Hype Index Methodology v1.0

Why publish a number at all?

Public discussion about AI mixes three different things: measured capability, capital markets, and cultural attention. Those can diverge for years. A weekly index is not a claim about welfare, morality, or inevitability; it is a disciplined attempt to summarize how hot the discourse–finance loop ran over a short window, in a way readers can audit and disagree with.

We publish scores because “vibes” scale poorly. A transparent composite forces editorial hygiene: name inputs, bound ranges, and version the recipe when definitions shift.

The four components (each 0–25)

Each week’s snapshot includes four integers, each between 0 and 25 inclusive. They are not model endpoints; they are weekly editorial judgments with documented notes.

1) Capability claims (0–25)

Measures how assertive the public conversation was about new abilities—agents, autonomy, reasoning, superhuman-this, scientist-that—relative to what third-party evaluations support. Product launches, benchmark battles, and credulous headlines raise this component; careful replication and narrow claims lower it.

2) Funding / valuation pressure (0–25)

Tracks the temperature of money: capex guides, mega-rounds, strategic investments, GPU financing stories, and expectations of future valuations. This component rises when markets treat AI spend as non-discretionary and compress discount rates on distant payoffs.

3) Media saturation (0–25)

Captures mainstream attention: front pages, broadcast segments, politics-of-AI chatter, and non-specialist outlets leaping from anecdote to ontology. High saturation is not “bad,” but it raises the risk that base rates and uncertainties get flattened.

4) Skeptic friction (0–25)

Captures counter-pressure: rigorous ROI skepticism, methodology critiques, legal jeopardy that changes planning, employee pushback, or macro narratives arguing for slower diffusion.

The inversion rule (why skepticism lowers the headline)

“Skeptic friction” is stored as how much counter-pressure showed up this week (higher means more pushback). It does not add to hype. In v1.0, the weekly headline score is:

score = capabilityClaims + fundingPressure + mediaSaturation + (25 − skepticFriction)

So if friction is low (little organized skepticism), the last term is near 25 and the headline runs hot. If friction is high, the last term falls—sometimes sharply—reflecting that attention was contested rather than freely converted into exuberance.

Each component remains bounded 0–25; the headline score is bounded 0–100.

Driving claims

Each week we list 2–4 driving claims from the claims corpus. A claim is “driving” if, in editors’ judgment, it absorbed meaningful incremental evidence, legal action, measurable outcomes, or disproportionate attention during the window. Inclusion is not endorsement; a refuted claim can drive hype if people argue about it loudly.

Weekly aggregation

Editors collect source notes throughout the ISO week, then reconcile on publication day. The component scores are set jointly to satisfy two constraints:

1. Internal consistency with the week’s notes (each component should be defensible in a paragraph).
2. Range discipline (0–25 per component; total headline score on 0–100).

Headline score is computed from the four posted components using the v1.0 rule capabilityClaims + fundingPressure + mediaSaturation + (25 − skepticFriction). If inputs disagree—say, strong capability news but equally strong replication critiques—scores should reflect that tension rather than smoothing it away.

Why we version

Capability definitions drift, markets change their vocabulary, and what counted as “saturation” in 2024 may not serve readers in 2027. Versioning lets us preserve historical series honesty: when the recipe changes, the number’s meaning changes; readers should not be left to guess.

An invitation to interrogate

The index is meant to be * criticized. If you believe skeptic friction was underweighted, publish your counter-ledger. If you think funding pressure ignored rising real rates, show the mechanism. We would rather be corrected than mythologized.

The deeper editorial standard of AI Hype Tracker remains constant across versions: date claims, quote them faithfully, separate evidence from inference, and keep score without cosplaying either doom or booster tribalism.