An open-source research project

Lineage

The history of America's infrastructure networks, line by line, with the historical citations to prove it.

Code on GitHub at github.com/line-age · Built by Tight Line Software with a university research group

A historical Oil and Gas Journal map of the United States with oil trunk pipe lines drawn in red, converging on the Mid-Continent fields and running east to the refineries.
Map of Oil Trunk Pipe Lines, The Oil and Gas Journal. David Rumsey Map Collection, David Rumsey Map Center, Stanford Libraries. CC BY-NC-SA 3.0.

What it is

Nobody knows when most of America's pipelines were built

The same goes for its power lines, water mains and older roads. Today's datasets show where these networks run, but rarely when each line was built, who owned it, or when it was retired. That history survives in scanned maps in archives and in a century of trade journals that reported each new line as it was laid.

Lineage recovers it as map data that opens in QGIS, the free GIS application researchers already use. Every fact, from a construction date to an owner, carries a citation to the page that says so. The first users are energy researchers studying how US energy distribution grew.

The name is a pun (line + age). It is also the GIS term for the record of a dataset's sources and processing, which is the whole idea.

What the data has to be

  1. Ready for standard GIS toolsopens in QGIS with no conversion and no plugins
  2. Dated, even when the date is fuzzy"sometime before 1930" is a real value
  3. Cited, fact by factwith a confidence level where one is known
  4. Accountable for every changewho changed what, when, and on what evidence

How it fits together

Eight components, one database in the middle

Sources Scanned maps Trade press, 1859-1961 Existing spreadsheets
Producers georef → trace corpus → cite backport
Holds it storenetwork + evidence
Consumers review
people check it
release
QGIS gets it

There is no master data processing pipeline. Each component reads records in one state and writes the next, and a human review is just a state change. Citation generation is the first priority; everything else can start in parallel.

Components

One repository each

cite

Priority · now

Finds the page that proves it. Retrieval and assertion happen per attribute, and every citation resolves at the moment the claim is made.

In
A segment and whatever is known about it
Out
Citations to specific issues and pages, one per claim

corpus

Feeds cite · now

Trade-press ingest from Internet Archive and an authority-ranked search over it, with era-keyed vocabulary: "gas works" in 1887, "pipe line" in 1937.

In
Digitized journal runs
Out
Ranked passages with identifiers that resolve

store

Now

The network and evidence model, PostGIS, and an audited CRUD API. One model for roads, power, oil, gas and water: segments, junctions, direction and flow.

In
Writes from every producer and from review
Out
An API, and PostGIS that QGIS opens directly

georef

Now

IIIF image serving, a self-hosted Allmaps editor, and the annotation server it needs. The editor doubles as the georeferencing review tool.

In
Scanned map images
Out
A georeference annotation with a permalink

trace

Research · now

Finding the lines on a georeferenced map, and not the county boundaries and railroads that look just like them. Conservative by design.

In
A georeferenced map and its symbology
Out
Candidate segments, unreviewed

backport

Now

Existing hand-made spreadsheets and traces, brought into the model, mess and all. The earliest real test of the data model.

In
Spreadsheets and traces
Out
Structured records in store

review

Winter

An authenticated app where people accept, correct or reject machine output. No machine-generated claim reaches a release without passing through it.

In
Candidate traces and machine citations
Out
Verdicts, with the reviewer in the audit trail

release

Winter

Versioned, citable builds of the dataset that play as a timeline in QGIS straight out of the box.

In
store
Out
GeoPackage and GeoJSON

House rules

What keeps it credible

  • No citation, no claim. Every identifier resolves when the claim is made, or the claim is thrown away.
  • Absent stays absent. A field nobody has a source for stays empty.
  • A 1930 map proves "by 1930". Not "built in 1930". Enforced in code.
  • The model proposes; a person decides. Especially whether two lines are the same one.

Roadmap

Three quarters

Fall 2026

Citations first, file in and file out. The corpus, the data model and API, georeferencing infrastructure, backporting and first vision experiments, all in parallel.

Winter 2027

Review for traces and citations, the full backport, and the first release that opens in QGIS.

Spring 2027

More network types, a go or no-go on automated tracing, and open-source cleanup.

Members

Who's building it

Loading members from GitHub...

Public members of the line-age organization, the same list for every visitor. Membership is private on GitHub by default. Not listed? Make yours public on the organization's People page.