Nigeria Machinery: A Low-Resource Industrial Dataset with a Domain-Grounded Reasoning Layer

Jul 16, 2026

Industrial machinery data for African economies remains scarce and not readily model-ready, limiting quantitative analysis and training of language models on numeric tasks grounded in local industrial realities. This work releases the Nigeria Machinery Usage and Failures Dataset with 89 machine-level records, 28 indicators, and 2006–2025 coverage, fully sourced and decoded by a codebook. It also provides a method to create chain-of-thought reasoning prompts from sparse numeric values, plus provenance and reproducible evaluation tooling.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages10

Reading Time25 min

Abstract

Introduction

Related Work

Dataset Construction and Release

Domain-Grounded Reasoning Layer Method

Evaluation Tooling and Verification

Appendix B

Nigeria Machinery: A Low-Resource Industrial Dataset with a Domain-Grounded Reasoning Layer Gospel Bassey1, Vincent Fakiyesi2 1. Independent Researcher [rexkindy@gmail.com](rexkindy@gmail.com) 2. Engineering Education Transformations Institute, College of Engineering, University of Georgia. [Vincent.Fakiyesi@uga.edu](Vincent.Fakiyesi@uga.edu) Dataset adaptation by Adaption Labs ([https://adaptionlabs.ai](https://adaptionlabs.ai)) Abstract There is relatively little, public, and model-ready data on industrial machinery for African economies. This makes it hard to do quantitative analysis or to train language models on numeric tasks grounded in that setting. We release two things to help with part of this problem. The first is the Nigeria Machinery Usage and Failures Dataset: 89 machine-level records across 28 indicators, covering Nigeria's manufacturing and oil and gas sectors from 2006 to 2025. Every record names a public source and is decoded by a codebook. The second is a method for building chain-of-thought (CoT) reasoning examples from these sparse numeric values. The result is 94 prompt, completion, and reasoning-trace rows. In every row, the prompt names the real indicator, subsector, year, and source of the record it comes from. The data adaptation work was carried out by Adaption Labs. Along the way we describe a problem that is common when language models are used to build datasets. The prompts can match the real numbers while saying nothing about the real domain. We show that fixing this raises the share of domain-grounded prompts from 1 out of 78 in an earlier release to 94 out of 94, and that every retrieval answer now matches its source value (84 out of 84) . We release the data, the reasoning layer, and a per-row provenance file under CC-BY-4.0. We are clear about the limits. With 89 records and 17 indicators that have only one observation, this is a reference and seed dataset, not a large training set. Most reasoning rows are retrieval rather than multistep computation. 1. Introduction Predictive maintenance and fault diagnosis depend on machine-level data: failure counts, downtime, capacity utilization, and maintenance spend, recorded over time. For most of Nigeria's industrial base, this data is not available in any form a model can use. The figures exist, but they are locked inside annual reports as static tables, held in regulator filings, or buried in national statistics that aggregate away everything specific. An analyst cannot easily see how a single refinery performed across a decade, only what a whole sector reported in a given year. This is a consequential gap. Nigeria is the largest oil producer in Africa, and the sector provides most government revenue and most foreign exchange earnings. When equipment fails, the effects reach past the plant into energy prices and public budgets. Yet there is almost no public, machine-readable record of how that failure unfolds over time, which means the data-driven methods now standard in industrial machine learning cannot be applied to the region at all. The problem is not only that data is scarce. It is also that the scarcity has a particular shape. The numbers that exist are scattered, inconsistently reported, and never assembled into a single structured resource. Building one requires manual collection from many separate official documents, careful crosschecking, and a consistent encoding so that a model can ingest it. This is slow, unglamorous work, and it is the reason the resource does not yet exist. This paper takes a first step. We make three contributions: 1. We construct and release the Nigeria Machinery Usage and Failures Dataset, a structured, fullysourced collection of 89 machine-level industrial records for Nigeria spanning 2006 to 2025, assembled by hand from official government and industry sources (Section 3) . 2. We describe a method for turning these sparse tabular values into domain-grounded chain-ofthought reas

FAQs

What problem does the Nigeria Machinery dataset address?

It targets the lack of public, machine-readable industrial machinery data in Nigeria, where relevant figures exist mainly in static tables across reports and aggregated statistics that are hard to structure for models.

What does the dataset provide in terms of records and coverage?

It releases 89 machine-level records across 28 indicators, covering Nigeria’s manufacturing and oil-and-gas sectors from 2006 to 2025, with each record linked to a public source and decoded via a codebook.

How does the work create domain-grounded chain-of-thought reasoning examples?

It provides a method that turns sparse numeric table values into prompt–completion–reasoning-trace rows that explicitly name the real indicator, subsector, year, and record source, addressing a grounding failure mode where prompts match numbers without real domain grounding.

Jul 16, 2026

Upload by Theodore