# Excerpt — Mantel & Bryan (1961): the original "virtually safe dose" (10⁻⁸), self-described as arbitrary

*Nathan Mantel & W. Ray Bryan, "'Safety' Testing of Carcinogenic Agents," **J. Natl. Cancer Inst.
27(2):455–470 (Aug 1961).** NCI Biometry Branch. The founding paper of quantitative cancer-risk
extrapolation — the source of the "virtually safe dose" at a lifetime risk of **1/100 million (10⁻⁸)**, the
number Hutt later moved up to 10⁻⁶ for the DES rule (Rodricks 2019). **Operator-supplied via Taildrop
2026-06-11; held & OCR-verified** in `papers/lateral/mantel-bryan1961_jnci_safety-testing-carcinogenic-agents.pdf`
(+`.txt`). Grade **[CONFIRMED-primary]**. This flips wishlist item #1 (`papers/lateral/_WISHLIST.md`) to held.*

## The thesis: there is no "safe," only a chosen level of risk
> "The problem of determining what dose levels of an agent are safe, e.g., non-carcinogenic, cannot be resolved
> unless one first defines some level of permissible risk, no matter how small, rather than insisting on
> absolute safety." (Summary, p. 455.)

> "How safe is safe? Absolute safety can never be unquestionably demonstrated experimentally… With the
> specification of some level of risk, no matter how small, the possibility of determining whether or not that
> risk is exceeded opens. We may, for example, assume that a risk of **1/100 million** is so low as to constitute
> '**virtual safety.**'" (p. 457.)

## The number is arbitrary — and they say so, repeatedly, in print
> "**Other arbitrary definitions of 'virtual safety' may be employed as conditions require.**" (p. 457.)

From the Summary (p. 455): the procedure requires "**an arbitrary definition of 'virtual safety**,'" plus "an
**arbitrarily** high statistical assurance level and a rule for extrapolation by use of an **arbitrarily** shallow
slope," and "**The arbitrary risks and definitions for so doing may change with circumstances.**"

> "In what follows we will take the defined level of 'virtual safety' to be 1/100 million, the statistical
> assurance level to be 99 percent." (p. 458 — the choice stated plainly, with no derivation.)

## Why it matters to the dossier / essay
- This is the **primary root of the 10⁻⁶ keystone** (`10` §3.5; `06` §5.1). Mantel & Bryan set the "virtually
  safe" target at **10⁻⁸**; FDA's **Peter Barton Hutt**, "after extensive discussions," moved it two orders of
  magnitude up to **10⁻⁶** ("less than 1 in 1 million") for the DES "no residue" redefinition (Rodricks 2019,
  `sources/Rodricks2019-Crump2018_10-6-origin_excerpt.md`). Both numbers are *appointed*, not measured.
- It is a **better, primary** substitute for the secondhand Mantel "we just pulled it out of a hat" anecdote
  (which is **[secondary]**, attributable only to Kelly 1991): the 1961 paper *itself* calls every input
  arbitrary. For the essay's "number they pulled out of a hat" block, quote the paper, not the anecdote.
- Same numerology grammar as Frawley's 0.1 ppm and the factor of 100: when the science cannot supply a
  threshold, a round number is **defined** into existence and an authority adopts it.

## Does the paper cite any *earlier* justification? (operator's question, 2026-06-11)
**Short answer: for the number itself, no — there is no prior authority; they stipulate 10⁻⁸ and call it
arbitrary.** What they anchor to is the prior *method* and the *policy moment*, not a precedent for the risk
level:

1. **The older "arbitrary safety factor" (the factor of 100) — which they ALSO call arbitrary, and say fails
   for carcinogens.** Intro, p. 456: "With a reasonable allowance for an extra margin of safety introduced in
   the form of an **arbitrary 'safety factor,'** the results obtained in laboratory animals can be successfully
   projected to humans. For the most part, modern pharmacology is based on just such usage of laboratory
   animals." They position the probit "virtually safe dose" as the carcinogen-specific successor *because*
   carcinogens "cannot be so confidently projected to man." → a **primary** bridge to essay one: the founders of
   cancer-risk extrapolation label the factor of 100 "arbitrary" in print.
2. **Earliest framing reference = a 1960 Delaney-era congressional cancer hearing.** Their opening premise (every
   agent that causes cancer in man does so in animals) is sourced to **ref (1): G. B. Mider, "The role of certain
   chemical and physical agents in the causation of cancers," Hearings before the [House] Committee on Interstate
   and Foreign Commerce, 86th Cong., 2d Sess., on H.R. 7624 and S. 2197, 26 Jan 1960** (Mider = NCI scientific
   director). So the founding risk-number paper anchors its premise to the **same 1958–60 cancer-legislation
   moment** that produced Frawley's problem and the Delaney Clause — a policy reference, not a statistical one.
   *(Bill identity beyond the reference's own words not separately verified.)*
3. **Borrowed statistical machinery (the method, not the number):** the probit / biological-assay lineage —
   Bryan & Shimkin 1943 (ref 11, the methylcholanthrene mouse data they re-analyze), Bryan 1959 review (ref 5),
   Bliss & Cattell 1943 (ref 2), Finney 1952 (ref 3), Abbott 1925 (ref 8), Cornfield & Mantel 1950 (ref 9).
4. **The no-threshold idea echoes radiation target theory, not chemistry:** the "single-hit / one-particle"
   model is taken from **D. E. Lea, *Actions of Radiations on Living Cells* (1946)** (ref 12); they note it would
   be even more conservative in the low-dose tail than their probit rule.

**Net:** the 10⁻⁸ risk *level* is a free stipulation ("we will take the defined level of 'virtual safety' to be
1/100 million," p. 458); its only "ancestor" is the equally-arbitrary safety factor it supersedes. The lineage
the dossier cares about therefore runs **factor-of-100 (arbitrary) → Mantel-Bryan 10⁻⁸ (arbitrary) → Hutt 10⁻⁶
(chosen) → 1995 ToR 0.5 ppb (a 10⁻⁶ calc)** — round numbers all the way down. **[CONFIRMED-primary]** for items
1–4 (read in the held PDF).

**Links:** [[frawley-dossier-project]] · `papers/lateral/_LIBRARY.md` · ties to `reckoningscience/sections.md`
§9 (no-threshold void) → the candidate "one in a million" block, and `10` §3.5 (10⁻⁶ inside Delaney).