Adrafinil was a wakefulness drug developed back in the late 1970s or so. It worked reasonably well, and interestingly enough, in a different way from most stimulants.1 Inside the body, adrafinil is metabolized by the liver into modafinil. So using adrafinil is inferior to using a straight dose of modafinil both because it stresses the liver and because reportedly you need about 3 times more adrafinil to be metabolized into an equivalent dose of modafinil. Its chief advantages are that all patents on it expired long ago, and it's very rarely proscribed - so it is relatively cheap & easy to get.

Modafinil is better, but the problem is that a number of forms are patented, and in addition, it's a proscribed drug in the US & Canada2. So you can either try to convince a doctor to give you a prescription or deal with dubious online pharmacies or suppliers in countries where modafinil is available over-the-counter. The former way is annoying and you pay full price, and the latter way is unreliable and still expensive - you may be buying at a lower price, but the risks of seizure at the border, shipping, possibly paying for laboratory assays etc. can more than make up for that.

Modafinil costs about $23-12 a day4; non-prescription sources are too unpredictable to say.5


"Modafinil can be reliably determined in plasma and urine (Schwertner and Kong, 2005; Tseng et al, 2005), and is readily absorbed (40–65%, as measured by urinary recovery) after single (Wong et al, 1999a) or multiple oral doses (Wong et al, 1999b), reaching peak plasma concentrations 2–4 h after administration (Wong et al, 1999a). The presence of food in the gastrointestinal tract can slow the rate but does not affect the total extent of absorption. Steady-state plasma concentrations are achieved between 2 and 4 days with repeated dosing. It is highly lipophilic, and approximately 60% bound to plasma proteins, primarily albumin. Major pharmacokinetic parameters are independent of doses in the range of 200–600 mg/day (Robertson and Hellriegel, 2003). The major circulating metabolites modafinil acid and modafinil sulfone do not appear to exert any significant activity in the brain or periphery (Robertson and Hellriegel, 2003). The elimination half-life is approximately 12–15 h (Wong et al, 1999a), and single daily dosing is adequate and common in clinical practice."6


Modafinil has a couple different but closely related benefits - assuming you are a person for whom modafinil works (genetic variation has been linked to responding or not responding). You can sum them up as pretty much it cuts your sleep need by about 2/3s, or it allows you to go without sleep for a day with no penalty. (If you aren't sleep-deprived, it seems to just increase your alertness and energy levels.)

Also, it may make you smarter - even if you're healthy:

"In addition, modafinil (at well-tolerated doses) improves function in several cognitive domains, including working memory and episodic memory, and other processes dependent on prefrontal cortex and cognitive control. These effects are observed in rodents, healthy adults, and across several psychiatric disorders."7

Interestingly, there seem to be some groups for which modafinil does little or nothing, and this ineffectiveness may not be due to counterfeit product or poor self-monitoring, but genetics:

"...Two-time 100 mg modafinil potently improved vigor and well-being, and maintained baseline performance with respect to executive functioning and vigilant attention throughout sleep deprivation in Val/Val genotype subjects but was hardly effective in subjects with the Met/Met genotype."8

The genotype variation specifically refers to the Rs4680 SNP, which is one of the SNPs that services like 23andMe test for. So someone could sign up for 23andMe testing and only start buying modafinil if the results are favorable; considering that 23andMe sometimes holds sales at $100 or $200 and that one could easily spend this much on a single order of modafinil, testing first is not a bad idea at all.

Side effects

Modafinil has a few side-effects. (I omit adrafinil's possible liver damage since it doesn't apply to modafinil.)

Don't believe me? Then try this quote from the abstract of a journal review of modafinil:

"Furthermore, modafinil appears to be well-tolerated, with a low rate of adverse events and a low liability to abuse."11

The most serious reported side-effect I know of is Stevens–Johnson syndrome (SJS); but the FDA report specifies that there was only 1 'possible' syndrome during the clinical trials of around 1,585 people. Presumably if modafinil really did cause SJS at a rate of 1 in 1500, then the hundreds of thousands or millions of users since would have caused ~667 or more cases of SJS, but I do not see any other reported cases of SJS.


Indeed, modafinil's lack of side-effects has led to it being called the most perfect nootropic around, in the etymological sense of a drug which has no bad aspects and only improves the mind. But there is no free lunch, after all. (There may be bargain lunches which we are thrilled to buy, but they aren't free. If they were, why didn't Evolution grab them already? For discussion of how drugs can be bargain lunches, see Drug heuristics.)

Anecdotally, the troll under the bridge for modafinil seems to be that one can develop tolerance, where it no longer has the stimulant or anti-sleep effects that made it so awesome. I have no medical references for this (references welcomed!), but reports of tolerance are fairly widespread anecdotally.

There are two pieces of good news in the anecdotes, though. It seems that it is only tolerance, and not dependence. Caffeine tolerance builds up rapidly, and worse, there is dependence: one declines to below baseline mental performance when one stops using the caffeine; but there seems only to be tolerance to modafinil - I have seen no anecdotes reported performance declines after tolerance and ceasing use. (But one's baseline productivity may be so low compared to productivity while using modafinil that one feels like there is dependence.) Secondly, some anecdotes report that modafinil can be used indefinitely on a weekly or more frequent basis without developing tolerance.

So even this downside isn't very large. There's no evidence that modafinil tolerance (if it exists, whatever it is) is linked to irreversible changes in the brain like use of MAO inhibitors is, so I'll ignore it in the following cost-benefit analysis. A drug which offers a high return on investment but can only be used once a week is still offering a high return on one's investment.


So what's the cost benefit analysis here? We need to take into account the financial expense of modafinil, the biological side-effects, and the benefit of less sleep.


As in Melatonin, we'll assume the value of our time is measured by the minimum wage. We'll also assume modafinil costs $3 a day (this adds 50% to be conservative). Finally, we'll assume the average sleep savings per day is 4 hours - roughly half one's sleep; this seems reasonable since sometimes people will use modafinil to skip that day's sleep and sometimes they'll sleep normally and will take it for greater performance while they are awake.

So, our very first calculation goes minimum wage×hours saved, or 6.55×4, or $26.2. $26.2 > $3, so off-hand modafinil seems worthwhile: the return is 873%.


But what of the other costs? There's the stinky pee, for example. I don't think this is even worth including, but let's assign it 5¢ (how long are you going to be in the bathroom anyway?). Then there are the unknown health effects. Sure, the skeptic says, the studies have turned up little to nothing, but what about the long-term effects?

Well, alright. We'll add it in. Let's consider the worst-case: modafinil is fatal. One human life is usually valued at around 10 million dollars. I personally can expect another 50 years of life. So per day, the expense of a fatal-modafinil habit (we assume the death could any day) would be $, or $548. Ouch. That is noticeably larger than the $26.2 we expected to gain. But this is the worst case; there's no way modafinil is actually 100% lethal.

So the expected value of modafinil is approximately 21 times less than it needs to be. Or, to put it another way, if modafinil had a less than 121 chance of killing you, it could be worth while. Do you think modafinil has a less than $ kill rate? I do.

(We could complicate the analysis by incorporating a discount rate and reduce the value of modafinil by assuming that one only uses it occasionally to avoid building up tolerance, but the basic point is made: there needs to be an improbably high risk to modafinil to neutralize its benefits.)

Suppliers & Prices

Fine, but how are we going to get any modafinil? Buying modafinil is very much a grey-market. And it's a black-market if one wants genuine Cephalon-manufactured Provigil/Nuvigil12.

So Indian generics it is. (India ignores any pharmaceutical patents before a certain year, which includes those on modafinil.)

The general feeling, which my own reading agrees with, is that the generics in descending order of desirability go:

  1. Modapro
  2. Modalert
  3. Alertec

Vendor notes:

It's worth noting that shipping can make a major difference. For example, Airsealed's $22 modafinil seems like an excellent deal - but the price is almost doubled by their $15 shipping. The above table assumes that one orders only one unit of whatever it is; if one wants to order several hundred dollars of modafinil from Airsealed, say, then Airsealed may look much better, and so for that purpose S&H is listed.

With all that in mind, my research into online vendors produced the follow table (last updated 5 October 2010):

mg Amt $ S&H mg/$ Provider
200 200 180 0 222
200 100 88 8 208 United Pharmacies
200 200 193 15 192 ThePharmacyExpress
200 180 195 15 171 ThePharmacyExpress
200 360 413 10 170 Good Health Pharmacy
200 360 415 15 167 EuroDrugStore
200 500 600 0 167 Pharmland
200 100 120 0 167
200 60 58 15 164
200 400 483 10 162 Good Health Pharmacy
100 400 246 10 156 Good Health Pharmacy
200 90 103 15 153 EuroDrugStore
200 200 255 10 150 Good Health Pharmacy
200 240 319 0 150
100 100 60 8 147 United Pharmacies
200 300 408 0 147 Pharmland
150 200 210 0 143
200 240 339 0 142 Pharmland
200 90 114 15 140 EuroDrugStore
200 120 171 0 140 Pharmland
200 30 28 15 140 Pharmland
200 60 75 15 133 EuroDrugStore
200 30 45 0 133
120 200 170 20 127
200 90 128 15 126 Pharmland
200 60 83 15 122 EuroDrugStore
90 200 131 20 120
200 60 86 15 119 Pharmland
60 200 89 20 111
100 100 84 10 106 Good Health Pharmacy
200 30 45 15 100 EuroDrugStore
100 300 329 0 91 Pharmland
100 90 85 15 90 EuroDrugStore
100 120 140 0 86 Pharmland
100 240 279 0 86 Pharmland
30 200 50 20 86
200 20 32 15 85 EuroDrugStore
100 60 62 15 78 EuroDrugStore
10000 N/A 118 15 75 Airsealed
100 90 106 15 74 Pharmland
200 20 40 15 73 EuroDrugStore
100 60 71 15 70 Pharmland
200 50 142 15 63 Airsealed
200 50 155 7 62 Nubrain
100 30 39 15 56 Pharmland
200 10 23 15 53 Airsealed
200 30 105 7 53 Nubrain
100 50 104 7 45 Nubrain
100 20 33 15 42 EuroDrugStore
100 30 88 8 31 United Pharmacies
100 30 102 16.5 25 QHI
100 30 125 10 22 AuraPharm
100 30 158 15 17 Biogenesis AntiAging
100 30 160 15 17
100 20 180 16.5 17 QHI
100 30 168 15 16 Biogenesis AntiAging

Ordering behavior

Internet suppliers are not known for their transparency or reliability; and the few retailers online of modafinil which are trustworthy tend to be more expensive (as a look at the above chart shows) or they require prescriptions for US buyers (like the Canadian online pharmacies). So we are faced with a classic problem of risk: given that any supplier may shaft us, how should we rationally order products to reduce our risk?

For modafinil we can come up with a few distinct ordering scenarios:

  1. a one-shot order
  2. multiple orders, with oneself as an oracle (perhaps one has used genuine Provigil in the past and is sufficiently rational to avoid mistakes caused by things like the placebo effect)
  3. multiple orders, with laboratory assays as the oracle

One-shot ordering

For #1, the simple answer is best. You decide roughly how much you trust each supplier based on factors like how their website looks, what descriptions of them you found online, whether they seem 'too good to be true', etc - what is the percentage that they will pay? An obvious scam is 0%, a totally trusted source is 100% and everyone else is a shade of gray. Then you multiply their trustworthiness against their modafinil/mg price. Whomever comes out on top, you order from.

So, if I thought EDAndMore had only a 60% chance of delivering real modafinil rather than caffeine pills, I would multiply their unit-price 222 against the trust-penalty of 0.6 and get 133.2 (222×0.6). And then I might decide United Pharmacies smells a bit better and give them a 70% chance, for a total of 145.6 (208×0.7); and perhaps Airsealed gets a whopping 95% trust from me, for a total of 71.25 (75×0.95). In this scenario, Airsealed is highly trustworthy compared to the other two, but because Airsealed costs nearly 3 times its competitors, it comes out poorly against United Pharmacies; UP is only a little more expensive than EDAndMore, but has a noticeable edge over EDAndMore in trust and so wins. For Airsealed to win, I would have to trust both EDAndMore and United Pharmacies somewhere less than 33%.

I order and I get... something. This is where the one-shot strategy ends. You calculated, did your best, and either won or lost.

If you plan to order again (and modafinil is such a useful drug that it is hard to see why one would not plan to use it indefinitely as tolerance and finances permit), you have a chance to update based on how the first order went.

Presumably one has learned from one's own experiences whether the product was modafinil or not. Modafinil is fairly distinct from caffeine; one gains tolerance to caffeine very quickly, and caffeine will not keep one going overnight with little mental penalty. So we'll assume one knows.

If it was modafinil, then great. You've found an honest supplier. You might want to order a few years' supply while you can. If you really want an optimally cheap supply, you can try some of the other suppliers. Naturally you will not try any supplier whose per-unit price is more expensive, because you trust your current supplier 100% and must distrust the others at least a little, so the list of possible suppliers has been cut down.

Ordering with learning

If it was not, then you have a problem. Are you the vengeful sort who assumes that suppliers are all rotten? Then you blacklist them and order from the next cheapest supplier (by unit-price adjusted for perceived risk). If you're not vengeful, if you believe the suppliers who say that they don't have control of their supply chain and sometimes bad products slip in, then you have to do more work.

If we interpret our distrust probability as instead 'what percentage of delivered product is genuine', then to continue the previous example, my trust probability would drop only a little if EDAndMore sent me caffeine/fake modafinil. Why? Because I estimated a 60% chance of them sending me modafinil, and 60% is another way of saying there's a 40% chance they'll send me not-modafinil. 40% chances happen quite often and I wouldn't be very surprised if one happened when I ordered. On the other hand, I expected mostly pure product out of Airsealed (95% of their shipments being good), so if Airsealed sent fakes to me, then I would be quite shocked - 5% chances don't happen all that often (it's like rolling a 20 in D&D). How much do I knock down my estimate of Airsealed? By more than EDAndMore, obviously. But how much more? Well, you have to apply Bayes' theorem. (One basic property of Bayes' theorem is that extreme probabilities are hugely damaged by opposed observations, while equivocal probabilities like 51% take a lot of data to knock down. For a good explanation, see "An Intuitive Explanation of Bayes' Theorem".)


The former is a little complex, so we'll simplify down again. Here's our scenario: 95% of the orders delivered by a 'good' supplier will be real (all modafinil); but bad suppliers have only 5% of the orders be real. (Apparently this isn't all that unrealistic; whether it's because supply chains are unreliable or deliberate profit-maximizing behavior, often neither good or bad suppliers ship all fakes or all genuine modafinil.)

Now, supplier Z, whom you had calculated was probably good with 90% probability, sent you a fake. Keeping in mind that you might just be one of the unlucky 5%, now how much do you think that Z is good?

A rearrangement of Bayes' theorem from the end of Eliezer Yudkowsky's "Intuitive Explanation of Bayes' Theorem" (he explains its derivation if you don't trust me):


How confusing and intimidating! Where does one start, with all the differing entries?

Let's break it down step by step. If you didn't read either Wikipedia or Yudkowsky, you should have, but remember the pipe is read backwards: P(foobar) means 'how likely foo is now that I have observed bar' (you could mentally rewrite it to something like P(barfoo)). x represents our observation, whatever it is. In this case, it's the not-modafinil, the fake pills. a represents our new, reduced, belief that Z is a good place to order from. So at the beginning we can make a few definitions:

If you look, the right-hand side of that equation has exactly 4 pieces in its puzzle:

  1. P(a)

    This is something we already know, 'probability of being a good supplier'. Well, we specified a few paragraphs above that we had somehow concluded that Z was a good supplier with 90% odds.
  2. P(¬a)

    This is the opposite of the previous. Now logically, if something has a 90% chance of being true, then obviously it has a 10% chance of not being true. Either one or the other. So this is simply equal to 10%.
  3. P(xa)

    Remember, we read the pipe notation backwards, so this is 'the probability that a good supplier (a) will send us fakes (x)'. We also said that good suppliers send 95% good orders; by the same logic as above, that's another way of saying they send 5% fakes. So this is simply 5%.
  4. P(x¬a)

    Finally, we have 'the probability that a bad supplier will send us fakes'. We said bad suppliers send 5% good orders, so again, subtracting from 100 and we know they must send 95% fakes. So this is simply 95%.

To put all these definitions in a list:

  1. a = good supplier
  2. x = fakes
  3. P(a) = probability of being a good supplier = 90% = 0.90
  4. P(¬a) = probability of being a bad supplier = 10% = 0.10
  5. P(xa) = probability a good supplier will send fakes = 5% = 0.05
  6. P(x¬a) = probability a bad supplier will send fakes = 95% = 0.95

We substitute in to the original equation:


32% seems like a reasonable answer. Intuitively, I'd expect my trust to drop considerably below 50%, but still noticeably above 5%, and 32% is well within that range.

Phew! So, now we have a full system for situation #2. First, go through the expected utility calculation for all the prices you've gathered in which you trade off risk versus price. Then order, test for modafinil or not-modafinil, and do a Bayesian update on the supplier. Rinse, and repeat. This procedure terminates if there are any good suppliers (suppose you luck out and the second supplier you test, #4 on the per-unit list, is honest; now you only need to test 3 more suppliers).

Ordering when learning isn't free

How do we extend this procedure to handle situation #3, where we no longer trust ourselves to test the supposed modafinil (for free)?

Let's assume it costs $1000 to assay some pills and like in #2, we'll assume the assay is infallible. Then we just repeat the #2 procedure except with the assay replacing our own subjective testing. Fair enough, right?

But a wrinkle comes to mind. $1000 is a lot. Quite a lot, really. You could buy a lot of modafinil with $1000; even if we played it safe and bought from Nubrain, $1000 would get us 62×1000, or 62,000 mg, or at 200mg a day, 310 days' worth14.

Are we sure we want to spend that? I mean, what if the price difference between the last 2 suppliers is just a few pennies - would we still want to spend $1000 just to be sure? Even if we plan to buy for years, decades, or the rest of our life, that $1000 might not be worth spending to optimize and save a few pennies. We could ask ourselves - is the possible penalty of a few pennies every order over our lifetime worth the $1000 right now? As it happens, there's a nifty formula for this.

We want to compare an infinite stream of small savings against a single large expenditure. The savings would make us slightly wealthier each time period. In fact, you could imagine that we were actually discussing savings bonds or annuities: is the small payoff each year worth the large upfront investment? In economics terminology, the question is whether the 'net present value' of that payoff stream is larger than our investment.

Net present value can be approximated based on knowing one's discount rate (which is a percentage) and then plugging in the numbers to the following formula:


What is my discount rate? If I introspect and ask myself, would I prefer $150 in a year to $100 right now; I say yes, so I know my discount rate is less than 0.5/50%; would I prefer $125 to $100 now? Yes, so it's less than 0.25/25%; and so on down to around $107 versus $100. So let's say 7% for me.

What do our two suppliers look like? Maybe one charges me $220 for a year's supply and the other, the one I haven't yet tested, supplies it for $200. So $20 a year is at stake. Let's plug it in:


But it would cost us $1000 to assay and find out whether there was a savings or not! Clearly it's not a good idea to spend the money for the assay. $20 a year is just not enough.

As it happens, at a discount rate of 7%, we would have to potentially save about $68 before the assay became a good idea. (For $20 a year to be worthwhile, your discount rate has to fall down to around 2%, and very few people are that patient and self-sacrificing!) If you look back at the supplier chart, it runs the gamut from 222 mg per dollar to 16 mg; so for the first few purchases the assay might well be worthwhile.

Thus we solve situation #3. You apply the #1 method to decide at any point which supplier to order from next by their risk-adjusted unit-price; then you update the risk-adjusted unit-price through #2's idea of using Bayes' theorem; then if testing is not free, you decide to stop testing and stop searching suppliers when, as given by #3's present value formula, the testing (assays) start costing more than one can hope to gain.

A final thought about modafinil:

"I am not interested in talking about my ideas. I am interested in your application of them to your life."15

It goes without saying that #1-3 are all simplified models which may not apply to every situation; but at some point, the would-be users of nootropics must start thinking for themselves...

Coordination problems: assaying

"On the one hand, information wants to be expensive because it's so valuable. The right information in the right place just changes your life. On the other hand, information wants to be free, because the cost of getting it out is lower and lower all the time. So you have these 2 things fighting against each other."16

The obvious response to the price of laboratory assays is to somehow distribute the cost among interested persons.

Any user-based service would founder, I think. Look at my footnote about EDAndMore - there is a link to user feedback about EDAndMore, but it is littered with what smells like fake reviews. But how could I possibly prove that? If people really began to use the service, then the sellers have great incentive to corrupt it. One of the best user-generated sources is, and even there, if you look through threads, you will find endless comments by users banned for being shills or otherwise nonkosher.

Now, it's easy to avoid this problem and create a reliable rating system for modafinil sellers. All one has to do is order every few months, and send the product to a lab for assays. (This would be the Consumer Reports model for drugs.)

But this requires a spare couple thousand dollars, and is a public good susceptible to the free rider problem. I certainly don't have $2000 to buy $100 of product (plus S&H) from 10 retailers and then pay for 10 lab assays. So really, the best I can do is catalogue what they advertise.

It's a public good, because while rationally, modafinil buyers should be willing to pay a few dozen or even hundred dollars to obtain access to the results of dozens of assays of purchases (avoiding, in the long run, spending thousands on counterfeits), and there are many thousands of buyers, plenty to collectively pay the vendor enough to cover the purchases & assays and provide a profit, it would be even more rational for one buyer to find the result and resell the summarized info for less, or just release it for free - ' is the best' - and thereby ruining the vendor.

Knowing how vulnerable they are, no vendor will go into the business to begin with - keeping everyone ignorant and making everyone worse off than if they could just have agreed not to share that result. (Consumer Reports gets away with it because it covers so much, there's a lot of detail, and they seem to be largely funded by their automobile division.)

One of the canonical solutions to a public good problem like this is called the assurance contract: participants legally bind themselves to contribute a sum of money if there are enough other such promises that the necessary threshold is passed. One of the most successful facilitators of assurance contracts is the online site Kickstarter; but of course, it is illegal to order modafinil without a prescription, and it is highly unlikely Kickstarter would permit an assurance contract for modafinil assaying to be created or completed.

External links

  1. See Minzenberg et al:

    "It is currently approved by the United States Food and Drug Administration as a schedule IV agent to treat excessive daytime sleepiness in narcolepsy, shift work sleep disorder, and obstructive sleep apnea/hypopnea syndrome. It has been popularly categorized as a psychostimulant due to its wake-promoting properties. However, it has shown a number of effects on physiology and behavior in both animal models and in humans, which suggest a divergent mechanism of action compared to amphetamine (described in detail below). This includes a lower liability to abuse, and a lower risk of adverse effects on organ systems such as the cardiovascular system."

  2. Admittedly, Schedule IV drugs like modafinil are not one of the scheduled drugs which send people to federal pound-you-in-the-ass prison, but it still is troublesome.

  3. One American prescription user I spoke with said that his prescription was for 200mg pills; each box cost around $50 for 30 pills. He took one in a day, or about $1.70 (120 mg/USD).

  4. from Businessweek's "The 10Q Detective: Losing Sleep Over Cephalon":

    "Supplementing the coupon efforts: discounted pricing. The prescription cost to the patient for a 30-day supply of daily recommended dosages of Provigil (200 mg) and Nuvigil (150 mg) is $361 and $295, respectively, a savings to cash-paying patients and health-care payers of 23%, according to pharma information site Destination Rx."

    36130=12.03; although it's unclear what of this is the actual price to the consumer.

  5. This is just conventional wisdom, echoed by many websites. An example:

    "When you buy modafinil online, make sure the company you are buying it from has a good track record and delivers orders promptly and securely. Modafinil is available in 100-mg and 200-mg doses and in packets of 30 tablets to 100 tablets. Prices may vary depending on where you are in the world. But in the United States, prices may range from $50 to more than $300. Make sure that you compare online modafinil prices to get a better deal."

  6. Minzenberg again.

  7. "Modafinil: A Review of Neurochemical Actions and Effects on Cognition"; Michael J Minzenberg, and Cameron S Carter. Neuropsychopharmacology (2008) 33, 1477–1502; local copy

  8. Quote from the abstract of "Pharmacogenetics of modafinil after sleep loss: catechol-O-methyltransferase genotype modulates waking functions but not recovery sleep", March 2009

  9. See, eg.,

  10. "Urine smells of suplhur"

  11. "Modafinil: A Review of Neurochemical Actions and Effects on Cognition"; Michael J Minzenberg, and Cameron S Carter. Neuropsychopharmacology (2008) 33, 1477–1502; local copy

  12. The real deal runs upwards of $10 per pill; so we won't even bother to include it in the price table

  13. But Airsealed itself is a cautionary anecdote; it may have a very good reputation, but it's still not hard to find bad reviews. This is a point that should not need to be made, but I still will anyway: everything in this page deals with risky & uncertain probabilities. If you want guarantees and certainty, go read a logic textbook.

  14. Which is more than a year's supply because taking it every single day is asking for tolerance to build up.

  15. Miyamoto Musashi, "Book of Water", _Book of Five Rings

  16. Stewart Brand to Steve Wozniak (1984; First Hacker's Conference)