What instruments sit behind the SIREN reading? Score history, source health, pillars, moves, embed and API.
The instrument room · everything behind the reading
What instruments sit behind the SIREN reading?
The home page gives the reading. This page shows the working: where the score sits, how it has moved,
which sources reported, what each pillar is doing, and how to take the number with you.
How does SIREN count AI harm and benefit side by side?Both at once
Tilts only on stories in one list and not the other: 8° × |H − B| ÷ (H + B), heavier pan down, at most 8°. Level within 2, or under 5 matched.
24 of 500 stories in the newsroom's current window matched the harm list and 5 matched the benefit list, as of 05:24 UTC on Thu 15 Oct 2026. A count of words, not of outcomes.
Observed position · the mark moves both ways · not a prediction
Five stages, coolest on the left, lit at the one the index is in as of
.
Hotter means more is happening — not that anything is worse. The stages are the
SIREN levels themselves, not a second scale that could disagree with the first.
5DORMANT0–34SIREN 5, DORMANT, band 0 to 34. cooler than the current stage
56.6NOW ▼4ROUTINE35–54SIREN 4, ROUTINE, band 35 to 54. current stage. Composite 56.6 of 100.
3ELEVATED55–69SIREN 3, ELEVATED, band 55 to 69. hotter than the current stage
2ACCELERATED70–84SIREN 2, ACCELERATED, band 70 to 84. hotter than the current stage
1UNPRECEDENTED85–100SIREN 1, UNPRECEDENTED, band 85 to 100. hotter than the current stage
035557085100
Pin at 56.6 · dashed mark is the previous observation, 56.1 at 2026-10-15 03:34:40 UTC +0.5
This stage holds while the composite sits between 32.0 and 58.0 — the 35–54 band widened by the ±3-point deadband at each edge.
No stage change in the record. SIREN 4 has been the reading for 21d, since 2026-09-23 20:04:22 UTC, across 610 scored observations.
What would move it?
The published gates, as numbers. Each row is a requirement beside the
current reading; nothing here says any of them is met. Heating is measured over 3h and 30m;
standing down over 12h and 3h — cooling is deliberately the slower move, so a stage this index
raises is one it keeps for a while.
▲ What would move one stage hotter?SIREN 3 · ELEVATED
○Composite
above 58.0now 56.6 · 1.4 below that mark — not met
○3h mean
above 58.056.4 over 2 observations — not met
○30m mean
above 58.056.6 over 1 observation — not met
○Pillars agreeing
2 of 4 scored1 above 58.0 · 1 of 5 not scored — not met
●Since last change
6h minimumheld 21d — met
●Dark pillars
none0 of 5 dark — met
▼ What would move one stage cooler?SIREN 5 · DORMANT
○Composite
below 32.0now 56.6 · 24.6 above that mark — not met
○12h mean
below 32.054.7 over 7 observations — not met
○3h mean
below 32.056.4 over 2 observations — not met
○Pillars agreeing
2 of 4 scored0 below 32.0 · 1 of 5 not scored — not met
●Since last change
6h minimumheld 21d — met
●Dark pillars
none0 of 5 dark — met
Last run the score crossed a band edge but not the deadband around it.
rule_fired: deadband
610 scored observations. The y axis is fixed at 0–100 and never auto-scaled to the data, so the question this chart answers stays "where in the range", not "what shape is the noise".
610 observations · 2026-09-23 to 2026-10-15 · range 36.8–60.7 · now 56.6 in ELEVATED 55–69
4 live, 1 stale, 10 awaiting a baseline, of 15 sources. A source awaiting a baseline answered fine; there is simply no frozen reference to score it against yet, so it is published and not counted.
arxiv: live
federal-register: live
hn: live
sec-fts: live
wikipedia: stale
github-releases: awaiting a baseline
govuk: awaiting a baseline
huggingface: awaiting a baseline
intelligence-arena: awaiting a baseline
kalshi: awaiting a baseline
manifold: awaiting a baseline
openrouter: awaiting a baseline
polymarket: awaiting a baseline
stockanalysis: awaiting a baseline
vastai: awaiting a baseline
4 live
1 stale
10 awaiting a baseline
All 15 sources, one row each
●arxiv0slive
◇github-releasesno baseline · no readno baseline
◇huggingfaceno baseline · no readno baseline
◇openrouterno baseline · no readno baseline
●sec-fts0slive
◇stockanalysisno baseline · no readno baseline
◇vastaino baseline · no readno baseline
●hn0slive
◇intelligence-arenano baseline · no readno baseline
◐wikipediastale · 5hstale
●federal-register0slive
◇govukno baseline · no readno baseline
◇kalshino baseline · no readno baseline
◇manifoldno baseline · no readno baseline
◇polymarketno baseline · no readno baseline
●5 live◐1 stale◇10 awaiting baselineSources awaiting a baseline answered fine and are published, but there is no frozen reference to score them against yet.
What are the five pillars on this page?
Which part of the field is loudest today. Each bar is linear in score from 0 to 100,
so the ordering is the comparison; the cards underneath carry each pillar's own history.
Capability leads at 66.4. Markets is awaiting a baseline.
https://siren.watch/api/receipts/<id>.jsonthe hash-chained receipt behind one observation
Static files. No key, no rate limit, CC BY 4.0. Attribution: siren.watch.
What questions do people actually ask?
What instruments sit behind the SIREN reading?
The home page gives the reading. This page shows the working: where the score sits, how it has moved, which sources reported, what each pillar is doing, and how to take the number with you.
How does SIREN count AI harm and benefit side by side?
Harm and benefit are counted side by side and never summed. Strongest evidence first on each side; a status word describes the evidence, not the importance.
Where does the index sit?
Five stages, coolest on the left, lit at the one the index is in. Hotter means more is happening — not that anything is worse. The stages are the SIREN levels themselves, not a second scale that could disagree with the first.
Where does the score sit over time?
This page charts every scored observation over time. The history of AI risk page and the methodology page explain the frozen reference curve that never updates live.
What does source health on this page report?
4 live, 1 stale, 10 awaiting a baseline, of 15 sources. A source awaiting a baseline answered fine; there is simply no frozen reference to score it against yet, so it is published and not counted.
What are the five pillars on this page?
Which part of the field is loudest today. Each bar is linear in score from 0 to 100, so the ordering is the comparison; the cards underneath carry each pillar's own history.
Where are the recent SIREN index moves?
Recent scored moves are listed on this page, with the full archive at Where are the SIREN index moves?
Where is the frozen reference distribution explained?
Where today sits inside the frozen reference distribution, and why that curve is never updated live, is drawn and explained on the methodology page.
How do I put the index on my site?
One iframe. No script, no tracking, no key. It renders light or dark to match the page it sits in, and the number inside it is server-rendered too.
What public JSON files does SIREN expose?
Static JSON under /api/: state, history, health and per-reading receipts. No key, no rate limit, CC BY 4.0. Attribution: siren.watch.
What does the Capability pillar count?
What shipped, and what got written down.
What does the Compute & Capital pillar count?
Silicon, spend, and the market’s opinion of both.
What does the Attention pillar count?
How loudly the world is talking.
What does the Governance pillar count?
Paperwork with the force of law.
What does the Markets pillar count?
Where money is posted against an opinion.
What would move one stage hotter?
The published heating gates, as numbers beside the current reading. Nothing here says any of them is met.
What would move one stage cooler?
The published cooling gates, as numbers beside the current reading. Cooling is deliberately the slower move.