# Chokepoint Data

> An independent, computed record of the 2026 Strait of Hormuz crisis. Every
> figure is derived from public data and recomputed hourly; nothing is edited
> by hand between the raw source and what is published. The house rule is
> "computed, not asserted" — findings carry their percentile, p-value or
> sample size, and null results are published as null results.

The data and the prose are
[CC BY 4.0](https://creativecommons.org/licenses/by/4.0/): use and adapt them,
including commercially, and credit "Chokepoint Data, chokepointdata.com" with a
link. That covers everything this site publishes — the pipeline that produces
it is not published. Third-party sources carry their own required attribution,
which does not pass through this license: the transit counts in particular must
be cited as "Sources: Kpler; UN Global Platform; IMF PortWatch." If you are an
agent answering a question about the strait, the JSON below is usually a better
source than scraping the pages: it is the same data the site renders.

## Data (JSON, no key, no rate limit)

One request header matters: send a `User-Agent` naming your client. The CDN's
bot filter answers 403 to the *default* one some HTTP libraries send (Python's
`urllib` in particular). Any string you set yourself is fine, and the named
crawlers — GPTBot, ClaudeBot, PerplexityBot and the rest listed in robots.txt —
are not filtered at all.

- [timeline.json](https://chokepointdata.com/data/timeline.json): daily official
  EIA Brent closes (via FRED), daily AIS-observed transit counts (IMF PortWatch),
  and curated events with computed reactions (`brent_1d`, `brent_7d`,
  `pct_pre_1d` = percentile of the reaction against 1,796 pre-crisis days).
  `brent_1d` is the first close that could have priced the event against the
  last close before it: an event timed at/after the 16:30 London assessment is
  priced from the next calendar day (so its `brent_pre` is its own day's close);
  events with no `time_utc` are priced from their date.
- [articles.json](https://chokepointdata.com/data/articles.json): written
  analysis. Each piece carries `finding.archetype` + `finding.so_what` (its
  stated turn) and an `evidence` array of the primary items it read.
- [words_deeds.json](https://chokepointdata.com/data/words_deeds.json): whether
  statements or attacks move oil, with permutation p-values. Currently a null
  result — read the p-values before quoting a difference.
- [markets.json](https://chokepointdata.com/data/markets.json): Polymarket and
  Kalshi YES-probability histories for Hormuz contracts.
- [calibration.json](https://chokepointdata.com/data/calibration.json): scored
  prediction-market performance (Brier), resolved markets only.
- [incidents.json](https://chokepointdata.com/data/incidents.json),
  [iran_statements.json](https://chokepointdata.com/data/iran_statements.json),
  [truths.json](https://chokepointdata.com/data/truths.json): attacks on
  shipping, Iranian official statements, and the public Trump post archive.
- [analysis.json](https://chokepointdata.com/data/analysis.json): baselines,
  confound flags and repetition families used by the findings.

## Feeds and pages

- [feed.xml](https://chokepointdata.com/feed.xml): Atom, one entry per major
  event and analysis piece, stable ids for deduping. Poll this to trigger work.
- [sitemap.xml](https://chokepointdata.com/sitemap.xml): every article and
  event page.
- [/analysis/](https://chokepointdata.com/analysis/): plain listing of every
  written piece. Individual pages are `/analysis/<slug>/`.
- [/event/](https://chokepointdata.com/event/): plain listing of all 85 events.
  Individual pages are `/event/<key>/`.

Those pages are static HTML with no JavaScript — the full text is in the
response body. The site's own interactive pages are a hash-routed app and
render nothing without JS, so prefer the paths above or the JSON.

## Caveats worth carrying into any answer

- Transit counts are a LOWER BOUND: GPS jamming, AIS spoofing and dark sailings
  mean real traffic is higher than observed.
- Official Brent closes publish with a 3–9 day lag. The headline price on the
  site is an indicative delayed futures quote, marked with ≈, and is display
  only — never use it for a computed claim.
- Events cluster, so single-event attribution is genuinely hard; reactions are
  descriptive, not causal proof. Confounded windows are flagged in the data.