1,560 materials have a boiling point and no longevity figure. Does the boiling point substitute?

A supplier datasheet usually gives you a boiling point. It often does not give you a longevity figure. In this database that gap is most of the problem: of 4,620 accessible materials, 1,560 have an atmospheric boiling point and no substantivity figure at all — a third of everything you can buy.

So the practical question is not whether volatility relates to longevity. It is whether the number you have can stand in for the number you want.

What the comparison needs before it means anything

Neither field is usable as it sits, and the cleaning matters more than the result.

Substantivity carries the concentration it was measured at. Of 1,778 figures, 1,601 are neat at 100% and the rest are at 20%, 10% or 1%. It also carries censoring: 180 of the values surviving the first filter read > 48 hour(s) or < 1 hour(s) rather than a number.

Boiling point is often reported twice, once under vacuum and once at atmosphere, and 131 of the atmospheric values are marked (est) rather than measured:

isoeugenol   125.00 °C @ 14.00 mm Hg | 266.00 °C @ 760.00 mm Hg

Applying all of it — neat measurement, atmospheric pressure, no censored values, no estimated boiling points, one row per CAS — leaves 770 rows. Not 770 compounds in any strict sense: several are essential oils and absolutes, which are mixtures with boiling ranges rather than single molecules, and "100%" describes the concentration it was smelled at, not its purity.

Two thirds of the boiling points are reported as a range, 250.00 to 258.00 °C, and the figures below take the lower end. Taking the midpoint moves Spearman by less than 0.01, so that choice does not carry the result.

The ranking holds

Atmospheric boiling point	Rows	Median substantivity	Middle half
37–190 °C	154	8 hours	4–24 hr
190–222 °C	154	16 hours	8–52 hr
223–250 °C	154	44 hours	19–144 hr
250–277 °C	154	124 hours	48–284 hr
277–497 °C	154	252 hours	100–400 hr

Monotonic across every band, about thirty-fold from bottom to top, Spearman ρ 0.62.

So the top-heart-base ordering is not folklore. Group materials by boiling point and the groups come out in longevity order.

And then it stops being useful

The medians hide that every band spans nearly the whole range. The lowest runs from 1 hour to 400; the third from 3 to 746. Nineteen rows boil below 200 °C and last 200 hours or more; seven boil above 280 °C and are gone within a day.

ρ of 0.62 says the two rank together strongly. It does not say how tightly they track, and squaring it into a share of variance borrows a reading belonging to a different statistic. The band ranges are the honest measure of the looseness. A boiling point tells you which end of the palette a row sits at. It does not tell you how long it lasts.

The numbers are coarser than they look

Substantivity is rounded hard. In the clean set the commonest values are 4 hours (10.3%), then 400 hours (8.8%), then 12 and 8. In the highest band, 40 of 154 sit at exactly 400.

Worth being careful about what that implies. It does not push the thirty-fold figure in a knowable direction: those 400s all sit above their band's median, so replacing them with any larger number leaves the median where it is. An earlier version of this article claimed the pile-up made the ratio a floor. It does not, and the arithmetic is simple enough that the claim should not have been made.

What the rounding does mean is that these are category labels wearing numeric clothing. The evaporation article already says to treat substantivity as an ordering scale rather than absolute values, and that survives everything here.

Why the two come apart

Substantivity as measured here is how long a smelling strip still smells of the material. That depends on how fast the molecule leaves — which boiling point speaks to — and also on how little of it you can still detect, which boiling point says nothing about.

A material that evaporates quickly but remains perceptible at very low concentration will outlast its boiling point. One that evaporates slowly but has a high detection threshold will disappear while there is still plenty of it there. That is reasoning about the mechanism rather than something measured here, and it is offered as the likely shape of the residual rather than as a result.

How to use it, and how not to

• Use it to place a group, not a material. The bands rank; individual rows inside them do not. The starter palette map sorts by measured substantivity above 72 hours for exactly this reason — of the rows here boiling between 289 and 291 °C, one lasts 40 hours and another under 1, so a boiling point cannot assign a single material to the base end.
• Do not build a formula ratio on it. The within-band spread is wider than the difference between adjacent bands.
• Prefer a measured figure when one exists. For any row that already carries an uncensored substantivity value, the boiling point adds nothing.

The longevity figures used here are recorded neat at 100%, which is not the condition anyone wears a fragrance in — what that number is and is not measuring works through the gap between a materials figure and a wearer's experience.

A note on how this nearly went wrong

The first version of this analysis came out non-monotonic: the second boiling point band showed a lower median than the first, which reads as the folk model breaking down among volatiles. An interesting result, and worth an article.

It was a parsing error. Filtering for strings containing 760.00 mm Hg passed materials that also carried a vacuum figure, and taking the first number in the string took the vacuum one. Isoeugenol entered the analysis at 125 °C instead of 266.

Fixing that produced a monotonic table on 1,113 rows, and the version of this article built on it stated a ratio of 61-fold and claimed the 400-hour pile-up made that a floor. A review found the second claim is arithmetically false, and that the 1,113 still contained 180 censored values, 131 estimated boiling points and 54 duplicate CAS entries treated as separate materials. Cleaning those leaves the 770 rows above; the ordering survives, the ratio drops to about thirty, and the floor claim is withdrawn.

Worth stating in full because none of the three errors produced noise. Each produced a cleaner, more quotable number than the truth — which is when to go back and look at what the raw field contains.

The same field caught me a third time, when I ranked a one-tenth dilution against neat materials without noticing.