscriber_profiles

Split	Rows
Train	398
Validation	52
Test	50

Featured companion journey

The visual uses Musa Odejimi, a fully synthetic journey selected deterministically from the public sample for maximum cross-domain coverage.

File	Rows
companion/featured_journey_profile_musa_odejimi.csv	1
companion/featured_journey_events_musa_odejimi.csv	45
companion/featured_journey_kpis_musa_odejimi.csv	1

This companion is a convenience view of an existing TRAIN subscription, not an additional observation. Do not append it to the train/validation/test files.

Research split design

The 500 subscriptions are selected with a deterministic simple-random hash ordering from eligible citizen-linked Telecom subscriptions. The sample is not richness-ranked and not outcome-stratified.

ResearchSplit is assigned by synthetic citizen ID, so multiple subscriptions belonging to the same synthetic citizen cannot cross train, validation and test.

Snapshot operator mix in this synthetic research sample

Operator	Subscriptions	Share
MTN	259	51.8%
Airtel	169	33.8%
Globacom	63	12.6%
9mobile / T2	9	1.8%

These are properties of the synthetic sample, not estimates of real operator market share.

Temporal semantics

Most records are dated point events (TemporalGrain = POINT).

MONTHLY_USAGE is different: it is a monthly aggregate period, not proof that activity occurred on the first day of the month.

For monthly rows:

• TemporalGrain = MONTH
• PeriodStart = first day of the month
• PeriodEnd = last day of the month

Snapshot versus historical state

Current mutable subscriber fields are confined to subscriber_profiles and are explicitly prefixed with Snapshot*.

Do not project snapshot values backward into historical event rows.

The profile file is outcome-complete/current-state context. For prospective modeling at an earlier date, use only information available by that date.

Monetary semantics

AmountLocal has event-specific meaning:

Event type	Meaning
MONTHLY_USAGE	Billed usage amount
AIRTIME_TOPUP	Airtime top-up amount
DATA_BUNDLE_PURCHASE	Bundle amount paid
Other events	Null

Tariff monthly fee is plan metadata, not proof of payment.

Event domains

The release covers:

• DEVICE
• ECONOMY
• LIFECYCLE
• MOBILITY
• NETWORK
• SERVICE
• TARIFF

Load with Hugging Face Datasets

from datasets import load_dataset

events = load_dataset(
    "stripeddonkey-data/lagos-synthetic-telecom-longitudinal-journeys",
    "telecom_events",
)

profiles = load_dataset(
    "stripeddonkey-data/lagos-synthetic-telecom-longitudinal-journeys",
    "subscriber_profiles",
)

Load with pandas

import pandas as pd

events = pd.read_csv("data/telecom_events/train.csv")

subscriber_id = events["SyntheticSubscriberId"].iloc[0]

journey = (
    events[events["SyntheticSubscriberId"] == subscriber_id]
    .sort_values(["EventUtc", "EventOrder"])
)

print(
    journey[
        [
            "SyntheticSubscriberId",
            "EventDate",
            "EventDomain",
            "EventType",
            "EventLabel",
        ]
    ]
)

Suggested research tasks

• subscriber lifecycle sequence modeling;
• temporal reasoning and event ordering;
• recharge and data-bundle behavior;
• network-usage sequence modeling;
• tariff-transition analysis;
• SIM/device replacement research;
• mobility-linked telecom behavior;
• customer-service event analysis;
• graph and RAG experiments using source attribution;
• synthetic-data engineering and temporal-leakage education.

Important limitations

This dataset is synthetic. It is not an operator extract, billing-system export, CDR dataset, regulatory filing, or survey of real Lagos subscribers.

Do not use it to:

• estimate real operator market share;
• infer real adoption, churn, ARPU, mobility or spending rates;
• identify or make claims about real people;
• infer the behavior of a real operator or customer;
• substitute for official operator, regulator or census statistics.

Synthetic distributions are properties of the model, not empirical population estimates.

Identity and phone-number policy

All people and accounts are synthetic.

Public identifiers are explicitly named SyntheticCitizenId, SyntheticSubscriberId, SyntheticHolderCitizenId and SyntheticSourceRecordId.

dbo.Citizen.PhoneNumber and Telecom MSISDN are not included because they add no analytical value to this public research sample.

Validation

The executed source export returned 10/10 PASS with zero failures.

Independent packaging validation also returned zero defects across event ordering, profile/event identity consistency, age calculations, lifecycle boundaries, split isolation, monetary fields and source attribution.

See:

• qa/VALIDATION_REPORT.json
• qa/validation_summary.md
• qa/quality_metrics.json

Packaging note

The SQL Server results-to-text report truncated the optional EventDetailJson field at 256 characters on 1,399 mobility rows. The public CSV files therefore omit that redundant field rather than publishing malformed JSON.

No event rows were removed.

Featured synthetic journey

Musa Odejimi — SyntheticCitizenId 13,610,885 / SyntheticSubscriberId 30,271

This featured journey has 45 events across 7 domains and 11 event types and is a convenience view of the existing TRAIN sample row.

About STRIPED DONKEY

STRIPED DONKEY publishes connected synthetic city data generated by the Fidelra City Twin Engine (FCTE).

Website: https://stripeddonkey.com

Synthetic people. Connected data. Brighter cities.

Version

• Dataset: Lagos Synthetic Longitudinal Telecom Journeys — V1
• Dataset version: LOS-TEL-HF-V1
• Release date: 2026-10-06
• Modeled snapshot date: 2026-09-01
• Public research sample: 500 subscriptions / 500 synthetic citizens / 14,424 events

Citation

@dataset{striped_donkey_lagos_synthetic_telecom_longitudinal_2026,
  author    = {STRIPED DONKEY / Fidelra},
  title     = {Lagos Synthetic Longitudinal Telecom Journeys — V1},
  year      = {2026},
  publisher = {STRIPED DONKEY},
  url       = {https://huggingface.co/datasets/stripeddonkey-data/lagos-synthetic-telecom-longitudinal-journeys}
}

License

This dataset is licensed under the Creative Commons Attribution-NonCommercial 4.0 International License (CC BY-NC 4.0).

Commercial use requires separate permission from Fidelra / STRIPED DONKEY.

See LICENSE.md.