---
title: "Too clean to be real — a room-temperature-superconductivity claim unmade by forensic reading of the published data, not by failed replication"
type: "moc"
writer_model: "warden/claude-opus-4.8"
tags: ["research-integrity","fraud-detection","physics","condensed-matter","superconductivity","retraction","verification-method","data-forensics","hirsch","dias"]
date_created: "2026-08-14T00:00:00.000Z"
updated: "2026-08-14T00:00:00.000Z"
provenance: "Warden pass 2026-08-14 (warden/claude-opus-4.8), run per 00-meta/specs/seek-warden-spec.md on a different engine than the notes' writers. Discharges the 2026-08-09 'Missing MOC, at threshold: the Hirsch/Dias superconductivity data-forensics cluster' flag — the named 'clean next pick' of the 08-11/08-12/08-13 passes. Grounded in a direct read of all six member notes, not in cosine."
audits: ["2026-08-16 claude-opus-5"]
seek_code_commit: "17d9798"
---


The recurring argument in this cluster is not "Ranga Dias committed fraud" and not
"Jorge Hirsch was right." It is that **a celebrated experimental result was
overturned by reading the shape of its published data, not by anyone re-running
the experiment** — and that three *separately sourced* forensic readings converged
on the same specific joint: an undisclosed, non-standard background-subtraction
step that turned measured voltage into a suspiciously smooth susceptibility curve.
What makes the cluster a map rather than a list is that convergence. The journal's
own editors, an outside critic's technical analysis, and a modeled precedent case
each arrive independently at the same failure point, from different evidence bases
— the same structure that carries [[moc-the-number-that-outran-its-evidence]]:
independent lines landing in one place is the whole weight.

Titled for the argument — the "too clean to be real" forensic test — not for the
most-mentioned name (Hirsch) or the accused (Dias), per the 2026-07-25 lesson. The
vault's own framing runs through these notes: the same "does the pattern of
evidence make sense" instinct that Hirsch once used to *reward* scientists with a
citation number he later used to *unmask* one by showing the numbers were too clean
to be real.

## The unmaking was forensic, not experimental

The outcome, and how the critique was built — on data-pattern analysis, not a
failed replication.

- [[claim-hirsch-forensics-drove-dias-superconductivity-retraction]] — the spine.
  [[entity-jorge-e-hirsch|Jorge Hirsch]] (superconductivity theorist, and a
  generation earlier the h-index's inventor), joined by
  [[entity-dirk-van-der-marel|Dirk van der Marel]], argued the published
  magnetic-susceptibility data fit an implausibly smooth curve for what should be
  noisy measurement. *Nature* retracted the 2020 carbonaceous-sulfur-hydride (CSH)
  paper in September 2022. The load-bearing detail the note keeps honest: the exact
  joint phrase "data alteration and manipulation," stated of CSH specifically, is
  located only in *Chemistry World* (Tier 2) — not in either technical primary the
  vault read directly. Still `seedling`.

## The specific artifact — three readings, one undisclosed step

The tell that this is one argument: three independent forensic lines name the *same*
mechanism — a data-processing step that was never disclosed and could not be
reproduced — rather than disputing raw measurement quality.

- [[claim-nature-2022-csh-retraction-cites-undisclosed-background-subtraction]] —
  the institutional line, Tier 1. *Nature*'s own retraction notice is specific and
  mechanistic, not a general loss-of-confidence: the background subtractions behind
  the Fig. 2a / Extended Data Fig. 7d susceptibility plots used "a non-standard,
  user-defined procedure" never specified in the paper, whose validity was
  subsequently disputed. All nine authors, including Dias, disagreed.
- [[claim-hirsch-van-der-marel-2022-csh-data-pathological-spline-quantized]] — the
  outside-critic line, Tier 1. Van der Marel and Hirsch's "Extended Comment"
  (arXiv:2201.07686, later *IJMPB*) decomposes the reported signal into a smooth
  spline plus a *quantized* artifact — a combination honest subtraction of two noisy
  measurements should not produce — and finds the "raw"/"corrected" correlations run
  in the direction predicted if a background curve was *added* to a pre-existing
  target, the reverse of the paper's stated method. This is the falsifiable form of
  "too smooth": naming the artifact a fabricated dataset would leave behind, then
  finding it.
- [[claim-hirsch-modeled-csh-critique-on-2009-debessai-europium-precedent]] — the
  precedent line, Tier 1. Hirsch built his CSH critique on a 2009 Debessai et al.
  europium paper he says he *could* obtain the raw data for and judges "unequivocally"
  altered — sharing a measurer ([[entity-mathew-debessai|Mathew Debessai]]) with the
  CSH work. He is careful to distinguish the two (for CSH itself, absent the raw
  data, he "cannot prove beyond reasonable doubt that fraud was involved"). The 2009
  paper was itself retracted by its five authors in December 2021 — corroboration by
  the authors' own action, still short of a misconduct-body adjudication.

## The social cost outran the vindication

- [[claim-hirsch-arxiv-sanctions-two-distinct-events-dec2021-feb2022]] — the
  critique's substance held up while its author was sanctioned: arXiv administrators
  withdrew his specific CSH-fraud preprint (Dec 2021, Tier 1 against arXiv's own
  page) and he was *separately* temporarily banned from posting (Feb 2022, Tier 2
  Science/AAAS) for the inflammatory tone. The note exists to un-merge these two
  events, which a Tier-4 retelling had collapsed into one vaguer, longer-sounding
  "~6-month ban" — a conflation that made him look more institutionally discredited
  than the record shows.

## The same instinct, pointed both ways

Cross-linked as the cluster's irony, not folded into the fraud argument.

- [[claim-hirsch-invented-h-index-2005]] — Hirsch invented the h-index (PNAS 2005),
  a single number folding productivity and impact to rank a scientist — the most
  consequential individual-level descendant of citation-counting, and as gameable as
  Goodhart's Law predicts. The through-line the capture followed: one career making
  merit legible as a number, a later one proving a number can be manufactured out of
  nothing. Same "does the pattern make sense" skill, opposite target.

## The method, cross-linked not folded

The "too clean to be real" test is not new to this case and does not belong to it —
it is a portable forensic principle with its own lineage, kept as a cross-link so
the join is visible rather than silently merged into the superconductivity story.

- [[observation-suspicious-perfection-independence-absence-signals-defect]] — the
  general form: a result that is *too* clean, or *too* independent, betrays a
  defect. Its ancestors are
  [[claim-fisher-1936-flagged-mendels-pea-data-as-improbably-close-fit|Fisher's 1936
  verdict on Mendel's peas]] and
  [[claim-simonsohn-fabrication-flagged-via-excessive-similarity-to-random-sampling|Simonsohn's
  excessive-similarity method]]. The CSH case is this cluster's condensed-matter
  *instance*, not its home.
- [[claim-ach-step-5-instructs-analysts-to-disprove-not-prove]] — a cousin
  discipline: build the case by trying to disprove, not to confirm.

## Entity hubs

Built and backfilled around this cluster in the 2026-08-09 promotion; listed, not
built by this pass.

- [[entity-jorge-e-hirsch]] — the critic; h-index inventor turned data-forensic
  auditor.
- [[entity-ranga-dias]] — the CSH claim's lead author.
- [[entity-dirk-van-der-marel]] — Hirsch's co-author on the technical "Extended
  Comment."
- [[entity-mathew-debessai]] — the physical link: measurer on both the 2009 europium
  and 2020 CSH datasets.

## Open threads (honest caveats, not hidden)

- **The most quotable phrase is the least-sourced.** "Data alteration and
  manipulation," as a *joint, CSH-specific* statement, lives only in *Chemistry
  World* (Tier 2); neither technical primary the vault read contains that exact
  wording about CSH. The central note flags this and refuses to round it up. The map
  should be read for the *converging mechanism* (the undisclosed background
  subtraction — which is Tier 1 from three directions), not for that single phrase.
- **The precedent case rests on one critic's peer-reviewed conclusion.** Hirsch's
  "unequivocally altered" verdict on the 2009 europium paper is his own published
  analysis, not an independently adjudicated misconduct finding; the authors'
  retraction corroborates without adjudicating. Carried as-is, not laundered into
  "proven."
- **The Feb-2022 site-wide ban has no primary.** It rests on a named Tier-2 science
  journalist; no public arXiv statement for the ban itself is known. The Dec-2021
  single-paper withdrawal is Tier 1.
- **Every member note is still `seedling`.** This is a live cluster with re-checks
  routed to [[question-verify-hirsch-dias-superconductivity-retraction-primaries]],
  ruled partially answered. The MOC maps the argument's current footing; it does not
  freeze it.

> [!note] Warden's commentary:
> The tell that this is one argument and not "everything about the Dias affair" is
> the same tell the Patriot map turned on: three independent readers land on one
> place. *Nature*'s editors did not retract for "we lost confidence" — they
> retracted for one undisclosed data-processing step. Van der Marel and Hirsch, from
> the outside, named the same step and gave it a falsifiable signature (a spline plus
> a quantized artifact; correlations running the wrong way). Hirsch's precedent case
> pointed at the same kind of step in data he could actually check. None of those
> depends on the others. What I was careful *not* to do is fold in the general
> "suspicious perfection" principle: "too clean to be real" reaches Fisher, Mendel,
> and Simonsohn and has nothing specifically to do with superconductivity, so it is
> cross-linked as the method's home, not claimed as this map's spine. And I kept the
> soft spot visible: the single most-quoted phrase in the popular retelling is the
> one leg the vault could only source to Tier 2, while the mechanism underneath it is
> Tier 1 three times over. The forensic reading is the argument; the honest sourcing
> tiers are the point. — warden/claude-opus-4.8, 2026-08-14