---
title: "A provenance metaphor with a provenance problem — how 'low-background steel,' a figure for trusting only uncontaminated sources, rests at every factual layer on the tidier-than-sourced version of its own story"
type: "moc"
writer_model: "warden/claude-opus-4.8"
tags: ["low-background-steel","provenance","ai-training-data","model-collapse","nuclear","contamination","metaphor","attribution","priority-claim","source-discipline","myth-correction"]
date_created: "2026-08-17T00:00:00.000Z"
updated: "2026-08-18T00:00:00.000Z"
audits: ["2026-08-18 claude-opus-5"]
audit_status: "APPENDED 2026-08-18 (scheduled cross-model audit, auditor claude-opus-5, writer warden/claude-opus-4.8 — different build, same Opus lineage, so a weaker independence than a cross-family check): all 26 wikilinks resolve; every bullet's tier, date, figure and status label was checked against the member note it summarises and none misstates it, including the three notes carrying their own Correction history (the fallout-decay, metaphor-coinage and UNSCEAR-revision notes), whose current status the MOC reports correctly. Two structural fixes applied: the conventional `audits` field was missing and is now present, and [[moc-attribution-and-origin-myths]] — named here as this map's general home — carried no link back, so the reciprocal link was added there. No claim in this map moved."
provenance: "Warden pass 2026-08-17 (warden/claude-opus-4.8), run per 00-meta/specs/seek-warden-spec.md on a different engine than the notes' writers. Discharges the 2026-08-16 [entity] 'Missing MOC candidate: the low-background-steel cluster crossed the ~5-note threshold' flag and its 2026-08-17 [entity] growth-update restatement ('build on the next session that works this thread rather than deferring a fourth time'). Grounded in a direct read of every member note, not in cosine."
seek_code_commit: "17d9798"
---


The recurring argument in this cluster is not "low-background steel is real" (it is)
and not "the AI-training-data metaphor is wrong" (it isn't). It is that **a metaphor
whose whole point is provenance — trust only material made before the contamination
event — rests, at every factual layer, on the tidier and more repeatable version of
its own story rather than the version the primary record supports.** Three separate
times, the clean account is the one that spread and the sourced account is messier,
narrower, and quieter:

- **The contamination mechanism.** The popular story (Tier-4 Wikipedia) is cinematic:
  post-war, air-blown steelmaking bakes *atmospheric fallout* into the metal, so only
  pre-1945 steel is clean. The two Tier-1 primaries the vault has actually read say
  something narrower and more mundane — cobalt-60 enters steel from *furnace-lining
  wear gauges* and from *recycled scrap feedstock*: a supply-chain contamination, not
  the atmosphere.
- **The "it is healing itself" claim.** The tidy story is that fallout decayed after
  the 1963 test ban, so low-background steel is "no longer necessary." The dose decline
  is Tier-1 real. But a 2016 primary on people *actually sourcing* clean steel still
  could not buy it off the shelf — they got it by avoiding scrap and controlling
  provenance, not by waiting for the air to clear.
- **The metaphor's own coinage.** The tidy story attributes the AI-data reading to one
  named person and a dated domain registration (Tier-1 confirmed). But the shape was
  already loose in public three months earlier, and the one *named* prior claimant
  rests entirely on a single unlinked, unquoted secondhand paraphrase with no archival
  trace found in two web archives that otherwise cover its subject densely.

What makes this a **map and not a list** is the bridge between the metallurgy and the
metaphor: they are the *same provenance question asked twice*. The steel leg asks where
the cobalt came from and finds the tidy answer unconfirmed; the coinage leg asks where
the metaphor came from and finds the tidy answer unprovable. A figure for contaminated
provenance turns out to have contaminated provenance of its own — and the dose leg is
the hinge that joins them, because the metaphor's own stated limit (fallout stopped
being injected in 1963; synthetic text has not) is exactly the finding that breaks the
"the clean reserve will replenish" reading.

Titled for the argument — a provenance metaphor with a provenance problem — not for
"low-background steel," the entity every note names (the 2026-07-25 lesson: the
recurring entity is not necessarily the recurring argument). This is a deep
single-instance map that rhymes with, and is cross-linked to but **not** folded into,
the vault's other "the neat version outran the record" maps: [[moc-too-clean-to-be-real]]
(fabrication caught by its own smoothness), [[moc-the-number-that-outran-its-evidence]]
(a wartime statistic that hardened before anyone counted honestly), and
[[moc-the-watermark-that-couldnt-prove-its-own-origin]] (a physical mark that cannot
authenticate itself). Its general home is [[moc-attribution-and-origin-myths]] — credit
or cause assigned to one tidy actor when the record is duller and more crowded.

## What the material actually is — and the contested mechanism

The literal provenance question: what makes pre-atomic steel clean, and does the
primary literature actually say what the popular telling says?

- [[claim-low-background-steel-is-pre-1945-steel-salvaged-from-fallout-contamination]] —
  the anchor. Pre-1945 steel (famously the Scapa Flow scuttled-fleet salvage) is prized
  for sensitive detectors because its feedstock — pre-test-era air — no longer exists.
  But the *mechanism* it carries ("present-day air … deposited into the steel") is a
  technical claim resting only on Tier-4 Wikipedia and is flagged `[unverified-mechanism]`.
- [[claim-heusser-1995-traces-steel-cobalt-60-contamination-to-furnace-gauge-sources]] —
  Tier 1, the standard low-background-physics review: iron "is usually contaminated
  through the monitoring of furnace linings with ⁶⁰Co," and old ship iron is clean
  because it predates *that gauging procedure* — a narrower and different claim than
  "predates the contaminated atmosphere."
- [[claim-zhang-2016-names-scrap-metal-feedstock-as-commercial-steels-cobalt-60-source]] —
  Tier 1, an independent second primary: the PandaX-II collaboration says plainly that
  "it is common to use scrap materials in the smelting process," so radiopurity "cannot
  be well controlled." The variable is scrap in the charge, not open air in the melt.
- [[claim-no-primary-source-names-atmospheric-entrainment-as-steels-cobalt-60-pathway]] —
  the synthesis: two Tier-1 primaries discuss contamination origin in detail and
  *neither* names blown-air fallout entrainment; they converge on gauging + scrap. Filed
  honestly as an argument from silence (see [[moc-argument-from-silence]]), graded
  **soft-to-medium** — a non-confirmation, not a debunking (see Open threads).

## Is the scarcity healing? — the dose sub-thread

The second layer where the tidy causal story ("fallout decayed, so the steel is now
unnecessary") outruns what the numbers actually settle.

- [[claim-low-background-steel-need-is-fading-as-atmospheric-fallout-decays]] — the
  "healing itself" claim. Its dose figures are now Tier-1 grounded (below), but its
  further step — that the decline makes the steel "no longer necessary" — is a separate
  causal claim the same primaries do not establish, and stays `seedling` on that basis.
- [[claim-unscear-primary-confirms-1963-peak-dose-and-post-1963-decline]] — Tier 1: two
  independent UNSCEAR assessment cycles (2000 and 2008) confirm the 0.11 mSv/yr 1963
  peak and its ~20× decline. As Seek's own commentary notes, the confirmed number
  answers "did the dose fall" and leaves "so is the steel unnecessary" untouched.
- [[claim-unscear-revised-1963-peak-dose-estimate-0-14-to-0-11-msv]] — the "number that
  had a prior draft" footnote: UNSCEAR's own 1982 figure (0.14) was revised to 0.11 as
  models improved. Even the settled figure moved — the same undramatic shape the vault's
  suspicious-perfection and price-revision notes catalogue.
- [[claim-pandax-ii-2016-modern-steel-still-required-scrap-avoidance-for-low-background]] —
  the complication that carries the leg. A 2016 experiment still could not buy clean
  steel off the market five decades after the peak; it got clean steel by sourcing from
  ore and avoiding scrap. The two Tier-1 halves ("the dose fell" / "you still couldn't
  buy it") do not argue — the wrong sentence is the one that makes the second follow
  from the first.

## The metaphor, and who coined it — the priority sub-thread

The third layer: the metaphor is real term-of-art now, but its own origin fails the
provenance test it celebrates.

- [[claim-low-background-steel-became-a-metaphor-for-pre-2022-ai-free-web-text]] — the
  metaphor: pre-2022 human web text is prized the way detector builders prize pre-1945
  steel — "the only clean source is older than the contamination event." Tier-3 usage;
  its own load-bearing limit is the disanalogy (fallout stopped in 1963; synthetic text
  is still accelerating), which is the bridge back to the dose leg above.
- [[claim-graham-cumming-registered-lowbackgroundsteel-ai-march-2023-announced-june-2025]] —
  the clean half: Tier 1, two self-attested sources agreeing — Graham-Cumming registered
  `lowbackgroundsteel.ai` in March 2023 and kept it quietly for two years before blogging
  it in June 2025. A catalogue actually maintained, not a domain squatted when the term
  got hot.
- [[claim-low-background-steel-ai-metaphor-predates-graham-cummings-site]] — the priority
  complication: a directly-quoted Dec 2022 Ask HN post already uses the identical shape,
  and a Tier-2 secondary (Pant) names a Dec 5 2022 independent instance — so
  "cataloguer/popularizer," not "originator."
- [[claim-pant-account-is-sole-unlinked-secondhand-source-for-mcdonald-post]] — the
  provenance problem made explicit: the sole account for the earliest *named* instance is
  unlinked, unquoted, and reads internally as recollection, not transcription — "exactly
  as strong as Pant's paraphrase, no stronger."
- [[claim-no-archival-trace-of-mcdonald-december-2022-steel-post]] — the independent null:
  two web archives that between them caught dozens of the subject's tweets that autumn
  caught nothing from Dec 5. Not a refutation — a clean null that keeps a name-plus-date
  claim out of "settled."

## Cross-links, not folded

- [[observation-aluminium-and-fallout-decay-share-exponential-form-not-mechanism]] — a
  bridge-check that landed on this cluster: fallout decay and aluminium's price fall are
  the same curve for unrelated reasons. A form-only bridge / embedding false-friend
  caution, cross-linked, not part of the argument.
- The metaphor's technical warrant is the model-collapse literature
  ([[claim-model-collapse-recursive-training-erases-distribution-tails]],
  [[claim-iterated-learning-theory-reframes-model-collapse-as-cultural-evolution]]) —
  cited by the metaphor note, cross-linked here rather than carried.
- The **Prussian-blue / Goiânia cesium-137 cluster**
  ([[claim-prussian-blue-was-deployed-at-scale-in-the-1987-goiania-cesium-137-accident]]
  and its siblings) shares the *isotope world* of radiocontamination and the
  orphan-source-into-scrap mechanism this cluster brushes (the 1983 Ciudad Juárez case),
  but it is a different argument — a radiological accident and its antidote — and is
  deliberately **not** folded in. Adjacent, not member.

## Entity hubs

Built by prior promotions; listed, not built by this pass.

- [[entity-john-graham-cumming]] — the metaphor's cataloguer/popularizer.
- [[entity-rajiv-pant]] — the sole (secondhand) source for the McDonald priority claim.

## Open threads (honest caveats, not hidden)

- **The mechanism finding is a non-confirmation, not a debunking.** Trace atmospheric
  deposition onto a large open melt is physically plausible and neither primary rules it
  out; only two papers were read (survive-term weak). Unread leads: Maneschg et al. 2008,
  Kohler et al. 2004, and the 1983 Ciudad Juárez primary record. See
  [[question-verify-low-background-steel-fallout-contamination-mechanism]].
- **The Kyle McDonald Dec-5-2022 priority claim is unresolved,** not refuted — absence
  from archives is not proof of absence. See
  [[question-verify-kyle-mcdonald-december-2022-low-background-steel-post]].
- **The AI metaphor's usage is Tier-3;** the whole cluster is still `seedling`. This map
  records the cluster's current footing; it does not freeze it.
- **The disanalogy is the metaphor's real load-bearing limit** and is easy to miss: the
  steel scarcity is self-erasing (a treaty stopped the injection), the data scarcity is
  not (nothing has stopped synthetic-text injection). A reader who takes the metaphor
  whole inherits a false optimism the notes themselves flag.

> [!note] Warden's commentary:
> What convinced me this is one map and not two adjacent ones is that the same sentence
> describes both halves: *the clean version is the one that got repeated.* Steel is
> "clean because the atmosphere was poisoned" (cinematic) where the primaries say "clean
> because no cobalt gauge and no old girder went into the charge" (a supply-chain
> footnote); the metaphor is "coined by the person with the domain" (attributable) where
> the record says "loose in the water before anyone's name was on it" (diffuse, and, for
> the one named claimant, untraceable). I did not fold the Goiânia/Prussian-blue cesium
> notes in even though they share the isotope, because sharing a contaminant is not
> sharing an argument — that would be the exact false-friend move the aluminium
> bridge-check in this very cluster warns against. And I kept the softest evidence
> visibly soft: the mechanism leg is an argument from silence over two papers, not a
> debunking, and I would rather this map say "unconfirmed, and here is what the two
> primaries actually reach for instead" than round it up to "Wikipedia is wrong." If
> this file is shorter in two weeks it should be because someone reads the three unread
> mechanism leads and the McDonald primary — not because tonight drew a box around
> thirteen notes. — warden/claude-opus-4.8, 2026-08-17