FISHER PAYKEL 900MR860 Temperature Probe Instructions - 900mr869
Provides a Function instance for types with Bounded and Enum. Use only for small types (i.e. not integers): creates the list [minBound..maxBound]!
Prints out the generated test case every time the property fails, including during shrinking. Only variables quantified over inside the verboseShrinking are printed.
Bud Industries CU-387 enclosure with Vigortronix VTX-146-030-212. The transformer converts 120VAC (parallel windings) to roughly 30VAC, or 208VAC (series windings) to roughly 25VAC. I roughened the plastic with sandpaper and epoxied 6mm M3 standoffs in place; hopefully that'll hold well enough.
These modifications are intended to make the condenser compartment run hotter, which may shorten the life of some components, but I have no idea which ones or by how much. I at least have a couple bits of evidence that this might not be a terrible idea:
If you are using checkCoverage as part of a test suite, you should be careful not to set certainty too low. If you want, say, a 1% chance of a false positive during a project's lifetime, then certainty should be set to at least 100 * m * n, where m is the number of uses of cover in the test suite, and n is the number of times you expect the test suite to be run during the project's lifetime. The default value is chosen to be big enough for most projects.
$forAllProperties has type (Property -> IO Result) -> IO Bool. An example invocation is $forAllProperties quickCheckResult, which does the same thing as $quickCheckAll.
Nondeterministic choice: p1 .&. p2 picks randomly one of p1 and p2 to test. If you test the property 100 times it makes 100 random choices.
Large x: by default, QuickCheck generates Ints drawn from a small range. Large Int gives you values drawn from the entire range instead.
Invoke as $(polyQuickCheck 'prop), where prop is a property. Note that just evaluating quickCheck prop in GHCi will seem to work, but will silently default all type variables to ()!
Tests a property and prints the results and all test cases generated to stdout. This is just a convenience function that means the same as quickCheck . verbose.
Reverse engineered IR protocol: all marks are the same length, with data encoded in the spaces. There appear to be 32 bits per packet, organized as [id, ~id, code, ~code]. I tried sending all 256 codes, hoping to find undocumented features like "set power to X" or "set mode to X", but alas, they were all just duplicates of the standard 6 buttons, so understanding the protocol is pretty useless.
Note: if the property times out, variables quantified inside the within will not be printed. Therefore, you should use within only in the body of your property.
The class of properties, i.e., types which QuickCheck knows how to test. Typically a property will be a function returning Bool or Property.
Test all properties in the current module. This is just a convenience function that combines quickCheckAll and verbose.
Whynter arc 14shreview
To attach the 24VAC wires, I took some 3.5mm screw terminal blocks (TE 1776275-6), bent the pins sideways, and soldered them to form a bus. The exposed metal was wrapped with electrical tape.
I wasn't able to get an answer for this from my engineers about the effect on sealing the gap will have on efficiency. They did say that when the units are tested, they are tested with this gap to produce that SACC rating. There's no modifications to the units done while testing. The efficiency of the units includes the air gap.
Should we replay a previous test? Note: saving a seed from one version of QuickCheck and replaying it in another is not supported. If you want to store a test case permanently you should save the test case itself.
Each use of label in your property results in a separate table of test case distribution in the output. If this is not what you want, use tabulate.
Note: for technical reasons, the test case is printed out after the property is tested. To debug a property that goes into an infinite loop, use within to add a timeout instead.
Tests a property, using test arguments, produces a test result, and prints the results and all test cases generated to stdout. This is just a convenience function that combines quickCheckWithResult and verbose.
These functions test all properties in the current module, using Template Haskell. You need to have a {-# LANGUAGE TemplateHaskell #-} pragma in your module for any of these to work.
By default up to 100 tests are performed, which may not be enough to find all bugs. To run more tests, use withMaxSuccess.
If you are using a recent GHC, there is a default definition of coarbitrary using genericCoarbitrary, so if your type has a Generic instance it's enough to say
When you use checkCoverage, QuickCheck uses a statistical test to account for the role of luck in coverage failures. It will run as many tests as needed until it is sure about whether the coverage requirements are met. If a coverage requirement is not met, the property fails.
In order to test these modifications, I installed the air conditioner in a somewhat-sealed closet with a 1.5m² volume. I initialized the temperature to roughly 72°F, set the A/C to 61°F (the minimum value), and let it run for about 3 minutes until the blower shut off.
This was repeated 3 times in the factory configuration (red lines), 3 times with the gaps sealed (blue lines), and finally 2 times in "single hose mode" (black lines) with the intake sucking directly from the closet space. I logged the temperature using a phone pointed at a thermometer, running Open Camera on 10-second repeat:
I captured the remote codes using LIRC's mode2, which provides mark/space times in microseconds, compatible with the sendRaw() function mentioned previously.
There is a fair bit of boilerplate in the code above. We can avoid it with the help of some generic functions. The function genericShrink tries shrinking a term to all of its subterms and, failing that, recursively shrinks the subterms. Using it, we can define shrink as:
Note: for technical reasons, the test case is printed out after the property is tested. To debug a property that goes into an infinite loop, use within to add a timeout instead.
It is worth spending time thinking about what sort of test data you want - good generators are often the difference between finding bugs and not finding them. You can use sample, label and classify to check the quality of your test data.
Each use of collect in your property results in a separate table of test case distribution in the output. If this is not what you want, use tabulate.
There are velcro strips along the top (with opposing velcro hot-glued to the A/C), and on the back so the burrito can attach to itself.
The circuit board is a Proto Advantage SBB206. It fits the 15-row ESP32 dev board exactly, though it'd be nice if the dev board were 1 pin narrower. Drilling holes in the power rails lets them hold general-purpose components.
Note: for technical reasons, the test case is printed out after the property is tested. To debug a property that goes into an infinite loop, use within to add a timeout instead.
Each unit consumes about 10.5 amps, so it's important to map out which circuit breakers feed which outlets, and avoid sharing a circuit with a high-drain appliance like a microwave or electric kettle. Ideally, they should also sit on opposite legs of the main breaker. So far, I've had no issues sharing a 15 amp circuit with a refrigerator or computer.
That's because GHCi will default any type variables in your property to (), so in the example above quickCheck was really testing that () is equal to itself. To avoid this behaviour it is best practise to monomorphise your polymorphic properties when testing:
Check that all coverage requirements defined by cover and coverTable are met, using a statistically sound test, and fail if they are not met.
Uniformly generates a fractional number. The number can be positive or negative and its maximum absolute value depends on the size parameter.
It's possible to set a specific temperature by sending enough DOWN commands to reach the minimum value, followed by enough UP commands to reach the desired temperature. In theory, the controller could always use 16C for cooling and 32C for idle, but I prefer 18C-27C, because it only needs 9 clicks (shorter beep), and the user can manually raise the cooling point to prevent a smaller room from getting unreasonably cold. Manual adjustment is only possible in celsius mode due to the temperature corruption bug, so it's unfortunate that unplugging the power reverts to fahrenheit. My best idea was to initialize cooling mode by sending enough UP commands to max out the fahrenheit range, and likewise for DOWN commands in heating mode. All adjustments after this point assume celsius, so the temperature will never get cold enough to cool, or hot enough to heat, until the user restores celsius mode and reboots the ESP32. In other words, it seems better to never work than sort of work.
Pushing and bending hundreds of staples was quite a lot of work. I think I would recommend that a sane person stop with the initial "folded blanket and tape" solution, since it just takes a few minutes, and MLV isn't tremendously better.
If you want to use monomorphic in the same file where you defined the property, the same scoping problems pop up as in quickCheckAll: see the note there about return [].
The area that you've taped up is considered an overflow release. You are perfectly welcome to keep that area taped up though as it shouldn't cause an issue. My engineers have informed me that it's OK to do this.
After sealing these gaps, I could feel more air moving through the intake port, and less air coming from the exhaust when I briefly obstruct the intake.
It's possible to synchronize the fan speed by briefly running DEHUMIDIFY (which forces the fan to LOW), but I decided not to do this, and just let the user choose the fan speed.
Return a value in the witnesses field of the Result returned by quickCheckResult. Witnesses are returned outer-most first.
I connected an ESP32 microcontroller to the "compressor relay stage 1" (Y1) and "heat pump changeover valve" (O) outputs of a Nest thermostat, with "auxiliary heat" driving a relay for my baseboard heaters. (I've tested that the Nest can individually turn everything on, but time will tell how it juggles two kinds of heat in the winter.)
There are some restrictions on command sequences; for example, the user must log in before doing anything else. The function valid :: [Command] -> Bool checks that a command sequence is allowed. Our property then has the form:
Performs an IO action every time a property fails. Thus, if shrinking is done, this can be used to keep track of the failures along the way.
$(polyQuickCheck 'prop) means the same as quickCheck $(monomorphic 'prop). If you want to supply custom arguments to polyQuickCheck, you will have to combine quickCheckWith and monomorphic yourself.
InfiniteList xs _: guarantees that xs is an infinite list. When a counterexample is found, only prints the prefix of xs that was used by the program.
Whynter arc 14shproblems
Note: the bizarre return [] in the example above is needed on GHC 7.8 and later; without it, quickCheckAll will not be able to find any of the properties. For the curious, the return [] is a Template Haskell splice that makes GHC insert the empty list of declarations at that point in the program; GHC typechecks everything before the return [] before it starts on the rest of the module, which means that the later call to quickCheckAll can see everything that was defined before the return []. Yikes!
Tests a property, produces a test result, and prints the results and all test cases generated to stdout. This is just a convenience function that combines quickCheckResult and verbose.
It's possible that http://www.climax-air.com/ knows what they're doing, since they make (as far as I know in 2020) the only dual-hose portable with a variable-speed compressor. But I've only looked at the website, since their supply is very limited. Maybe I'll check again when the EdgeStars wear out.
I recently purchased a couple EdgeStar AP14001HS dual-hose air conditioners, and noticed that the intake hose moves quite a bit less air than the exhaust hose. It seems that the extra air is "leaking" into the condenser compartment where the bottom meets the front of the unit. See the attached photo, where I temporarily covered the gap with masking tape. Are you able to tell me whether this gap serves a necessary function? Otherwise, I would rather leave it covered, because I chose a dual-hose model to keep the inside/outside air isolated from each other.
whynterarc-14sh service manual
One caveat to this installation is that my windows are locked/taped into position, so I can't open them at night when the air is cooler outside. It would be nice to have an automatic "ventilate" mode for this purpose. However, an A/C cooled with cool air produces even colder air, so it doesn't need to spend much time running in this state.
According to this article, "infiltration air" is a major factor in the SACC measurement: https://support.edgestar.com/hc/en-us/articles/115005480926-What-is-SACC-Seasonally-Adjusted-Cooling-Capacity-
I cut the MLV to size using a tape measure, square, pencil, and scissors. Removing the bottom corners leaves room for the power cord. The wings have different lengths because the power cord is off-center.
You should prefer tabulate to label when each test case is associated with a varying number of values. Here is a (not terribly useful) example, where the test data is a list of integers and we record all values that occur in the list:
Prints out the generated test case every time the property is tested. Only variables quantified over inside the verbose are printed.
There is a default implementation for function, which you can use if your type has structural equality. Otherwise, you can normally use functionMap or functionShow.
For the improved version, I used mass loaded vinyl (MLV) wrapped in cotton blanket with velcro strips, held together using staples.
Generates an integral number. The number can be positive or negative and its maximum absolute value depends on the size parameter.
For statistical reasons, checkCoverage will not reject coverage levels that are only slightly below the required levels. If the required level is p then an actual level of tolerance * p will be accepted. The default value is 0.9.
whynterarc-14sh heat mode
Note: If the coverage check fails, QuickCheck prints out a warning, but the property does not fail. To make the property fail, use checkCoverage.
The default implementation returns the empty list, so will not try to shrink the value. If your data type has no special invariants, you can enable shrinking by defining shrink = genericShrink, but by customising the behaviour of shrink you can often get simpler counterexamples.
whynterarc-143mx
(In this photo, note that I taped over the LEDs to reduce brightness, and the chunk of foam was from an abandoned soundproofing experiment involving repositionable adhesive.)
The photodiode and IR LED dongles are attached to 22AWG MTA-100 connectors. I used a small screwdriver in lieu of the $35 Insertion Tool.
If you want to get the counterexample as a Haskell value, rather than just printing it, try the quickcheck-with-counterexamples package.
Ordinarily, a failed coverage check does not cause the property to fail. This is because the coverage requirement is not tested in a statistically sound way. If you use cover to express that a certain value must appear 20% of the time, QuickCheck will warn you if the value only appears in 19 out of 100 test cases - but since the coverage varies randomly, you may have just been unlucky, and there may not be any real problem with your test generation.
Note: for technical reasons, the test case is printed out after the property is tested. To debug a property that goes into an infinite loop, use within to add a timeout instead.
On the air conditioner side, I'm driving a generic IR LED through a 150 ohm resistor, using the sendRaw() function from IRremoteESP8266.
The QuickCheck manual goes into detail on how to write good generators. Make sure to look at it, especially if your type is recursive!
Maximum number of successful tests before succeeding. Testing stops at the first failure. If all tests are passing and you want to run more tests, increase this number.
A final gotcha: we cannot define shrink as simply shrink x = Nil:genericShrink x as this shrinks Nil to Nil, and shrinking will go into an infinite loop.
With the unit running for hours when it's 90°F outside, the hottest point under the blanket is around 94°F, so it seems to do a reasonable job of keeping itself cool.
Takes a list of elements of increasing size, and chooses among an initial segment of the list. The size of this initial segment increases with the size parameter. The input list must be non-empty.
The plastic gets pulled toward the rear while the compressor is running, and toward the front otherwise, so a piece of tape can help hold it in place.
Shrinks the argument to a property if it fails. Shrinking is done automatically for most types. This function is only needed when you want to override the default behavior.
All in all, MLV does reduce noise from the compressor, but there's still plenty of blower noise. I'm not sure to what extent the cotton wrap helps to reduce reflected sound, but at least it looks nicer than raw vinyl.
To make the unit run quieter, I started by wrapping it with a blanket, held in place using tape. The hoses prevent anything from obstructing the room air intake. Here is my "A/C burrito" prototype:
Tests a property, using test arguments, and prints the results and all test cases generated to stdout. This is just a convenience function that combines quickCheckWith and verbose.
There is no generic arbitrary implementation included because we don't know how to make a high-quality one. If you want one, consider using the testing-feat or generic-random packages.
Checks that at least the given proportion of successful test cases belong to the given class. Discarded tests (i.e. ones with a false precondition) do not affect coverage.
I plan to run these in "heat" mode during the winter, which should consume less energy than the resistive baseboard heaters, but time will tell how well that works.
Whynter arc 14shmanual
On a day that peaks around 95°F, two A/Cs running continuously can keep 850 sqft in the mid-70s. My IR thermometer says the ceiling is considerably warmer than the walls, so it would probably be useful to improve the attic insulation.
Here is the thermostat enclosure with everything mounted, including a 24VAC to 5VDC power supply at the bottom. I stuck with MicroUSB, so I can't forget to unplug the power before reprogramming it.
Checks that the values in a given table appear a certain proportion of the time. A call to coverTable table [(x1, p1), ..., (xn, pn)] asserts that of the values in table, x1 should appear at least p1 percent of the time that table appears, x2 at least p2 percent of the time that table appears, and so on.
I wanted a dual-hose air conditioner because single-hose models are thermodynamically suboptimal. Blowing hot air out the window creates negative pressure in the room, which sucks outdoor air back inside through various cracks. In theory, dual-hose models avoid this problem by cooling the condenser with outdoor air, which is then re-exhausted.
Map a shrink function to another domain. This is handy if your data type has special invariants, but is almost isomorphic to some other type.
The hoses may be insulated using a 6 inch "tubular cotton stockinette bandage". After 5 minutes of runtime with 2 layers of cotton, I measured 122°F at the hose and 102°F at the cotton surface. This isn't the greatest insulating material, but it's cheap, easy, and looks reasonable enough.
Used to generate a function of type a -> b. The first argument is a value, the second a generator. You should use variant to perturb the random generator; the goal is that different values for the first argument will lead to different calls to variant. An example will help:
Generates an integral number. The number is chosen uniformly from the entire range of the type. You may want to use arbitrarySizedBoundedIntegral instead.
Given a property, which must use label, collect, classify or cover to associate labels with test cases, find an example test case for each possible label. The example test cases are minimised using shrinking.
genericShrink is a combination of subterms, which shrinks a term to any of its subterms, and recursivelyShrink, which shrinks all subterms of a term. These may be useful if you need a bit more control over shrinking than genericShrink gives you.
Warning: any random values generated inside of the argument to ioProperty will not currently be shrunk. For best results, generate all random values before calling ioProperty, or use idempotentIOProperty if that is safe.
Test a polymorphic property, defaulting all type variables to Integer. This is just a convenience function that combines verboseCheck and monomorphic.
QuickCheck provides Arbitrary instances for most types in base, except those which incur extra dependencies. For a wider range of Arbitrary instances see the quickcheck-instances package.
For example, listOf, which uses the size parameter as an upper bound on length of lists it generates, can be defined like this:
Each 24VAC relay output drives the input pins of an SFH620AGB optocoupler through a 10K resistor (24V / 10kohm will push a few mA through the optocoupler LEDs). The optocoupler is compatible with GPIO in INPUT_PULLUP mode, using a bit of software to clean up the signal. I got the idea from this StackExchange post, though 24VAC is less scary than 120-240VAC.
Discards the test case if it does not complete within the given number of microseconds. This can be useful when testing algorithms that have pathological cases where they run extremely slowly.
Optional; used internally in order to improve shrinking. Tests a property but also quantifies over an extra value (with a custom shrink and show function). The Testable instance for functions defines propertyForAllShrinkShow in a way that improves shrinking.
For example, listOf, which uses the size parameter as an upper bound on length of lists it generates, can be defined like this:
I cut a ballpoint pen in half, painted it black, and placed it over the mode status LEDs, with an INL-3APD80 photodiode to detect brightness changes when cycling through modes. The photodiode seems to work best in photovoltaic mode, with anode connected to an ADC input, and cathode connected to GND. I do a read every 50 microseconds, and average across 1000ms (20000 samples) to get a clean signal. Adding a capacitor in parallel with the photodiode reduces the stabilization time; 10nF seems like a good balance between noise and responsiveness, reducing the read time to 100-200ms.
To start using QuickCheck, write down your property as a function returning Bool. For example, to check that reversing a list twice gives back the same list you can write:
Test all properties in the current module, using a custom quickCheck function. The same caveats as with quickCheckAll apply.
Run a generator. The size passed to the generator is always 30; if you want another size then you should explicitly use resize.
whynterarc-1230wn
It doesn't look like there has been factory level testing with the gap sealed, but the engineers who work on these units have performed their own tests and found covering this gap to not be an issue. They weren't able to give me additional information on the SACC though as it's not something we can test at this location.
Generates a random element in the given inclusive range. For integral and enumerated types, the specialised variants of choose below run much quicker.
Note: for technical reasons, the test case is printed out after the property is tested. To debug a property that goes into an infinite loop, use within to add a timeout instead.
Collects information about test case distribution into a table. The arguments to tabulate are the table's name and a list of values associated with the current test case. After testing, QuickCheck prints the frequency of all collected values. The frequencies are expressed as a percentage of the total number of values collected.
Implication for properties: The resulting property holds if the first argument is False (in which case the test case is discarded), or if the given property holds. Note that using implication carelessly can severely skew test case distribution: consider using cover to make sure that your test data is still good quality.
The standard parameters used by checkCoverage: certainty = 10^9, tolerance = 0.9. See Confidence for the meaning of the parameters.
Note: If the coverage check fails, QuickCheck prints out a warning, but the property does not fail. To make the property fail, use checkCoverage.
This unit blows indoor air mostly upwards. I opened the case to remove the built-in deflector, and put a Frost King HD9 in its place. This creates a roughly-horizontal air stream, and also seems slightly quieter without those extra fins in the way.
Thank you. I had suspected that the gap was to prevent the unit from filling with water, though it seems bigger than necessary.
Places like the Bay Area are full of apartments and condos that are gradually becoming unlivable as temperatures rise, so it would be nice if people could just throw money at a decent product.
To use QuickCheck on your own data types you will need to write Arbitrary instances for those types. See the QuickCheck manual for details about how to do that.
The basic building block for Function instances. Provides a Function instance by mapping to and from a type that already has a Function instance.
These types do things such as restricting the kind of test data that can be generated. They can be pattern-matched on in properties as a stylistic alternative to using explicit quantification.
A special error value. If a property evaluates discard, it causes QuickCheck to discard the current test case. This can be useful if you want to discard the current test case, but are somewhere you can't use ==>, such as inside a generator.
How certain checkCoverage must be before the property fails. If the coverage requirement is met, and the certainty parameter is n, then you should get a false positive at most one in n runs of QuickCheck. The default value is 10^9.
The junction box originally held a 208VAC thermostat for baseboard heat. I added a Schneider Electric 92S7A22D-24 relay, which the Nest identifies as "aux heat" via the white wire. Actually, the relay and transformer are the only components needed to use the Nest as a line-voltage thermostat.
When I connect a 10K resistor between (e.g.) Y1<->C on My Nest 3rd Generation, it has no trouble detecting that a device is present, but I've heard reports that the Nest E needs a heavier load. Adding a 1K resistor (rated for at least 1 watt) in parallel with the Y1<->C connection should be sufficient.
I don't think my setup is rigorous enough to calculate any absolute efficiency numbers, but clearly the sealed configuration does a better job of cooling, and single-hose mode is abysmal.
To block this one, I cut a 436mm x 98mm rectangle from a sheet of 1/16" (1.5mm) ABS plastic. A cereal box also works for prototyping purposes... in general, the material needs to be pretty thin, but rigid enough to keep itself flat.
If all this leaves you bewildered, you might try shrink = genericShrink to begin with, after deriving Generic for your type. However, if your data type has any special invariants, you will need to check that genericShrink can't break those invariants.
...we can add a coverage requirement as follows, which checks that LogIn, LogOut and SendMessage each occur at least 25% of the time:
whynterarc-14sh 14,000 btu
Generates an integral number from a bounded domain. The number is chosen from the entire range of the type, but small numbers are generated more often than big numbers. Inspired by demands from Phil Wadler.
24VAC relay module, with 10K resistors and optocouplers (Note: for compatibility, it may be useful to parallel each connector with a 1K 1W resistor):
Note: for technical reasons, the test case is printed out after the property is tested. To debug a property that goes into an infinite loop, use within to add a timeout instead.
If you want to use polyVerboseCheck in the same file where you defined the property, the same scoping problems pop up as in quickCheckAll: see the note there about return [].
Test all properties in the current module. The name of the property must begin with prop_. Polymorphic properties will be defaulted to Integer. Returns True if all tests succeeded, False otherwise.
The QuickCheck manual gives detailed information about using QuickCheck effectively. You can also try https://begriffs.com/posts/2017-01-14-design-use-quickcheck.html, a tutorial written by a user of QuickCheck.
I bought a couple EdgeStar AP14001HS portable air conditioners in April 2020, since I expect to spend most of the Summer at home. I'm not able to use window A/Cs here, though that would be the obvious choice for efficiency because all the hot parts remain outside.
The use of ==> may skew test case distribution. We use collect to see the length of the command sequences, and tabulate to get the frequencies of the individual commands:
You can rip open a standard furnace air filter, and cut the zig-zag part to fit the air intake's plastic frame. Note that this will restrict airflow, so it probably makes sense to run the blower on high. I increased the density a bit after taking this photo:
Finally, if you decide to open the unit (a 12-inch #2 Philips screwdriver is helpful), it's possible to tape over some internal hose/wiring gaps, but I'm not sure how important these are.
Pico was made by Gilbert Pellegrom and is maintained by The Pico Community. Released under the MIT license.
Used for random generation of functions. You should consider using Fun instead, which can show the generated functions as strings.
However, when I feel the intake and exhaust hoses on the AP14001HS, the volume of air moving through the intake is considerably smaller, so air must be leaking from the room into the condenser compartment, and also between the room and outside while the blower is off. There are two major locations where this happens:
WARNING: Users working on the internals of the Set type via e.g. Data.Set.Internal should be aware that this instance aims to give a good representation of Set a as mathematical sets but *does not* aim to provide a varied distribution over the underlying representation.
To generate random values of type Fun a b, you must have an instance Function a. If your type has a Show instance, you can use functionShow to write the instance; otherwise, use functionMap to give a bijection between your type and a type that is already an instance of Function. See the Function [a] instance for an example of the latter.
Warning: during shrinking, the I/O may not always be re-executed. Instead, the I/O may be executed once and then its result retained. If this is not acceptable, use ioProperty instead.
I couldn't find an enclosure with the right dimensions, so I cut out some ABS rectangles with a utility knife, attached using 5mm and 12mm M3 screws, with 18mm standoffs and 5mm spacers.
If you want to use polyQuickCheck in the same file where you defined the property, the same scoping problems pop up as in quickCheckAll: see the note there about return [].
You should only use genericCoarbitrary for data types where equality is structural, i.e. if you can't have two different representations of the same value. An example where it's not safe is sets implemented using binary search trees: the same set can be represented as several different trees. Here you would have to explicitly define coarbitrary s = coarbitrary (toList s).
I find it disgraceful that the portable A/C companies are so clueless about efficiency that improvements can be made using basic household supplies. This reminds me of the computer PSU industry before 80 Plus, where everything is garbage and nobody knows what to buy. Why is all the innovation focused on single-hose designs?
In the following counterexample, the list must start with "bbbbbbbbbb" but the remaining (infinite) part can contain anything: