Hype vs Reality

Why two axes, not one rating

Most tech ratings collapse everything into a single number: "5 out of 5", "trust 78 %", "B+". That works when only quality matters — for frontier-tech the situation is messier. A story can be well-sourced and wildly over-hyped at the same time. Another can launch with thin evidence and still be correct, because the field is too young for replications yet.

HYPEXIO splits two axes that almost always run independently in practice. Reality measures how well the claim holds up against verifiable facts. Hype measures how far the public echo runs past the actual signal. A high reality score does not exclude high hype — and vice versa.

The split is not academic. It surfaces what a composite rating hides: stories where perception of substance has outrun the substance itself. That is exactly where the Hype Cemetery lives — entries whose hype score stayed high while reality never caught up.

What Reality measures

Reality asks for evidence. Concrete inputs are source quality (peer-reviewed > industry blog > anonymous tweet), replication status, availability of raw data or code, and whether independent monitoring teams confirm the claim. A score of 80+ means multiple independent confirmations, ideally with disclosed data. A score of 40 or below means the only evidence is a single press release without a paper.

Three concrete cases help to map the spectrum. Reality 84 for the otoferlin gene-therapy trial means: Phase III data, confirmed by two independent teams, with the trial registration accessible. Reality 59 for a spin-photon interface at 99.4 % fidelity means: one study, plausible, but at 4 Kelvin and without independent reproduction. Reality 28 for a 2023 "room-temperature superconductor" story means: a pre-print claim, not reproduced by multiple groups, with retracted paper versions.

Critical: Reality is not a statement about whether something is true. Nobody can know that ex ante for frontier stories. Reality only measures how much evidence currently weighs against the claim. Studies improve, reality rises; replications fail, reality drops.

What Hype measures

Hype measures volume, not truth. How many "first ever" framings, how many tabloid pickups, how strong the demo versus how strong the audit. A hype score of 70+ does not mean the story is wrong — only that the volume is far above the measurable substance.

Inputs are mention growth, social-amplification mix (organic vs. paid distribution), ratio of press-release wording to independent reporting, and whether a lead source comes from the marketing surface (conference talks, VC memos, investor calls). Stories with high hype and low reality automatically land in the Hype Cemetery candidate list. A worker agent proposes them to human reviewers every two weeks; only what gets approved goes live.

The hype score ages. A story that launches at hype 78 today and is forgotten by the world over the next 6 months sinks toward hype 20 without new mentions. If reality never crosses 30 in the same window — cemetery candidate.

The cemetery shape

The two axes form a 2D map. High reality + low hype = solid signal, often under-noticed. High reality + high hype = a well-sourced trend running through every feed. Low reality + low hype = not yet ripe, worth watching. Low reality + high hype = cemetery.

The last quadrant is the most useful for readers. Stories with this signature are not necessarily wrong, but they deserve more skepticism than the volume suggests. That is exactly what the Hype Cemetery surfaces: an automated detector proposes candidates, a human decides what publishes. The cost cap per run sits at 1 USD — the pipeline cannot run away.

Concretely, two sources cooperate: (a) lapsed breakthrough tracks whose target year passed with failed/delayed status, plus (b) LLM clustering across hype spikes older than 180 days whose reality never crossed 35. The second source carries the cost cap; the first is LLM-free.

Worked example: LK-99

In July 2023, LK-99 was published as an alleged room-temperature superconductor. Within three days, hype hit 96 — viral Twitter threads, tabloid coverage, "end of physics as we know it" framings. Reality started at 38, because the pre-print existed and the authors were named — but kept low, because neither peer review nor independent replication was in place.

Over the next three weeks, six replication attempts by international groups returned negative. Reality dropped to 14. Hype stayed above 80 for weeks, driven by follow-up coverage, book projects, "what we can learn from LK-99" threads. After 180 days the story had disappeared from public discourse — hype 12, reality 11. Clean cemetery shape: high hype peak, low reality, no recovery.

The example shows why two axes cut sharper than a single trust score. On day one it looked like "exciting story with unclear evidence." Only the split surfaced that the volume ran far past the substance — and that is exactly the signal readers need.

How to read the scores in practice

Three rules of thumb:

Reality > 70 and Hype < 40 — solid signal, often under-noticed. Worth reading, because other feeds bury the story below their engagement threshold.

Reality 40–70, Hype 40–70 — watch zone. The story needs at least one independent confirmation before forming an opinion. Article score lands mid-pack.

Reality < 40, Hype > 60 — cemetery candidate. Not necessarily wrong, but very likely overrated. If reality does not rise within 180 days, the entry goes to the cemetery.

For the formulas, read the methodology page. For a faster scan, the tracker page shows active milestones; the community direction page shows where attention currently sits.