Dataset Sourcing Engineer: Assemble License-Cleared, Pre-2020 Human Text Corpus

We are assembling a large corpus of human-written English text for a machine learning training dataset. The defining constraint is provenance: every piece of text must be verifiably human-written and cleared for commercial ML training. We need someone who can source, verify, and document text at volume, not write it or type it. WHAT THE WORK INVOLVES - Source English text written and published before 2020. The date is the proof of human authorship, so this cutoff is not negotiable. - Confirm each source's license actually permits commercial machine learning training. Publicly visible is not the same as licensed to train on. This is the core requirement of the role. - Document provenance for every dataset: original source, publication date, license terms, and a citation or link. - Match text to the specific domains we need (below), not just general prose. - Identify and remove text that is duplicated across many sites or datasets, so the corpus stays clean. - Deliver organized batches with a provenance sheet: one row per source, with date, license, and origin recorded. DOMAINS WE NEED (variety is essential) Product reviews, marketing copy, cryptocurrency and finance forum posts, customer support messages, academic writing, casual social media, and non-native (ESL) business English. The ability to find the harder, less-obvious registers is what separates a strong candidate from an average one. HARD REQUIREMENTS - Demonstrated experience assembling or curating datasets where source and licensing mattered. - Comfort reading and interpreting dataset licenses and terms of use (Creative Commons variants, dataset cards, archive terms). - A systematic, auditable method. We need a documented trail, not a bulk dump. - Clear written English for documenting sources. WHAT THIS IS NOT - Not data entry or transcription. - Not labeling, judging, or rating text. - Not building, training, or advising on the machine learning model. That work is handled separately; this role sources the input data only. - Please do not pitch AI strategy or model-building services. Proposals that do will be declined. TO APPLY, ANSWER BRIEFLY 1. Describe one corpus or dataset you have sourced where licensing and provenance were requirements. What was the source, and how did you confirm the rights to use it? 2. Name two or three sources or archives you would start with for pre-2020, license-cleared English text. 3. Confirm you understand the deliverable is sourced, documented, license-cleared text, with a pre-2020 date cutoff.