Nobody draws it out.
Animated videos about how software systems behave and how they fail, worked out from real numbers and real documents. Newest first; each one plays on YouTube.
- Nobody Explains
- One idea from running systems in production, in a few minutes: queues, latency, retries, caches.
- Nobody Reads the Documents
- One real story told from the record: regulators' orders, court rulings, postmortems. I read the boring documents so you don't have to.
-
Nobody Explains · S1 E12
The deploy is the riskiest change you make: canaries, progressive rollout and fast rollback
Google's SRE book says roughly 70% of outages come from changes to a live system. This video puts numbers on its three answers.
· · Watch on YouTube
-
Nobody Explains · S1 E11
Capacity planning with one spreadsheet: peak to average, headroom and the N+1 rule
How many servers do you need? This video builds a five-line capacity model from real traffic: four weeks of English Wikipedia page views by the hour, from Wikimedia's public API.
· · Watch on YouTube
-
Nobody Explains · S1 E10
Load balancing is scheduling: round robin vs least outstanding vs power of two choices
Ten servers, each sent exactly the same number of requests, and halfway through one holds 26 while another holds 1.
· · Watch on YouTube
-
Nobody Explains · S1 E09
The consistency you can afford: read-your-writes, monotonic reads and when eventual is fine
You save a new display name, the page reloads, and it shows the old one: the write went to the primary and the read went to a replica that hadn't applied it yet.
· · Watch on YouTube
-
Nobody Tutorials · Tutorial 05
How to format a Word manuscript for a KDP paperback
To format a Word manuscript for a KDP paperback: set the page to your trim size (KDP says 6 x 9 in is the most common in the US), keep the top, bottom and outside margins at 0.25 in or more (0.375 in with bleed), and set the inside margin…
· · Watch on YouTube
-
Nobody Explains · S1 E08
Blast radius: cells, bulkheads and the math of shuffle sharding
If every server handles every customer, one poison request or one bad deploy can reach 100% of your users.
-
Nobody Tutorials · Tutorial 04
How to convert a PNG to SVG for Cricut
A Cricut cuts along closed outlines, and a PNG is pixels: uploaded as is, Design Space brings it in as a single layer, and a Cut image keeps only the outer silhouette. Trace it into an SVG with one layer per color and no specks first.
· · Watch on YouTube
-
Nobody Explains · S1 E07
Make "do it twice" safe: idempotency keys, at-least-once delivery and dedup windows
A payment request times out after 1 second, but the charge took 1.2 and went through. The client retries, and the customer pays 40 dollars twice. With retries at three layers of three attempts, that's up to 27 charges.
-
Nobody Tutorials · Tutorial 03
How to make a scanned form fillable without Acrobat
A scanned form isn't fillable because it's a picture of the page: it has no form fields. To make it fillable, put fields on top of the scan (a PDF interactive form, also called an AcroForm).
· · Watch on YouTube
-
Nobody Tutorials · Tutorial 02
What are the JSON files in Google Takeout and how to merge them back into photos
The JSON files in a Google Takeout hold each photo's details from Google Photos: when it was taken (photoTakenTime), where (geoData), and your caption (description). They aren't inside the photos, so other apps can't see them.
· · Watch on YouTube
-
Nobody Reads the Documents · No. 05
The servers were the cheap part: the AWS line items that grow a cloud bill
I read AWS's pricing pages and built one small, made up app's monthly bill from them, at US East (N. Virginia) list prices as I read them on 25 September 2026.
-
Nobody Tutorials · Tutorial 01
How to fix wrong dates in Google Takeout photos
Your Google Takeout photos show the export day because the real date is in the JSON file next to each photo (photoTakenTime), not in the photo.
· · Watch on YouTube
-
Nobody Explains · S1 E06
Caches lie: hit rates, cache stampedes, request coalescing and stale data
Your database only sees the cache's misses. At 10,000 reads a second, a 99% hit rate sends it 100 a second; drop to 90% and it gets 1,000, ten times the load.
-
Nobody Reads the Documents · No. 04
What happens after you tap a card: 923 pages of rules and 10 seconds to answer
You tap a card and the terminal beeps. I read the documents that run that beep so you don't have to, starting with Visa's rulebook: 923 pages, April 2026 edition.
-
Nobody Explains · S1 E05
Timeouts: the number nobody sets (deadlines, p99.9 and deadline propagation)
A dependency that hangs is worse than one that fails.
-
Nobody Reads the Documents · No. 02
Quibi raised $1.75 billion and announced its shutdown 198 days after launch
Quibi raised $1.75 billion, launched on April 6, 2020 with shows made for phones, the scripted ones told in chapters of seven to ten minutes, and announced on October 21, 2020, 198 days later, that it was shutting down.
-
Nobody Explains · S1 E04
Retries can take you down: retry amplification, jitter and retry budgets
A retry is extra load sent to a service that is already struggling.
-
Nobody Reads the Documents · No. 03
One extra second: how 23:59:60 overloaded Linux servers in 2012
Reddit, Mozilla and the Amadeus Altea airline system (used for Qantas check-in) all reported trouble when the leap second 23:59:60 UTC was inserted at the end of 30 June 2012. The IERS had announced it on 5 January 2012.
-
Nobody Explains · S1 E03
Backpressure: saying no is a feature (bounded queues and load shedding)
What happens when 1,100 requests a second arrive at a service that handles 1,000?
-
Nobody Explains · S1 E02
Latency is a distribution, not a number: p99, fan-out and the tail at scale
Average latency hides your slowest requests.
· · Watch on YouTube
-
Nobody Reads the Documents · No. 01
One server missed a deploy. 45 minutes later, Knight Capital was on its way to a $460 million loss.
On August 1, 2012, Knight Capital's order router, SMARS, had 212 small retail orders to fill.
· · Watch on YouTube
-
Nobody Explains · S1 E01
Everything is a queue: why latency explodes near 100% utilization
Every system you run is a queue: connection pools, thread pools, Kafka partitions, GPUs.
· · Watch on YouTube
