Everyone has an opinion about garlic breath and almost nobody has a number. The folk answers range from "brush your teeth" to "a day or so", and both are wrong in ways that are easy to demonstrate.
The laboratory answer is better. Researchers have tracked allyl methyl sulfide — the systemic compound responsible for the long tail — in exhaled breath using gas chromatography and mass spectrometry, and have measured it persisting for something in the order of thirty to forty-eight hours after a meaningful dose. That work is solid, and we are not claiming to have discovered it.
What that work does not do is three things. It measures with instruments, not with human noses. It works almost entirely with raw or crushed garlic, which is not how most people eat it. And it measures one channel.
So we built a protocol that does the other thing: a human olfactory measurement, on whole roasted garlic, across two channels at once. Instruments tell you what molecules are present. A panel tells you what a person in the room actually perceives. These are different questions, and the second one is the one that ruins dinner parties.
- Sample: VI-001, our reference garlic — Aglio Bianco, Delfanti, Monticelli d'Ongina. The same sample opens every panel we run.
- Preparation: whole bulb, uncut, sealed in foil, 180 °C for 45 minutes. Uncut matters — cutting the top before roasting ruptures cells and lets alliinase work before the heat destroys it, producing something closer to raw garlic.
- Timing: cooked the day before, served at 18–20 °C, probe-measured at the centre.
- Assessor: one fixed third-party assessor, blind to the dose. Never the subject.
- T0: ingestion on an empty stomach, no toothbrushing. First reading at T0+5 minutes.
- Calibration: before every T0 the assessor calibrates on whole coffee beans from a dedicated jar, anchoring the top of the scale.
- Two measurement points: breath, at 30 cm, on a 2–3 second exhalation; and wrist skin, at 1–2 cm, single inhalation.
- Scale: 0 to 10, relative to that session. The 10 of one session is not the 10 of another.
- Readings: every 4 hours for the first 24, then every 6. Nights declared unmeasured, never silently interpolated.
- Breakfast: fixed, identical, never before 90 minutes from T0.
The results
| Dose | T0 | +4h | +8h | +12h | +16h | +24h | +30h |
|---|---|---|---|---|---|---|---|
| 1.5 g | 9 | 10 | 9 | 7 | 6 | 5 | 0 |
| 3 g | 10 | 10 | 7 | 7 | 7 | 1 | 0 |
| 6 g | 10 | 10 | 9 | 5 | 5 | 3 | 0 |
| Dose | Half-life | Total persistence |
|---|---|---|
| 1.5 g | 24 h | 30 h |
| 3 g | ~18–19 h (interpolated) | 30 h |
| 6 g | 12 h (measured) | 30 h |
Two things in those tables are worth stopping for.
One: thirty hours, every time
The total persistence did not move. Not by four hours, not by two. Three doses spanning a factor of four, and the trace disappeared at the same point in all three sessions.
We have since run a fourth session with a different variety, at the largest dose, and it also ended at thirty hours. Four independent sessions, one number.
This looks like a threshold rather than a dose effect: within this range, the dose determines how loud the signal is, not how long it lasts. Something else — clearance rate, or the perceptual floor at which a human nose stops registering the compound at all — appears to set the endpoint. We are not claiming this holds beyond our range. We are saying it held four times out of four, and that we did not expect it.
Two: the half-life runs backwards
The intuition is that a larger dose should take longer to halve. It did the opposite. Smallest dose, longest half-life: 24 hours at 1.5 g, down to 12 hours at 6 g.
The explanation is the same one we met during the very first self-experiment. At a large dose the assessor's reading sits pinned at the top of the scale while the actual concentration is already falling. The perceived peak is a plateau, the plateau ends in a cliff, and the half-life computed from those readings is compressed. At a small dose there is no ceiling effect, the curve declines gently from the start, and the half-life stretches.
In other words: the perceptual half-life is a property of the measurement, not only of the molecule. It is exactly the kind of artefact a human panel produces and an instrument does not, and rather than hide it we would rather name it, because anyone reproducing this work will meet it on day one.
There is an editorial temptation here to write "less garlic lasts longer" as a headline. It is true of the perception and false of the chemistry, and we are not going to write it.
The wrist told a different story
The skin channel was far less orderly than the breath.
| Dose | T0 | +4h | +8h | +12h | +16h | +24h | +30h |
|---|---|---|---|---|---|---|---|
| 1.5 g | 0 | 0 | 0 | 0 | 0 | 0 | 0 |
| 3 g | 10 | 10 | 0 | 0 | 0 | 0 | 0 |
| 6 g | 0 | 7 | 5 | 0 | 0 | 0 | 0 |
Three doses, three unrelated patterns. What we can say is that the skin channel has a different dose threshold from the lungs — 1.5 g produced twenty-four hours of breath and no skin signal whatsoever. What we cannot yet say is what governs it. Ambient temperature, perspiration, and time since washing are all uncontrolled in this protocol, and any of them could dominate. This is the part of the work that is not finished.
What we got wrong
Two things, both declared, because a protocol you cannot criticise is not a protocol.
The doses were not the ones we planned. We designed the study at 15, 30 and 60 grams. A misread decimal point on the precision scale in session one meant the real doses were 1.5, 3 and 6 grams — discovered during session two. We considered restarting at the intended weights and decided against it: the three sessions were internally consistent, and the accidental range turned out to be the more interesting one. Under two grams of garlic — less than a single clove — produced a twenty-four hour half-life. That is a better finding than anything sixty grams would have given us.
Session two deviated on breakfast. The fixed breakfast was replaced with a different one. The protocol has since been re-fixed and held.
The limits
One subject. One assessor. A scale that is relative within each session and therefore not directly comparable across them in absolute terms. Nights unmeasured. No instrumental control alongside the human readings.
This is not a clinical study and we will not dress it as one. It is a tasting protocol, built the way a tasting protocol is built: fixed variables, a blind assessor, declared deviations, and results published including the ones that embarrass us.
What we would say is this. The chemistry of garlic breath has been described in laboratories for years. Nobody, as far as we can find, had sat down and translated it into a repeatable human sensory protocol — the kind an association of tasters can hand to a panel in another region and expect the same numbers back.
That is the part we did. The comparative variety work is built on top of it, and that is the next article.