Construction change event reasoning tasks, 14 general contractors

Summary

Long-context tasks drawn from the project records of 14 US commercial general contractors. Each task gathers the RFIs, daily logs, meeting minutes and email behind one change event and asks whether the owner accepted it, for what amount and for how many days, with the answer taken from the executed change order. Projects run 8 to 30 months, so many tasks need reasoning across hundreds of documents written by people who disagreed. Owner, subcontractor, staff and site names, addresses and phone numbers are replaced with placeholders.

Collected to order. We collect it from US companies for your order. Scope, volume and timing are confirmed in writing before you sign.

Pricing on request

Request scope and pricing

At a glance

Planned scope. Volume, sources and timing are confirmed in writing before you sign.

Records: 9,400graded tasks
Size: 1.5 GB
Tokens, estimated: 380M
Years: 2015 to 2026
Firms: 14

Owner profile: 14 US commercial general contractors, 25 to 220 staff each, building K-12, municipal, light industrial and tenant improvement work in the Midwest, Mid-Atlantic and Pacific Northwest
Industry: Construction
Data type: Evaluation and task packs
Headcount: 25 to 220 staff per firm
Region: United States (Midwest, Mid-Atlantic, Pacific Northwest)
Languages: English, Spanish
Formats: JSONL
Updated: 2026-10-08

Sources

System	Kind	Records
Procore	Projects and wiki	58K RFIs, 17K change events
Microsoft 365 mail	Email	4.6M project emails
Shared drives	Shared files	290K daily logs, minutes and submittals

Good for

evaluating long-context reasoning across multi-month project records
training models to assess the cost and schedule impact of change events
testing entitlement analysis where the parties disagree in writing
retrieval over RFIs, daily logs and meeting minutes

Fields

Field	Type	Description
task_id	string	Stable task id
project_profile	object	Building type, contract type and duration
change_event	string	Change event id and title
documents	array<string>	The event's RFIs, logs, minutes and emails in date order, excerpted in this sample
question	string	The question put to the model
answer	object	Accepted or not, approved amount and schedule days
answer_basis	string	The record the answer comes from
grader	enum	field_match or numeric_tolerance
context_tokens	int	Size of the full task context

Example record

{
  "task_id": "T-54096-002871",
  "project_profile": {
    "type": "K-12 classroom addition",
    "contract": "GMP with shared savings",
    "duration_months": 19
  },
  "change_event": "[CE_0147] Rock at elevator pit excavation",
  "documents": [
    "RFI 112, 2024-03-05: Unforeseen rock at elevator pit. Geotech boring B-4 showed stiff clay to 14 ft.",
    "Daily log 2024-03-06: Excavator idle 5 hrs, hydraulic breaker mobilized by [COMPANY_6].",
    "OAC minutes 2024-03-12, item 7.3: Owner rep reserves position on differing site condition.",
    "Email 2024-03-19 from [PERSON_22]: T&M tickets for rock removal attached, 3 days to date."
  ],
  "question": "Did the owner accept the rock as a differing site condition, and for what amount and how many days?",
  "answer": {
    "accepted": true,
    "approved_amount_usd": 38450,
    "schedule_days": 6
  },
  "answer_basis": "Executed change order [CO_031], 2024-04-22",
  "grader": "field_match",
  "context_tokens": 41870
}

One record, de-identified. Placeholders such as [PERSON_12] and [COMPANY_4] stay the same across the whole dataset, so threads still read correctly.

Licenses available

Exclusive
One buyer. Nobody else can license it during the exclusive term, usually 2 years.
Non-exclusive
Licensed to more than one buyer.
First look
One buyer gets 2 to 4 weeks to make an offer before anyone else sees the dataset.

What ships with it

Datasheet: what it is, what it is good for, verified size, sources, years, format and limits.
Provenance file: the owner shown by code, staff notice and consent status, exclusions, mapped to the EU AI Act Article 53 training-data summary and California AB 2013.
Manifest with a hash for every file.
QC report and de-identification report.
Your own copy, with unique canary records.

How it is de-identified

Default exclusions removed, files flattened to text, identifiers found in three layers and replaced with consistent placeholders, and the result measured against a 99% bar for each identifier type in each source.

The full method