A map of what technology is thinking about.
What it is
SYMTRI is an open source map of research, software, and ideas. Explore regions, follow a thread, and read the original sources behind connections.
How collection and classification work
Hourly ingestion collects public observations. Classification places them into regions and threads using the source title, available summary, and categories. The classifier is versioned; saved inputs allow older records to be reclassified as the rules improve. A signal can belong to more than one subject.
Links to the same canonical source are deduplicated into one signal while separate source observations retain their attribution. Content-identifying URL parameters are preserved. Readable evidence is published only after the source passes the reading pipeline; the map can count classified signals that do not yet have readable evidence.
The live map follows a rolling fourteen-day window based on each feed’s event date: paper publication, repository creation, model/dataset activity, discussion posts, or catalog additions. It is not a count of every recent project update or every newly discovered document. Region activity and shared-signal connection counts use the full classified window, independently of the overview’s 300-detail display cap. Topic pages list readable sources from the full archive in publication-date order. Source publication time, first archive discovery, ingestion time, and report publication time describe different events. “First archived” is the signal’s recorded first-seen time in SYMTRI, rather than the date its publisher created or updated the work.
Saved daily snapshots preserve the evidence and coverage recorded at that time. Weekly reports identify their reporting period and available snapshot coverage. A later-discovered source cannot be assumed to have been known in an earlier snapshot.
Counts use unique canonical signals, rather than the number of feed observation rows. Momentum compares weighted activity in the latest twenty-four hours with the preceding twenty-four hours. Weights combine classification relevance and the source’s importance proxy. “Rising” requires a recent weight of at least 1.5 and more than 1.5 times the previous weight plus 0.5; “quiet” means neither twenty-four-hour window has activity. The visual activity score uses a twenty-four-hour half-life and is scaled relative to the strongest region.
Activity measures attention in the sources we cover. It is not a complete picture of the internet, a quality ranking, or a prediction. Quiet threads may reflect limited coverage. Sources can be healthy, partially available, or unavailable in a run; missing coverage limits comparisons.
What do connections mean?
A measured connection counts signals classified into both regions. Shared subject matter and semantic similarity can also guide exploration. These links do not show that one development caused another, that researchers agree, or that a project works as claimed. Editorial suggestions for related regions are labeled separately from measured connections.
Sources and licenses
Sources include Hacker News, GitHub, arXiv, OpenAlex, NASA, NASA JPL, CISA, Europe PMC, Hugging Face. Availability varies by run. Each story links to its original publisher.
The SYMTRI software is MIT licensed. Articles, abstracts, repositories, and other source materials retain their own licenses and attribution. SYMTRI’s classification metadata and annotations are also MIT licensed. This license does not grant rights to republish third-party content.
Ask SYMTRI
Ask explores the archive and uses AI to synthesize available evidence. Answers include source links so you can check the claims. AI can misread evidence, and an empty result means we lack matching sources. It does not prove that a subject has no activity.
Limitations and corrections
Classification can miss relevant material or place a source in the wrong thread. Preprints may be revised, public repositories may be unmaintained, and publisher claims may be disputed. AI-assisted reading and synthesis can introduce mistakes; original sources remain the authority for what they actually report.
To request a correction, open a GitHub issue with the SYMTRI page or source URL, the statement or classification in question, and supporting evidence. Avoid including private information. Corrections to classification and reading evidence can change current results; saved reports retain their own reporting context.
Open source
Explore the code, report an issue, or contribute on GitHub ↗
About the creator
Built by Stanley Yang.