Skip to main content
Research Insights serves what outside analysts, specialist trade press and industry podcasts said, restated as insight records. Each observation is one source document — an article or one podcast episode — with three to five insights inside it, and every insight carries a verbatim quote from the source that was verified character by character before the record was written. Rows are returned newest first.
Plan: Free and above · Credits: 1 per call; the sources list is free

​
What You Can Use Research Insights For

See what specialist publications are saying about a company, a supply chain or a technology. Follow one source over time. Filter to the pieces that took a negative stance on a name this week. Quote the exact sentence behind a claim, with a link to the piece.

​
This Is Analyst Speech, Not Company Speech

“TSMC says CoWoS capacity doubles in 2027” is Management Guidance — the company speaking under its own disclosure obligations. “A trade publication says TSMC’s CoWoS capacity doubles in 2027” is this dataset — an outside source’s claim about the company.The two are kept apart and never merged. Attribute every insight here to its publisher, never to the company it is about.
It is also not the news feed: Company News scores a wire article for direction and impact, while this dataset restates an analysis piece without scoring it.

​
Research Is Not Keyed by Company

A research piece is often about several companies at once, and sometimes about none. company, source, topic and stance are filters that combine, not addresses. With no filter, the endpoint returns the whole stream newest first. company accepts a ticker, company_id, CIK or ISIN and is expanded to every symbol the company is known by; the response lists them in symbols_searched.

​
Every Insight Is Quote-Anchored

Each insight carries a claim, a verbatim quote and, where the record has them, bullets or a summary, plus a type (estimate, forecast, mechanism, event or opinion) and the source’s own confidence in its claim (asserted, estimated, hypothesis, unconfirmed or declined). An insight whose quote did not match the source is not served. dropped lists its id and claim, and insights_dropped counts them, so you know the source said more than the record serves.

​
Tickers in companies Are Our Resolution

companies lists the listed companies the source named, each with ticker, mentions, impact — one sentence on what the source establishes about that name — and basis: named, product_mention:<product> when the source named only a product we mapped to the company, or subsidiary_of:<parent>. The tickers are resolved from the names the source printed and are not yet confirmed against the company registry. Unlisted names are in private_entities.

​
Podcast Rows Carry Two Timestamps

source_type is analyst_newsletter, trade_press, industry_forum or podcast. A podcast row also carries video_id, duration_sec, main_guest and guests.
published_at_utc is when a podcast episode went live and captured_at is when we obtained its transcript. For a back-catalogue episode they can be years apart, so an as-of question needs captured_at, not the publication date.
Podcast text is an automatic speech transcript: no editorial punctuation, with filler words and repeats left in. Quotes are verified against it character by character, so a quote that reads awkwardly is correct, not garbled.

​
What Is Served and What Is Not

Everything we produced is served in full: each insight’s claim, bullets and verbatim quote, and record_md, the whole record. The article body is never served — follow url to read the piece.

​
Research Insights Language

source_title and every quote keep the source language — Korean, Japanese and Chinese included — and are never translated. lang names the source language.

​
Research Insights Coverage

A few dozen research sources — call /v1/research/sources for the current list: analyst letters, semiconductor and data-center trade press (including Korean and Japanese outlets), and industry podcasts. The pipeline runs hourly.
The pipeline started on 2026-09-23 and is being backfilled. Read coverage.available_from on the response for the current floor. An empty result for an earlier date means the pipeline had not reached it, not that nobody wrote about the company.

​
Research Insights Date Convention

since and until filter on published_on: the source’s printed publication date or, where the source prints none, the day we extracted it. published_on is also the sort key.

​
GET /v1/research

Insight records, newest first.
​
string
A ticker (NVDA, 2330.TW, 6758.T), company_id, CIK or ISIN.
​
string
A feed or publisher name, exactly as /v1/research/sources lists it.
​
string
A topic tag, case-insensitive, for example cowos, hbm or agents.
​
string
The source’s own stance toward its subject: positive, negative, mixed or neutral.
​
string
Earliest publication day, YYYY-MM-DD. until sets the ceiling.
​
integer
default:"20"
Rows per page, between 1 and 100. Page through with cursor.
Each row carries doc_id, feed and publisher, url, source_title (the piece’s own title, verbatim), claim_title (the source’s actual claim in one sentence), published_on and published_at, lang, topics, source_stance, stance_downgraded (true when the stance’s evidence quote did not verify and the stance was reset to neutral), source_type, insights, companies, private_entities, dropped, insights_dropped, quotes_total, record_md and extracted_at.

​
GET /v1/research/sources

The source vocabulary — every feed and publisher source accepts, with its record count and newest publication date. Free. This endpoint takes no parameters. See also Management Guidance for what companies say about themselves, Company News for scored wire articles, and Company News Events for what happened to a company this week.