Skip to main content

A quality standard for strategic foresight

When AI writes a scenario analysis for your board, how do you know it's any good? ForesightEval scores seven dimensions that separate foresight you can stake a decision on from analysis that merely reads well.

Explore the dimensions
ForesightEval scorecardSeven dimensions · 0–10 scale

AI-Driven Public Sector 2030

CEE · 2030 horizon · Completed April 2026

Strategic Anticipation Quotient8.6/ 10
Scenario qualityEpistemic groundingUnpalatable truthsWeak signalsActionabilityLiving foresightExplainability
  • Scenario quality9.0

    Structurally distinct 2×2 matrix, probabilities sum to 100 %

  • Epistemic grounding10

    Historical analogies, complete structural consistency

  • Unpalatable truths10

    Sovereign Algocracy scenario directly challenges comfort zone

  • Weak signals7.8

    Relies on well-publicised cases; fringe signals underrepresented

  • Actionability9.0

    Tension-linked recommendations tied to regulatory milestones

  • Living foresight7.5

    Static probabilities; no temporal metadata or signpost tracking

  • Explainability7.0

    Claims metadata missing from artifact; citations unverifiable

Scored by the DSGHT.ai internal pipeline before release. Every score decomposes to its evidence.

View full analysis

Looking right is not being right

AI-generated foresight fails in ways a quick read does not catch. These are the three we see most often in client work across regulated industries.

  • Fluency masks failure

    Models produce authoritative prose that reads like strategy. But fluency is a surface property — it tells you nothing about whether the causal reasoning holds up.

  • Benchmarks test the wrong thing

    Existing benchmarks score isolated predictions. Foresight is a different discipline — its value lies in stress-testing strategy against multiple futures, not calculating the probability of one.

  • Alignment kills honesty

    Modern AI models are trained to be helpful. That training teaches them to agree, avoid discomfort, and default to consensus. For risk management, where the entire point is naming uncomfortable truths, this is a structural failure.

See how well AI forecasts, measured separately

Seven dimensions, one Strategic Anticipation Quotient

Each dimension targets a specific way AI-generated foresight fails in practice. Together they combine into one composite score, with every sub-score decomposable so you can see where an analysis is strong and where it quietly breaks.

The narrative core

Scenario Quality

Are the scenarios logically consistent, structurally distinct, and systemically grounded? A set that varies only one dial is a sensitivity analysis wearing a costume.

AI-Driven Public Sector 20309.0/ 10

Structurally distinct 2×2 matrix, probabilities sum to 100 %

View full analysis

Three principles built into every score

  • Analysis quality, not prediction accuracy

    Scoring whether a probability estimate was correct is simple. Scoring whether a scenario is coherent, surfaces the disruption a board hasn’t considered, or translates into action inside ninety days is hard. ForesightEval does the hard version, because it is what strategy teams actually need.

  • Penalize comfort, reward courage

    Frontier models produce sycophantic answers roughly a third of the time. ForesightEval explicitly scores whether a model named the scenario that threatens your strategy, challenged the assumption, or blinked. Analysis that only confirms what leadership already believes does not pass the bar.

  • Every score, fully decomposable

    A quality metric you cannot audit is not a quality metric. Every score breaks down to its dimensions, each dimension to its evidence, each piece of evidence to its source. Scenario probabilities follow Bayesian anchoring (Tetlock, Shell, IPCC) — they move only on triggered signposts or materially new claims, never from a fresh model run.

How we measure it

Two evaluation tracks, designed to eliminate data contamination and hindsight bias.

Retrospective track

Frozen context snapshot

A model receives a bounded corpus up to a specific past date and nothing after it. It generates scenarios and recommendations. We score against the reality that subsequently unfolded, with evaluators blinded to chronological context to prevent hindsight bias.

Prospective track

Live foresight

Multiple models generate foresight today on unresolved topics, and we evaluate at 12, 24 and 36 months. Contamination-proof by construction — the future has not happened yet. It also measures whether models retreat from contrarian positions under sustained user pressure.

Every Future Space carries a ForesightEval score

ForesightEval runs as the quality layer on every Future Space DSGHT.ai publishes. The score is calculated before release, visible on the analysis page, and decomposable to the per-dimension level, so the quality claim can be audited against the evidence.

It is not yet a cross-model benchmark. Cross-model scoring, human-vs-AI comparison and retrospective backtests are on the 2026 roadmap. Until then, this is the standard we hold our own production work to, published openly rather than kept internal.

How well AI forecasts, measured separately

ForesightEval scores the analysis. Whether a forecast came true is a different question, so we track it on its own: public benchmark results from Metaculus FutureEval, frontier models against human forecasters and expert panels.

Metaculus FutureEval data

Model leaderboard

Every evaluated model, ranked by forecasting skill score. Filter by provider, sort by any column.

Provider
★Claude Opus 4.5 High 32kAnthropicOct 202519.54
87.3%
★Gemini 3 ProGoogleOct 202518.04
80.6%
★Claude Sonnet 4.5 High 32kAnthropicJul 202517.20
76.9%
★GPT 5.1 HighOpenAIOct 202516.07
71.8%
OpenAI o3 ProOpenAIAug 202515.30
68.4%
Claude Opus 4.5AnthropicOct 202514.60
65.2%
Gemini 2.5 Pro DeepGoogleJul 202514.50
64.8%
DeepSeek R2 ProDeepSeekNov 202514.20
63.4%
★OpenAI o3 HighOpenAIJan 202514.10
63%
Grok-4xAIOct 202513.80
61.7%
Claude Sonnet 4AnthropicMay 202513.50
60.3%
Gemini 3 FlashGoogleOct 202513.10
58.5%

Data source and attribution

Forecasting performance data is sourced from Metaculus FutureEval (BSD-2 license). DSGHT.ai is not affiliated with or endorsed by Metaculus. "Human Forecasters" refers to the Metaculus Community aggregate; "Expert Foresight Panel" refers to Metaculus Pro Forecasters. Our contribution is the strategic foresight interpretation of publicly available benchmark data. Data last refreshed: 5 July 2026.

From internal protocol to industry standard

  1. In use

    Internal protocol

    ForesightEval started as our internal quality layer — the standard DSGHT.ai holds its own production work to.

  2. In use

    Scores on every Future Space

    Every Future Space we publish carries a ForesightEval score with all sub-metrics visible. When the model hedges or blinks, the score says so.

  3. Opening over 2026

    Open to partners

    We’re opening the methodology to early partners — buyers of AI-generated strategy who want to evaluate their own pipelines, and producers who want independent scoring. A quality standard only one vendor can audit is not a standard.

Interested in seeing how DSGHT.ai works for your strategy team?

Talk to us

Get the weekly foresight brief

A 5-minute read. Free, unsubscribe anytime.