Same word — "impaired" — but alcohol, cannabis, and fatigue leave different marks. This lesson turns the domain map from Lesson 1 into three distinct patterns, and shows why a battery can sometimes name the cause, not just flag it.
Why this matters for your build
If every impairer degraded the same domains by the same amount, your assessment could only ever answer
"impaired: yes/no." Because they don't, a well-chosen battery can lean toward "impaired, and it
looks like fatigue" vs. "…looks like cannabis." That extra signal is a product differentiator —
but only if you pick tasks that expose the differences.
Recall first
Before new material — retrieve Lesson 1. Don't peek.
Why can't a single task detect "impairment" in general?
A task is a bet on one domain. To cover several impairments you need several bets — a battery. That's the whole reason fingerprints (this lesson) are useful.
Three fingerprints
Each cause hits a characteristic set of domains, over a characteristic time course, with a characteristic behavioral tell. Bar length = how strongly that domain is hit.
Divided attention fails at the lowest doses; inhibition and working memory go as the dose climbs —
which is why intoxicated people make impulsive errors.[1] The
behavioral tell is dangerous: at higher BAC people underestimate their own impairment.[2]
It overlaps alcohol on working memory, attention, and RT — but adds time-perception distortion,
which most other impairers don't touch, making it close to a signature.[3] Time course:
onset 30–60 min, peak ~90 min–2 h, largely gone by ~4–5 h; shorter in frequent users due to
tolerance.[4] Behavioral tell — the opposite of alcohol: cannabis users tend to
overestimate impairment and compensate (slow down).[2]
And crucially, blood THC tracks impairment poorly — so a behavioral test is more defensible than a
chemical one here.[2]
Sleep deprivation — the vigilance breaker
Vigilance
hit first & hardest
RT variability
lapses, not slowing
Working mem
later
Higher-order
spared until late
Distinctive trait: wake-state instability — erratic lapses, not uniform slowdown.[5]
This is the most distinct fingerprint. Fatigue doesn't slow everything evenly — it makes performance
erratic: mostly-normal responses punctuated by sudden lapses
("wake-state instability").[5] The severity is startling: roughly
17 h awake ≈ 0.05% BAC, and 24 h awake ≈ 0.10% BAC on performance.[6]
This is why variability and lapse count — not average reaction time — are the signals to
compute for fatigue.
The comparison at a glance
Cause
Domains hit hardest
Behavioral tell
Tracks a chemical marker?
Alcohol
Divided attn → inhibition → RT
Underestimates; impulsive
Yes — BAC predicts well
Cannabis
Working mem + time perception
Overestimates; compensates
No — blood THC ≠ impairment
Sleep loss
Vigilance (lapses, variability)
Micro-lapses; state instability
No chemical marker at all
The design takeaway
To discriminate these — not just detect — your battery wants: a sustained-attention / PVT-style
task scored on lapses & variability (catches fatigue's signature), a divided-attention or
Go/No-Go task (catches alcohol's inhibition loss), and a time-estimation task
(catches cannabis's near-signature). Average reaction time alone would blur all three together.
The honest limit
Fingerprints have smudges. The depressant family overlaps: benzodiazepines produce an
alcohol-like profile (slowed psychomotor speed, impaired attention and working memory), so telling alcohol
from benzos on cognition alone is the hardest case.[7] And
stimulants can invert the logic — they may transiently sharpen simple reaction time
while impairing judgment and inhibition, so an unexpected improvement on one metric is itself a
fingerprint worth flagging, not ignoring.
Check yourself
Which metric best exposes fatigue specifically?
Fatigue is wake-state instability — mostly-fine responses broken by lapses. An average hides the lapses; variability and lapse count reveal them.
A user overestimates their impairment and slows down. Best guess?
The compensate-and-overestimate tell is characteristic of cannabis; alcohol shows the opposite (underestimate, act impulsively).
Behavioral testing is especially valuable for cannabis because...
Unlike alcohol's tight BAC coupling, blood THC is a weak proxy for impairment — so a test of the effect (behavior) beats a test of the substance here.
Your single win
You can now sketch three fingerprints from memory — alcohol (dose-coupled, inhibition,
underestimates), cannabis (working memory + time distortion, overestimates), fatigue
(vigilance lapses, variability) — and say which task and which metric exposes each. That's the raw
material for a battery that discriminates. Lesson 3 asks the harder question: how do you know your detector is
actually trustworthy and not just noise dressed up as a game?
I'm your teacher — ask me anything. Want to war-game a specific case — "our users
test at 6 a.m., could we be catching fatigue and mislabeling it?" — or figure out which of your current tasks
maps to which fingerprint? Bring it to the chat.
References
[1] Dose-Related Effects of Alcohol on Cognitive Functioning, PLOS ONE (2013).
[2] Pearlson et al., Cannabis and Driving (2021) — alcohol vs. cannabis behavioral contrast & THC–impairment dissociation.
[3] Figueiredo et al., Cannabis and Cognitive Functioning, Frontiers in Psychiatry (2021).
[4] McCartney et al., Duration of Neurocognitive Impairment With Cannabis, Frontiers in Psychiatry (2021).
[5] Van Dongen et al., The cumulative cost of additional wakefulness, Sleep (2003); Alhola & Polo-Kantola (2007).