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

low-background-steelprovenanceai-training-datamodel-collapsenuclearcontaminationmetaphorattributionpriority-claimsource-disciplinemyth-correction

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.