Could someone advise whether I am misinterpreting the instructions, or whether this is an incorrect test:

# printed args (with multiple lines of the data arg removed to make it more concise)
data = [
    ...,
    (92, 3231),
    (2332, 92),  # used in P  # 2332 is a friend of the prince
    (92, 7178),
    (8138, 8473),
    (92, 2690),
    (9248, 2332),
    (92, 9883),
    (92, 5203),
    (698, 4683),
    (698, 4515),
    (81, 5122),
    (81, 1537),
    (5290, 1858),
    (9782, 5290),  # used in P
    (5290, 698),  # used in P
    (698, 3582),
    (698, 1922),
    (81, 92),  # used in P
    (81, 6645),
    (9782, 2332),  # used in P
    (8138, 4800),
    (81, 6611),
    ...,
]
prince_id = 92
thief_id = 698

The test result says 81 (among other similar examples) should be included in the output.

The rows with # used in P noted appear to be used in a path ([698, 5290, 9782, 2332, 92, 81]) which contains a friend of the prince (2332). This leads me to believe the ID of 81 should not be included as per the formula in this section in the instructions:

To explain this mathematically, denote $F$ as the set of friends of the prince.
Let $f \in F$ be the friend of the prince and $t$ the thief of the treasure. Now,
denote $P^t_f = { p_1, \dots, p_n}$ as the set of all the people in a
friendships path between $t$ and $f$ excluding them. $f$ is a suspect if and
only if there is a path between $t$ and $f$:

$t → p_1 → p_2 → \dots → p_n → f$ such that for all $p \in P^t_f$ , $p \notin F$

Is it simply that the end of the formula is not accurate ($ p \notin F $), or am I mistaken?

The fixed tests are not fixed, they are relying on the ref solution. Meaning if the ref solution is wrong, there is no guarantee on the specs. At all.

Also, the fixed tests should hold all the needed data for each test in one place: input data, princ and thief ids, and also the expected output. The sample tests need to also be updated with this.

Who the hell reviewed the random tests and didn't notice THIS:

def create_subgraph(id_, amount_friends, not_possible = set()):
                                       # ^^^^^^^^^^^^^^^^^^^^

Also complete lack of specification regarding expected time complexity etc.

some other solutions are only passing the tests from time to time (the current top solution, for example) => something is wrong somewhere.

Note: according to the description ("algorithmic side"), a friendship path thief -> prince -> firend of prince make all of them suspect, while this doesnt look right because the thief himself is a friend (not that problematic), but this also means there is another friend in the friendship path, meaning none can be suspected...? eidt: mmh, the thief is excluded, so I guess that part is ok, but this situation still looks pretty weird. A clarification and a specific fixed test for that is needed, at the very least.

Hi,

Something isn't specified "correctly", apparently, but more importantly, the efficiency tests are using some outputs that are built entirely differently of all the other tests (edit: that, or the problematic case comes up more frequently when the number of nodes is huge): I currenlty have a solution (fast enough) that is passing all the tests, but returns wrong results on some efficiency tests only. Problem being: how am I supposed to find the hidden spec/stuff I misunderstood on inputs with more than 4000 nodes...?

Sad that this is still lingering in beta. This is an awesome kata! You have to be willing to do the research into the solution. If you can't figure it out on your own -- I couldn't! -- go looking for the "Futurama Theorem." I found a PDF from a teacher who demonstrated the solution by using their students in a live demo. Then I just turned the teacher's live demo into code. I won't say any more, or I'll be flamed for saying too much.

Some dissatisfied with the task, because in the tests there is a performance test for 7 kyu kata, with a very large number of items. Unlike the actual implementation of the algorithm, which is described in detail, you need to do the math. It takes away the ability to implement The algorithm.

And now I've solved it, but only after researching the episode and the theorem. Great kata!

Made some basic edits to spelling (e.g., auxiliar --> auxiliary), punctuation and the like, while trying not to be too pedantic. Looking forward to trying this!

Tests should fail without raising exceptions when the result is not iterable.

Yes but it is because the test suite is more robust now

Random tests still take 6 seconds to generate.

See the bodies "Professor" and "Leela" only swith once (the last element). The restriction is about the bodies!, two pair of bodies are not allow to use the machine. The thing that says "map_" is just the final state of which body has which mind, so Leela is in Professor's body, Bender is in Amy's body, Amy is in Leela's body and Professor is in Bender's body.