quantum of sollazzo

quantum of sollazzo

Archives
Sponsor
Subscribe
November 10, 2026

682: quantum of sollazzo

Quantum of Sollazzo

Forwarded this email? Subscribe here  ·  View it online

No. 682  · 

Quantum of Sollazzo
 

Things that caught my attention

Not the National Data Library, but a National Data Library launched by the team at Open Innovations, presumably as a provocation for debate (which I, among others, always welcome), lists a series of datasets that they believe should be available for public use. In fact, they went a step further and started to collect an archive of the Bus Open Data Service (BODS) live data, and reveal it's A LOT of data, to the point they've had to stop collecting SIRI-VM data (a specific format for bus live data).

Now, this is admirable in many ways, as no official archive exist. I was part of the conversations around this when BODS was launched (I was Head of Data at the Department for Transport between 2018 and 2002 and, while not responsible for BODS itself, I had quite a bit of input on its early running). I'm obviously in favour of archiving data. The engagement with the sector revealed a lot of people were interested in the archiving, although opinion varied as to this should be a Government-run part of the service and if, maybe, someone impartial should be responsible for it. Nothing happened at the time, in part because it wasn't funded but also because there were more pressing questions about how viable operationally BODS needed to be.

I only speak as an external bystander now, but having worked in other departments since, the proportionality question is a strong one here, which can be articulated in two ways:

  1. should a service like BODS, intended to provide live data for the benefit of developer and users of transport, also cater to the uses of analysts and transport planners? And should this be free or not, given how big the archive will become over time?
  2. how do we capture and measure the benefit of an hypothetical data archiving facility provided directly by BODS?

I'd start this debate from the usual question: "What's the user need?". Features should always be prioritised according to user needs. And user needs, themselves, come with a cost/benefit analysis. Open Innovation argues that the cost of archiving is small, quoting £1,500/year vs £3b in public funding for buses. But the same could apply to many other features on the backlog and not. There's a better argument to "let's do it because it's cheap". It is "let's do it because we have evaluated the costs against the benefits and the ratio is very positive".

What do you think?

·

It was interesting to read about the Productivity Institute's The AI Ambition Gap report. One finding that stands out is that AI assistants roughly tripled programmers' code output, but released software rose only about 30% (page 12). The bottleneck? Obviously checking, testing, and actual releasing.

As I wrote last week, we can and should be measuring the benefit of AI-assisted software development. AI speeds up coding, there's no denying it. But it can't speed the parts of the process that are human-driven, and being those part for human-consumption, it's unlikely it ever will.

It's also worth noting that individual adoption of AI tools is much, much easier than enterprise-level adoption. Redesigning processes and implementing them is tough, AI or not.

·

I soft-launched Yesterpedia, a fully automated online guessing game, based on Wikimedia API. It asks you to guess the most read Wikipedia pages on the previous day, and if you score high you get a bonus location-based round. One quiz per day.

Enjoy, share your results on social media, and send me your feedback 😃

Yesterpedia

'till next week,
Giuseppe @puntofisso.bsky.social

Sponsored by Jane Street

Jane Street depends on all sorts of messy, real-world data to understand financial markets and the global economy: think world news, decades of weather patterns, deidentified credit card spending, or packet captures of stock exchange market data feeds.

We're hiring Data Engineers to turn datasets like these into reliable inputs for trading. Working closely with our researchers, you'll evaluate unfamiliar datasets, build robust ELT pipelines, develop deep domain expertise, and decide what's worth exploring next. 

The job requires a mix of engineering, data analysis, and product sense. If you love the detective work of investigating a weird dataset and figuring out what it actually means, we want to hear from you. No financial background is necessary.

Want to hear what the job is really like? Eric Mannes, who leads our alternative data team, talked with Ron Minsky about it on Signals & Threads.

We have openings in New York, London, and Hong Kong.

Jane Street

Topical

How to know a red or blue wave when you see one

USAFacts looks at midterms performance, noting that the 2010 midterms produced the largest wave in 50 years, with Republicans gaining 64 House seats and a 128-seat margin swing. The 2018 midterms saw the biggest blue wave, with Democrats winning 41 additional House seats despite losing one Senate seat.

How to know a red or blue wave when you see one

datakabas Instagram post on midterm election history

Brilliant dataviz of political swing, noting that 1998 and 2002 are the only modern midterms where the president's party gained House seats.

datakabas Instagram post on midterm election history

Trump vanishes from GOP campaign ads

Axios looks at what Republican and Democratic political ads mentioned. Illustrated by the ever incredible Erin Davis.

Trump vanishes from GOP campaign ads

Trains are still safe

After Spain's disaster, The European Correspondent asks if trains can still be considered safe, and it uses a brilliant chart to illustrate it.

Trains are still safe | The European Correspondent

Tools & Tutorials

Can LLMs Supercharge Geolocations? Our Tests Show Big Improvements But Humans Remain Key

Bellingcat tested LLMs like Claude, Gemini, GPT and Grok on 52 unpublished photos, finding Anthropic's Opus/Fable and Google's Gemini 4 Argon performed best. Models can now narrow down general locations quickly but remain far from perfect at pinpointing exact spots.
"AI models remain a useful assistant rather than a tool we can blindly delegate geolocations to".

Can LLMs Supercharge Geolocations? Our Tests Show Big Improvements But Humans Remain Key

Twelve years of distributed SQLite

  • Philip O'Toole created rqlite in 2014, combining Raft consensus with SQLite as the first open-source system of its kind.
  • The "distributed SQLite" space evolved into three design corners: consensus (rqlite, dqlite), streaming (Litestream), and CRDT-based merging (cr-sqlite).
  • Open-source projects pioneered the idea for eight years before venture-funded platforms like Turso and Cloudflare D1 turned it into a commercial product.

gralhix #004

"This is a photo of a resort located on an island.
a) What is the name of the resort?
b) What are the coordinates of the island?
c) In which cardinal direction was the camera facing when the photo was taken?
In my opinion, solving this challenge with google lens is wasting a fun opportunity, so decided to solve it with math and programming.
"
(via Geomob)

gralhix #004

nodal-data/nodal-context: Nodal Context Repo

Nodal is an open-source agent skill set that builds governed analytics context, letting analysts confirm metric definitions before agents answer data questions.

StackExchange Data Explorer

Exactly what it says on the tin.

Just a moment...

Spreading Liquid with three.js

This article describes implementing a liquid foundation spreading effect using a fluid simulation combined with a height map rendered via fragment shaders. "*I developed this liquid-spreading effect for a Japanese cosmetics brand's liquid foundation promotional project released in 2021. *".

Spreading Liquid with three.js

GetCassis/analytics-agent-articles: A curated list of articles from data teams building analytics agents.

A GitHub repository curating articles from data teams like Anthropic, OpenAI, Ramp, Uber, and LinkedIn about building internal analytics agents.

How to make interactive charts you can email

Nicola Rennie explains how to build interactive R charts using ggiraph, tooltips, and dropdown menus linked via JavaScript, and how to embed resources in a Quarto HTML document with embed-resources: true so charts work standalone.

A new Go API for JSON

The Go programming language team: "Last year, we wrote about a new experimental API for JSON designed to bring long-awaited improvements and fixes.
After additional feedback and refinement, the new encoding/json/v2 and encoding/json/jsontext packages are stable and generally available in Go 1.27!
"

Mirrored bar charts

I loved the mirrored bar chart from The European Correspondent (see "Dataviz" section), so with Claude's help I created a self-contained html page that allows you to create your own. I've put the source code in my new Quantum's Code Samples Github repository.

Mirrored bar charts

Data Thinking

Exploratory Data Analysis Can't Be Automated Away

Randy Au suggests that LLMs can't be good at curiosity-driven Exploratory Data Analysis: "Even in my attempt to actively force the bots to do my job, the end results were just so frustratingly bad that I'm confident in saying that, as of 2026, we are very much not there yet, if we ever are even going to get "there"."

What Happened to Data Products?

"Data products did not disappear with data-mesh hype; they are becoming the governed context and accountable offers that human and AI consumers need."
AI agents need explicit context ("purpose and semantics, quality and policy, plus ownership and provenance"), making portable, standardised data product definitions essential.

What Happened to Data Products?

Dataviz, Data Analysis, & Interactive

One Hundred Famous Views of Edo

Steven Feldman's latest Claude mapping attempt aims to build an interactive Hiroshige-style map of all 119 woodblock prints from Wikipedia's "One Hundred Famous Views of Edo" table.

One Hundred Famous Views of Edo - KnowWhere

Walkably: how walkable is any address?

Walkably scores every street in 31 European countries from 0 to 100 based on walking access to eight daily needs. Beautiful, interactive map, too.
(via Durand D'souza)

Walkably

Cartographers of North Korea :: 북한의 지도 제작자

"An analysis of the OpenStreetMap data in North Korea: Who are the contributors? How and Why did they contribute to North Korea"
(via Geomob)

Cartographers of North Korea :: 북한의 지도 제작자

MuGlobe — Explore World Music on an Interactive Globe

MuGlobe is a free interactive world music atlas combining NASA satellite imagery with Natural Earth geographic data, and linking to local YouTube recordings.

MuGlobe — Explore World Music on an Interactive Globe

Body

Wow. Neal has published an interactive 3D model of his body, external and internal, with personal stats like heart rate, steps, sleep, and calories, all pulled from his fitness watch and blood test.

Body (neal.fun)

Worst Halloween candies, according to the data

Nathan Yau (Flowingdata): "Some years ago, FiveThirtyEight placed candies head-to-head and calculated the win percentage of each. The site is closed, but some of the data lives on. Below are the results for chocolate and non-chocolate candies."

Worst Halloween candies, according to the data

Don't Lie to Us, Finland: The Happiness Report Has a Problem

Interesting approach (maybe with some cultural bias involved...) on using video recordings to detect a country's level of happiness by analysing smiles.

Don't Lie to Us, Finland: The Happiness Report Has a Problem

[OC] City blocks colored by the number of their corners

Barcelona's Eixample blocks are mostly octagons (32%), Midtown Manhattan is "77% plain four-cornered blocks", etc.
(via Durand D'souza)

[OC] City blocks colored by the number of their corners

I discovered a sound that leaks your exact location (and it's everywhere)

"This video examines the amount of location information carried by inaudible background sounds, using the Doppler effect to locate a reference recording within ~100m, and only needing (in some cases) less 1-min of audio. This video raises awareness of these capabilities and closes by discussing preventative measures one could take."
Basically, using a combo of publicly available aircraft data and Doppler curve spectrogram analysis, the authors show that you can pinpoint a video location to less than 100m distance.
In simple terms: showing an audio file on a spectrogram you can "see" inaudible frequencies left by aircrafts flying over the location of the recording, and cross-referencing these with ADS-B data about aircraft you can find their GPS coordinates. Fascinating, and clever.
(via Durand D'souza)

https://assets.buttondown.email/images/b0c7f409-5b1e-4342-b6cf-eb8bfb9b3d74.png

#30daymapchallenge #python #dataviz #maps

Lisa Hornung delivers on day 1 of this year's #30DayMapChallenge, by mapping German supermarkets in London, showing Lidl has almost double Aldi's shop count.

#30daymapchallenge #python #dataviz #maps | Lisa Hornung

A Family Tree of Software Job Titles

104 software job titles maped from the 1940s "Computer" to 2025's "AI Agent / Context Engineer" using Graphviz. Top findings: titles often split then rejoin (e.g. webmaster divided into web developer, sysadmin and editor before front-end/back-end reunited as "full-stack"). Many "new" roles recycle old ones, like today's "forward deployed engineer" basically being IBM's 1960s pre-sales "systems engineer".

A Family Tree of Software Job Titles – zwischenzugs.com

Economic and fiscal outlook – October 2026

Rising energy prices and interest rates are pushing CPI inflation up. But I link to this predominantly for the chart below, a dataviz style to capture multiple OBR forecasts over time versus the real outturn that reader Peter Wood refers to as "Hedgehog Plot".
(via Peter Wood)

Economic and fiscal outlook – October 2026

America on Data Centers

American Sentiment has been researching opposition to data centres in the US by running an analysis of 10,202 public comments at 417 local governments, showing that 73% of meetings had zero data center supporters.
We'll keep an eye on these folks at American Sentiment, because their dataviz approach is outstandingly good.

America on Data Centers · American Sentiment

The Full Moon Myth Takes the Field

Marcos Huerta: "Google ran a “Gemini with Google Cloud” commercial during the World Series that made 3 dubious claims about stats changing during the full moon. All are examples of small number statistics with large error bars, and analyzing more than one year of data (I used 1900-2025) shows any effect of the full moon to be spurious."
I had no idea about this full moon thing...

The Full Moon Myth Takes the Field – Marcos Huerta

Rioting for a reason

The European Corresponded suggests that only a minority "of over 10,000 European headlines about French protests mentioned protestors' demands". What I find very interesting is the brilliant choice of chart style: a mirrored bar chart, with the same values top and bottom, and with two different variables highlighted on either side.
I liked the chart very much, so you find a self-contained html file to generate your own in the Tools section.

Rioting for a reason

AI

OpenAI and Anthropic cut their LLM prices by up to 80% within a few months. Why?

OpenAI and Anthropic have cut token prices by 80% (in some cases) since June 2026, driven by competition from commoditised models and cheap open-weight alternatives like DeepSeek and Kimi. With Anthropic approaching IPO soon (and OpenAI rumoured to be heading there, too), the article notes that the cuts are puzzling given reported billions in losses.

OpenAI and Anthropic cut their LLM prices by up to 80% within a few months. Why?

Can we trust LLMs to judge whether a research idea is novel? | Ai2

"Today we’re releasing Old Ideas, Novel Problems, a new study of how reliably large language models can evaluate the novelty of scientific research ideas. Alongside the paper, we’re publishing a benchmark, data, and code for testing novelty judges, plus an automated pipeline researchers can use to build new evaluation sets."

Can we trust LLMs to judge whether a research idea is novel?  | Ai2

RAG for decision models is here!

Parallel has launched a Decisions API combining live web search with bounded decision-making in a single call. This follows the hype on decision models like Jev, Laya, and Rizzo-flow becoming popular, but adding live capabilities to them.

RAG for decision models is here!

Could a modern chess engine on a Pentium have defeated Kasparov in 1996?

"But was Deep Blue necessary to beat Kasparov? Or could we have gotten away with an off-the-shelf 1996 desktop running a smarter chess engine?"
Basically, a 1996 Pentium Pro running (a modernised port of) chess model engine Stockfish can reach an estimated Elo of 3300, far above Kasparov's near-2800 rating. The gain in performance comes from better programming rather than faster hardware.

Could a modern chess engine on a Pentium have defeated Kasparov in 1996?

How the AI boom might really end: 5 scenarios

Martin Alderson outlines five ways the AI hype might collaps, split between AI disappointing outright and AI succeeding but value flowing away from capital holders.

Did you like this issue?

Buy Me A Coffee

 

You're receiving this because you subscribed to Quantum of Sollazzo, a weekly newsletter covering all things data, written by Giuseppe Sollazzo (@puntofisso). Have a product or service data people should know about? Sponsor an issue.

The Quantum of Sollazzo grove now has 50 trees at Trees for Life, helping offset this newsletter's carbon footprint.

Proofread by Andy Redwood at Proof Red, who also makes motion graphics about climate change.

ProofRed Logo

© 2012–2026 Giuseppe Sollazzo · London · UK  ·  Unsubscribe

Don't miss what's next. Subscribe to quantum of sollazzo:
← Newer 683: quantum of sollazzo Older → 681: quantum of sollazzo
Powered by Buttondown, the easiest way to start and grow your newsletter.