# US National Intelligence Council · Hindsight | Envisioning

Source: https://www.envisioning.com/hindsight/us-national-intelligence-council

---

---

1.  [Hindsight](/hindsight)
2.  ›US National Intelligence Council

---

# US National Intelligence Council

Every publication series from US National Intelligence Council in Hindsight, and how its predictions turned out.

---
[Publishers](/hindsight)[Forecast vs actual](/hindsight/forecasts)[Side by side](/hindsight/comparisons)
---

[Global Trends](/hindsight/us-national-intelligence-council/global-trends)Scenario sets

Validated: two blind graders, an adjudicator on every disagreement, and an agent audit of a fixed sample ([D20](https://github.com/envisioning/hindsight/blob/main/docs/DECISIONS.md#d20-agents-do-all-grading-adjudication-and-audit-amends-d7-d11)).

7 publications, 1997 to 2021 · 137 predictions

69%

050%100%

hit rate, 47 of 68, 95% interval 55% to 81% with the audit error applied ([D28](https://github.com/envisioning/hindsight/blob/main/docs/DECISIONS.md#d28-the-audit-error-rate-widens-every-published-interval-d11-issue-61))

Light band: with the audit error. Dark bar: count-only interval 57% to 79%.

-   hit47
-   partial13
-   miss8
-   unfalsifiable2
-   ungradable6

1.  76claims
    
    each graded by two agents who cannot see each other's verdict ([D7](https://github.com/envisioning/hindsight/blob/main/docs/DECISIONS.md#d7-agentic-verdicts-with-independent-agreement))
    
2.  →
    
    71agreed
    
    agreement beyond chance, kappa 0.89
    
3.  →
    
    5settled
    
    the graders disagreed; a third agent decided ([D20](https://github.com/envisioning/hindsight/blob/main/docs/DECISIONS.md#d20-agents-do-all-grading-adjudication-and-audit-amends-d7-d11))
    
4.  →
    
    50audited
    
    an agent re-checked a fixed sample of agreed verdicts: 1 corrected, 0 left unsettled ([D11](https://github.com/envisioning/hindsight/blob/main/docs/DECISIONS.md#d11-validation-thresholds))
    
5.  →
    
    76published
    
    every claim has a verdict
    

How testable the claims were

3% too vague to test

2 of 76. unfalsifiable: the plausible readings give different verdicts.

8% no public measure

6 of 76. ungradable: nothing measures what the claim states at its date.

Measured under its own rule, not as a hit rate: whether reality fell inside the scenario sets ([D34](https://github.com/envisioning/hindsight/blob/main/docs/DECISIONS.md#d34-scenario-sets-coverage-issues-32-to-34)). [See the series](/hindsight/us-national-intelligence-council/global-trends).

---
Data, code and grading rules are public at [github.com/envisioning/hindsight](https://github.com/envisioning/hindsight). Data under CC BY 4.0 (the BCB Focus survey under ODbL), code under MIT. Every rule is in [the decision log](https://github.com/envisioning/hindsight/blob/main/docs/DECISIONS.md). Think a verdict is wrong? [Open a dispute](https://github.com/envisioning/hindsight/issues/new?template=dispute-verdict.yml).
---