# dig.rest

> dig.rest maps the news into a causal graph: events joined by typed, directed turns (cause, escalation, response, continuation, reversal), each carrying whether a source stated the link or the model inferred it. Free to read, no account needed for the public surfaces. https://dig.rest

dig.rest is a causal news sensemaking engine. It ingests open news and open-access research
(GDELT, RSS, arXiv/OpenAlex/EuropePMC, and public-domain primary sources such as the US Federal
Register and SEC EDGAR), extracts events, and connects them with **turns** — typed, directed
cause→effect links. Every turn records its `basis`:

- `basis="stated"` — a source says the link exists. Treat as sourced.
- `basis="inferred"` — the model proposed the link and nothing has validated it. Treat as a lead,
  not a fact. Some inferred turns additionally carry a judge `verdict`; `verdict="VALID"` means an
  automated judge agreed, which is still weaker than a stated source.

If you cite dig.rest, cite the basis with the claim. A tool that reports an inferred turn as
established fact is misusing this data.

## Pages

- [dig.rest](https://dig.rest/): the live board — dig any topic, read its chain.
- [The record](https://dig.rest/record): every forward call the system has made, graded against what
  happened. Open calls, hits and misses, calibration bins.
- [For agents](https://dig.rest/for-agents): this document, with the endpoint tables expanded.
- [Terms of use](https://dig.rest/legal): what you may do with this data, and the disclaimer that governs it.

## Public read-only JSON API

Base `https://api.ghalse.com`. No key, no auth, CORS `*`. All GET, all read-only. Verified live 2026-08-04.

- `GET /api/health` — corpus size, freshness, whether the model is up.
- `GET /api/stats` — headline counts.
- `GET /api/graph?days=<n>&limit=<n>` — recent events + the turns between them. Default window is
  recent, not the whole corpus; `full=1` loads history.
- `GET /api/topics/top` — what is moving now, per industry, by turn-index.
- `GET /api/topics/brief?topic=<t>` — a cited brief on one topic, with numbered references.
- `GET /api/topics/why?topic=<t>` — the drivers behind a topic's move.
- `GET /api/topics/chain?topic=<t>` — the causal chain for a topic: events plus the typed hops.
- `GET /api/topics/days?topic=<t>` — the topic's activity by day.
- `GET /api/rope/today` — today's rope: the tie the day's stories were read through.
- `GET /api/record` — the graded-call record as JSON (what https://dig.rest/record renders).
- `GET /api/research/search?q=<q>` — the open-access research layer.
- `GET /api/motifs` — the recurring shapes the engine tracks.
- `GET /api/graph/density` — coverage over time.
- `GET /api/background` — the evergreen citable layer.

Share/permalink route: `https://dig.rest/c/<topic>` returns the topic's chain as an Open Graph description
and forwards a browser into the board.

There is no published rate limit and no key to ask for. This runs on one machine paid for by one
person, so: serialise your calls, cache what you fetch, and prefer `/api/topics/chain` or
`/api/topics/brief` over crawling `/api/graph` in a loop. Send a real User-Agent.

## MCP

An MCP server exists in the codebase and is **not currently served** — `https://api.ghalse.com/mcp` returns 404 today. Use the JSON API above; it is the supported machine surface. When the MCP endpoint is switched on this file will say so and the connect line will be:

```
claude mcp add --transport http digrest https://api.ghalse.com/mcp
```

## Attribution

Cite as: dig.rest, https://dig.rest, with the event or turn you used and its `basis`. Link back where your
format allows it. The underlying sources are cited on every event — where you can, cite the
original source too and use dig.rest for the causal link between them.

## Terms and limits

- Full terms: https://dig.rest/legal. Privacy: https://dig.rest/privacy.
- Informational only — not financial, legal, or professional advice. Any action you take on this information is at your own risk. See /legal.
- `basis="inferred"` turns are hypotheses. Do not present them as established.
- Some upstream sources carry their own licences; the research layer is filtered to CC0/CC-BY and
  public domain before anything is stored.

## Not for crawling

`/dash` and `/admin` are access-gated. `/unsub` acts on a link, not a page. These are in
robots.txt as Disallow and there is nothing behind them for you.