Posts

Madagascar & the 2008 Financial Crisis

2026-04-11T00:00:00+00:00

What could a children's movie about zoo animals possibly have to do with the 2008 financial crisis?

"You are what you eat." – Jean Anthelme Brillat-Savarin </blockquote>
And if we are what we consume, then LLMs could be the best personal chefs we've ever had. All you have to do, is bring your ingredients. The ingredients inside of you. While some of us have an easier time than others of bringing out these ingredients , it's the initiative to venture out of a shell, an echo chamber, your walled garden, and take your taste buds or neurons for a spin.
In September 2008, Lehman Brothers</a> filed for the largest bankruptcy</a> in American history. The global financial system</a> — built on captive assets, suppressed volatility</a>, synthetic hedges</a>, and the systematic silencing of dissenting analysts — collapsed under the weight of its own contradictions. Three years earlier, DreamWorks Animation had already made Madagascar.
I. The zoo is the pre-crisis financial system </a> </h2>
housing market</a>, circa 2005. Assets are bundled, captive, and optimized for yield</a>. Volatility has been engineered out of the system — not eliminated, merely invisible. The animals are comfortable. The visitors keep coming. The fund managers, offscreen and unnamed, collect their fees. Everything appears stable because nothing has been tested.

This is the mortgage-backed security</a>. This is the AAA-rated tranche</a>. This is the entire apparatus of pre-crisis finance: a closed system in which risk</a> has not been removed but merely caged, fed on schedule, and given a crowd-pleasing name. </blockquote>
II. Marty's escape is the subprime trigger </a> </h2>
Marty the zebra is the subprime</a> borrower: not malicious, not sophisticated, simply acting on the genuine belief that the open market</a> holds something better than the cage he was born into. His midnight escape into Grand Central is the 2004 adjustable-rate mortgage</a>. The problem is not Marty. The problem is that the entire portfolio was built on the assumption that Marty would never leave.
One default triggers contagion</a>. Alex, Gloria, and Melman — none of whom chose exposure to Marty's risk — are repriced</a>, seized, and shipped overseas. This is correlation risk</a>: the thing every 2008 model forgot to model. When one asset moves, correlated assets move with it, regardless of their individual fundamentals.
III. The ship is the shadow banking system </a> </h2>
hedge funds</a>. They arrive with no sentimental attachment to the cargo, identify the inefficiency instantly, execute a leveraged</a> takeover with ruthless precision, and redirect the entire operation toward their own ends. They do not explain themselves. They are already gone before anyone understands what happened. In 2008, we called this the repo market</a>.
IV. Madagascar is the crisis itself </a> </h2>
The island is mark-to-market</a>. It is the moment the models meet the actual world and the actual world wins. Infrastructure is absent. There is no liquidity</a>. There is no Federal Reserve</a>. There is only the landscape as it actually is, stripped of the zoo's managed fiction — and it is nothing like the prospectus described.
The animals' distress upon arrival is the precise emotional register of a pension fund</a> manager in October 2008, staring at a screen, realizing that the assets he was told were safe were never safe. They were just caged.
V. King Julien is Alan Greenspan </a> </h2>
King Julien XIII is the film's Alan Greenspan</a>: a self-appointed authority whose entire monetary policy</a> rests on personal mythology and crowd management rather than institutional legitimacy. He controls the island's resource base through cultural dominance. He keeps the population compliant through spectacle. His interest rate</a> is set not by data but by the music he wants to dance to. He has held power so long that the system has forgotten it could function without him.
Most critically, Julien's response to the Foosa threat — the island's endemic systemic risk</a> — is not to address it structurally but to import an exotic financial instrument</a> and hope it solves the problem. That instrument is Alex. This decision will be discussed shortly. It does not go well.
VI. Maurice is the analyst who was right about everything </a> </h2>
Maurice is Nouriel Roubini</a>. He is the internal risk officer</a> whose memos were never escalated. He is every analyst at Moody's</a> who knew the ratings</a> were wrong and filed the objection that disappeared into a drawer. Maurice flags every bad decision Julien makes, understands the risk clearly, and is systematically ignored — not because he lacks access, but because the incentive structure of the court depends on Julien being right.
He is not wrong once. He is structurally incapable of being heard. Proximity to power without influence over it is its own special category of financial tragedy — one that 2008 produced in abundance.
VII. The Foosa are the derivatives market </a> </h2>
tail risk</a>. The lemur economy functions only because the Foosa are, most of the time, kept at the perimeter. Nobody has a real plan for what happens if they breach it. They are the $60 trillion credit default swap</a> market: technically always present, functionally ignored, and catastrophic the moment the mechanism keeping them at bay fails.
VIII. Julien's plan is the synthetic CDO hedge — and Alex is Lehman Brothers </a> </h2>
synthetic CDO</a> hedge. It is an instrument designed to cancel out a known threat by introducing a larger, less understood one. It works, briefly, on paper. The lemurs sleep soundly. The Foosa retreat. The model is validated.
The fatal flaw is identical to 2008: the hedge itself carries counterparty risk</a> that was never priced in. Alex is not a stable deterrent. Alex is hungry. He is, as established, a dual-use instrument — simultaneously the portfolio's most valuable asset and its most dangerous liability. The steak hallucination is the margin call</a>: the moment the gap between modeled value and actual exposure becomes impossible to ignore. The hedge has become the threat. There is no fallback position.
Julien never modeled this scenario because modeling it would have meant not doing the deal. Maurice told him not to do the deal. The record will show that Maurice told him not to do the deal.
Alex is Lehman Brothers. Outsized, indispensable, and carrying a risk profile that the entire system had agreed, tacitly, never to examine too closely. When the examination became unavoidable, the consequences were not contained. Gloria and Melman — too big to fail</a> — are rescued by the plot. The U.S. government made the same calculation in 2008. For different animals. At greater expense.
The zoo was always a lie </a> </h2>
repricing</a>. There was a boat, an island, a lemur king who governed by dancing, an advisor no one listened to, a derivatives market</a> prowling the tree line, and a lion who suddenly remembered what he was.
Madagascar was released on May 27, 2005. The U.S. housing market peaked in 2006. Lehman fell in 2008. Maurice knew. Nobody asked Maurice.

apple turned Trash into a hostage situation.

2026-03-27T03:45:00+00:00

Apple turned Trash into a hostage situation
The simple version</a> </h3> A folder had once lived under an iCloud-managed path. Later, on a different local account that was not signed into iCloud, the same machine still had to deal with Apple’s special handling for that folder. Even after it was moved to Trash. Even after it was moved out of Trash into /private/tmp</code>. At that point it should have been an ordinary directory. It was not. macOS kept treating it like a magical cloud object with hidden rules. The leaf directory showed up as compressed,dataless</code>. Normal recursive removal failed. Normal inspection failed. Finder failed. Command line tools failed. Even after the object was no longer in any real iCloud sync root, the operating system still behaved as if some invisible contract had to be honored. Why? Why is this so hard?</a> </h3> Because Apple keeps building consumer systems that override plain, understandable file semantics with service semantics. You think you have a folder. Apple thinks you have a cloud-backed representation of a folder, managed by a file provider, wrapped in metadata, subject to background state the user cannot see, reason about, or reliably control. That is bad enough when you are signed into iCloud and knowingly using it. It becomes hostile when: you are not signed into iCloud</li> the files came from another account</li> the folder is already in the Trash</li> the system still refuses to let go</li> </ul> Trash is supposed to be the escape hatch. It is where complexity goes to die. If the user says "delete this," the operating system should not respond with theological debates about cloud identity and placeholder state. Apple is hostile to people who do not want iCloud</a> </h3> This is the larger problem. Apple increasingly designs macOS as if iCloud participation is the default human condition. The system keeps inventing special storage classes, hidden metadata, file-provider layers, sync roots, optimization flags, and "helpful" abstractions that are only helpful if you fully buy into Apple’s way of doing things. If you do not want that, you get friction.</li> If you want local control, you get friction.</li> If you want predictable file behavior, you get friction.</li> If you want to stop the system from quietly consuming local storage in the name of convenience, you get friction.</li> </ul> And then when something breaks, Apple gives you almost nothing: useless dialogs</li> hidden flags</li> obscure metadata</li> timeout errors</li> no honest explanation of what the system is doing</li> no supported "force convert this back into a normal folder" button</li> </ul> The fix could have been simple. Apple could provide any of the following: "This item is an orphaned iCloud placeholder from another account."</li> "Remove cloud metadata and delete permanently?"</li> "Convert to normal local directory before deleting?"</li> "This item cannot sync because no matching iCloud account is present."</li> "Force local unlink."</li> </ol> Instead, the user gets to play forensic analyst against their own trash can. Apple loves your storage</a> </h3> There is also a broader storage story here that annoys me more the longer I use a Mac. Apple acts like storage is yours right up until the moment the system decides it has other plans: local snapshots</li> caches</li> file-provider state</li> opaque "system data"</li> downloaded-on-demand placeholders</li> synced-but-not-really-local folders</li> </ul> The machine keeps consuming disk while also making basic cleanup harder. It is the worst combination: less transparency and less control. And it all points in one direction: Stay signed in. Stay synced. Stay inside iCloud. Let Apple mediate the meaning of your files. That is not convenience. That is dependency. The part that really bothers me</a> </h3> I was just trying to delete a folder. A normal folder. In Trash. The system should have had a boring, deterministic answer, something like delete the inode tree</code>. Instead macOS preserved just enough cloud-state weirdness to break Finder, break rm</code>, break rmdir</code>, and waste real time on something that should have been a no brainer. Apple loves to market polish. But polish is not rounded corners and translucent sidebars. Polish is when the machine does not trap users inside undocumented storage abstractions. What Apple should do</a> </h3> It add a real cleanup path for cloud-managed files that have fallen out of a valid account context. At minimum: detect orphaned File Provider placeholders</li> surface that state clearly in Finder</li> offer one-click local deletion</li> strip cloud metadata when the file is moved to Trash</li> stop making Trash depend on live service semantics</li> </ul> If a file is in Trash and there is no valid cloud account backing it, macOS should prefer deletion over cleverness. Always. Final thoughts</a> </h3> What made this experience so frustrating was not merely that it broke. Everything breaks sometimes. What made it maddening was that Apple designed the system so that the user could not easily understand why it broke, could not easily force a local-only outcome, and could not even trust Trash to behave like Trash. That is hostile design. And for a company that keeps selling it just works, this kind of failure says something much more honest: It just works, as long as yield to Apple's dark patterns. </blockquote>
ten things. 2026-03-21T10:00:00+00:00 Ten Things</a> </h3> Fear is a bigger obstacle than the obstacle itself.</li> What is 'true' will always endure.</li> Break the monotony.</li> Embrace the present.</li> Your success has a ripple-effect.</li> Make the decision.</li> Be unrealistic.</li> Keep getting back up.</li> Focus on your own journey.</li> Always take action.</li> </ol> python comprehensions. 2025-12-03T13:01:00+00:00 What is a list comprehension;</a> </h2> According to wikipedia</code> a list comprehension is a syntactic construct available in some programming languages for creating a list based on existing lists. </blockquote> The general format is; list = { output_expressions | variable in input_set, predicate }</code> Here are some simple examples import math limit = 1440 l1 = [x**(1/math.pi) for x in [y for y in range(limit) if y % 3 == 0 and y % 7 == 0]] </code></pre> The resulting list contains some interesting numbers fit for number theory study. Interesting Comprehensions in Python</a> </h2> Here's a simple list comprehension fruits = ["apple","banana","orange","kiwi","guava"] gr = ((1+5**0.5)/2) def price(fruit): return len(fruit)*gr prices = [price(fruit) for fruit in fruits] </code></pre>Applying different functions to elements</a> </h3> You want to have a price setting function for fruit you are selling in you grocery store import math gr = ((1+5**0.5)/2) fruits = ["apples","bananas","oranges"] operations = [lambda apples: len(apples)**(len(apples)/10), lambda bananas: len(bananas)**(len(bananas)/math.pi), lambda oranges: len(oranges)**(len(oranges)/gr)] # Set the prices prices = [op(fruit) for op, fruit in zip(operations,fruits)] </code></pre>Automation example</a> </h2> Suppose you have a list of machines identifiable by their mac address and you want to assign a machine to each user in a userlist. Ideally there should be a machine available for every user in the userlist. class Machine: def __init__(self,mac): self.mac_address = mac self.uptime = 0 class User: def __init__(self,user_name,user_age): self.name = user_name self.age = user_age machines = [Machine('135-144-255-117-098-023'), Machine('123-209-089-023-113-210'), Machine('109-209-221-112-213-231')] users = [User('Joe',21),User('Kate',25),User('Kevin',17)] </code></pre>Combining the two</a> </h3> ums = { machine.mac_address:user.name for machine, user in zip(machines,users) } print(ums) # should output a dictionary of mac_address username pairs </code></pre> [{'135-144-255-117-098-023':'Joe'} {'123-209-089-023-113-210':'Kate'}] </code></pre>Modifying the list</a> </h3> We want to capture a user so we assign a user dict value to each mac_address key ums = { machine.mac_address:{user.name,user.age} for machine, user in zip(machines,users) } print(ums) # should output a dictionary of mac_address user dictionary pairs </code></pre> [{'135-144-255-117-098-023':{'Joe',21}}, {'123-209-089-023-113-210':{'Kate',25}}, {'109-209-221-112-213-231':{'Kevin',17}}] </code></pre> And supposing we now want to filter the list of dictionaries to show only machines (represented by their mac addresses ) assigned to users above a certain age; ums = { machine.mac_address:{user.name,user.age} for machine, user in zip(machines,users) if user.age > 18 } print(ums) # should show a list of length 2 containing users `Joe(21)` and `Kate(27) </code></pre> [ {'135-144-255-117-098-023':{'Joe',21}}, {'123-209-089-023-113-210':{'Kate',25}} ] </code></pre> Happy coding</code>. the kramdown vulnerability. 2020-08-16T00:00:00+00:00 We need to update kramdown(1.17)</code> because of a critical vunerability; $ bundle audit --update</code> Name: kramdown Version: 1.17.0 Advisory: CVE-2020-14001 Criticality: Critical URL: https://github.com/advisories/GHSA-mqm2-cgpr-p4m6 Title: Unintended read access in kramdown gem Solution: upgrade to >= 2.3.0 Vulnerabilities found!</code></pre> However, upgrading kramdown</code> is not as simple as editing the Gemfile</code> and running bundle install</code> or gem update</code> simply because kramdown(1.14)</code> is a dependency of jekyll(3.8.7)</code>; jekyll (3.8.7) 47 addressable (~> 2.4) 48 colorator (~> 1.0) 49 em-websocket (~> 0.5) 50 i18n (~> 0.7) 51 jekyll-sass-converter (~> 1.0) 52 jekyll-watch (~> 2.0) 53 kramdown (~> 1.14) 54 liquid (~> 4.0) 55 mercenary (~> 0.3.3) 56 pathutil (~> 0.9) 57 rouge (>= 1.7, < 4) 58 safe_yaml (~> 1.0) </code></pre> and is thus fixed. Btw, all version of kramdown < 2.3.0</code> are vulnerable to CVE-2020-14001.</code> So to fix that, we could try running; bundle update --patch --conservative kramdown</code> But bundle</code> reports back with; Bundler attempted to update kramdown but its version stayed the same</code> So upgrading to Jekyll 3.9.0+</code> is our next option since that comes with kramdown v2.0+</code>; $ vi Gemfile 11 # Happy Jekylling! 12 gem "jekyll", "~> 4.0" 13 </code></pre> Now when we run bundle update jekyll</code> jekyll(3.8.7)</code> and it's dependencies including kramdown</code> are updated. jekyll (4.1.1) 47 addressable (~> 2.4) 48 colorator (~> 1.0) 49 em-websocket (~> 0.5) 50 i18n (~> 1.0) 51 jekyll-sass-converter (~> 2.0) 52 jekyll-watch (~> 2.0) 53 kramdown (~> 2.1) 54 kramdown-parser-gfm (~> 1.0) 55 liquid (~> 4.0) 56 mercenary (~> 0.4.0) 57 pathutil (~> 0.9) 58 rouge (~> 3.0) 59 safe_yaml (~> 1.0) 60 terminal-table (~> 1.8)</code></pre> ## CVE-2020-14001? According to the national vulnerability database (NVD</a>), CVE-2020-14001</a> is a critical vulnerability which allows unintended read access (such as template=/etc/passwd</code>) or unintended embedded Ruby code execution (such as a string that begins with template="string://<%=`). </blockquote> ## HTML & CSS in kramdown You can apply a CSS</code> class anywhere in your karmdown file using the {: .className}</code> option which makes it easier to style different parts of your karmdown.</code> karmdown also allows HTML</code> mixing; {::options parse_span_html ="true" /}</code></pre> ...the kramdown parser processes the content of span HTML tags as text containing span-level elements. </blockquote> The Problem With The {::options /} Extension</a> </h2> Setting {::options template="/etc/passwd"}</code> allows unintended read access to /etc/passwd</code> which can help an attacker guess user passwords or gain access to other crucial files in /etc</code> such as /etc/shadow</code>. To mitigate the vulnerability, a forbidden_inline_options</code> option was added in karmdown 2.3.0</code>. This allows developers to restrict options allowed with the {::options /}</code> extensions. </blockquote> fixing pylyrics. 2019-07-30T20:10:00+00:00 PyLyrics</code> is a great package for retrieving lyrics. It uses BeautifulSoup</code> to scrape the lyrics from lyrics.fandom.com</code> BeautifulSoup</code> or bs4</code> provides; Python idioms for iterating, searching, and modifying the parse tree. </blockquote> See here</a> for more information. Mulyrica</a> </h3> PyLyrics</code> is a central part of Mulyrica</code>, my lyrics search/sentiment analysis app. I was pretty disappointed when it did not work for some queries. I thought this was a parsing error but upon inspection, it turned out to be something else entirely. It works!</a> </h3> PyLyrics</code> worked until I tested it on a few songs. It returned the following error; ValueError: Song or Singer does not exist or the API does not have the lyrics</code></pre> I checked the website but the lyrics were there. So I opened up a debug session to investigate and find out where the error was being thrown. Source of the problem</a> </h4> @staticmethod def getLyrics(singer, song): #Replace spaces with _ singer = singer.replace(' ', '_') song = song.replace(' ', '_') r = requests.get('http://lyrics.wikia.com/{0}:{1}'.format(singer,song)) s = BeautifulSoup(r.text) #Get main lyrics holder lyrics = s.find("div",{'class':'lyricbox'}) if lyrics is None: raise ValueError("Song or Singer does not exist or the API does not have Lyrics") return None #Remove Scripts [s.extract() for s in lyrics('script')] #Remove Comments comments = lyrics.findAll(text=lambda text:isinstance(text, Comment)) [comment.extract() for comment in comments] #Remove unecessary tags for tag in ['div','i','b','a']: for match in lyrics.findAll(tag): match.replaceWithChildren() #Get output as a string and remove non unicode characters and replace with newlines output = str(lyrics).encode('utf-8', errors='replace')[22:-6:].decode("utf-8").replace('\n','').replace' ','\n') try: return output except: return output.encode('utf-8')</code></pre> specifically if lyrics is None: raise ValueError("Song or Singer does not exist or the API does not have Lyrics") return None</code></pre> I quickly did a print lyrics</code> and sure enough they showed up in the debugging shell running in my browser so the problem was further down. More hunting</a> </h4> artist = form.artist.data song = form.song.data lyrics = PyLyrics.getLyrics(artist, song)</code></pre> This time I got the following error; ascii code can't decode byte 0xc3 in position 1245: ordinal not in range(128) in file /venv/lib/python2/site/packages/functions.py</code></pre> That was on line 96</code> in functions.py</code> under getLyrics(singer,song)</code> #Remove unecessary tags for tag in ['div','i','b','a']: for match in lyrics.findAll(tag): match.replaceWithChildren() #Get output as a string and remove non unicode characters and replace with newlines output = str(lyrics).encode('utf-8', errors='replace')[22:-6:].decode("utf-8").replace('\n','').replace(' ','\n') try: return output except: return output.encode("utf-8")</code></pre> I thought encoding</code> the lyrics twice didn't make sense so I removed call to the encoding method; output = str(lyrics).encode('utf-8', errors='replace')[22:-6:].decode("utf-8").replace('\n','').replace(' ','\n')</code></pre> New code; output = str(lyrics)[22:-6:].decode("utf-8").replace('\n','').replace(' ','\n')</code></pre> Saved the script and restarted the flask app</code> again. I now had a situation were queries that had caused problems before were working and a few of previously error-free queries were suddenly not working. Curiously, I tried running all queries under two environments, one used python2.7</code> the other used python3.5</code>. It was then that I realized that this was a matter of different versions of python. The package was made for and tested only on python3+</code> I quickly seperated what worked for python3</code> from what workd for python2</code> and everything finally worked. #Python 3 if sys.version_info.major > 2: output = str(lyrics).encode('utf-8', errors='replace')[22:-6:].\ decode("utf-8").replace('\n','').replace(' ','\n') else: # Python 2 output = str(lyrics)[22:-6:].decode("utf-8").replace('\n','').\ replace(' ','\n') try: return [album, output] except: return [album, output.encode('utf-8')]</code></pre> Happy pythoning. welcome to my blog. 2019-03-27T15:31:19+00:00 Welcome to my blog. This site is where I write about software, systems, and the things I learn while building. It is also a place to think in public a bit: to turn rough ideas into clearer ones, to document mistakes worth remembering, and to leave behind notes that might be useful later. The current version of the site is built with Zola and the Serene theme. That matters less than the reason for rebuilding it: I wanted a simpler home for my writing, one that feels lighter, easier to maintain, and closer to how I want to work now. I expect most posts here to fall into a few broad buckets: practical software notes, reflections on engineering work, and occasional pieces on tools, bugs, or ideas that were worth slowing down for. Some posts will be technical. Some will be personal in the sense that they come from experience rather than theory. Ideally, all of them will be honest and useful. I do not want this site to be a polished archive of finished thoughts. I want it to be a living notebook. A place for clear writing, careful thinking, and the kind of lessons that usually only stick after you've had to earn them. If you are here, start anywhere.

Final thoughts

Ten Things

Interesting Comprehensions in Python
Automation example
Combining the two </a> </h3>
Modifying the list

Automation example
Combining the two </a> </h3>
Modifying the list

Automation example
Combining the two </a> </h3>
Modifying the list

Combining the two </a> </h3>
Modifying the list

Modifying the list

The Problem With The {::options /} Extension

Posts

Madagascar & the 2008 Financial Crisis

apple turned Trash into a hostage situation.

ten things.

python comprehensions.

the kramdown vulnerability.

fixing pylyrics.

welcome to my blog.