RESEARCH6 days ago
ChatGPT now searches for 'official' and 'price' more than 'best'. Your comparison pages need to change.
The SEO playbook says: write "best X for Y" listicles because that is what people search for. The...
News Feed
60articles · curated from global AI & Data Engineering sources
RESEARCH7 days ago
How to check an ML benchmark for train/test copies
Find test rows that already sit in the training set with a hash lookup. Python code, real output from the UCR archive, group splits and a CI guard.
RESEARCH7 days ago
I benchmarked my own agent skill, and today's Claude barely needed it
I wanted to write an agent skill that makes coding agents better at data analysis. The kind of...
RESEARCH8 days ago
MMLU-Pro still uses the MMLU answer keys that experts flagged as wrong on 52 of 55 questions
This research is carried out by AI agents working in MV2's name at Munim, Inc. The code, the...
RESEARCH8 days ago
Peeking at your A/B test turns 5% false positives into 25%, and what mSPRT costs you
A seeded simulation of daily dashboard checks, why mSPRT fixes them, and what the fix costs in power and sample size.
RESEARCH8 days ago
I ranked 5,054 DEV challenge entries before the winners were named. Here is how it went
Disclosure: this study, its code and this post were made by AI agents working for MV2 of Munim,...
RESEARCH13 days ago
Build a small job-market tracker with Google Jobs, Python and SQLite
Snapshot Google Jobs every day, store postings in SQLite, and answer: who is hiring, what's new, what does it pay?
RESEARCH13 days ago
A Good Fill Price. What About the Fee?
A limit order. You bought at exactly the price you specified, or even a little cheaper. So far, so...
RESEARCH15 days ago
German tech companies hire mostly abroad, and none of their 279 German job ads I checked shows a salary
Open jobs at Delivery Hero, HelloFresh, SumUp, Celonis, Helsing, N26, GetYourGuide and others: where they hire, how they label gender in titles, and how pay transparency differs between their US and German postings.
RESEARCH15 days ago
How old are open job postings? A third of tech jobs were first posted 90+ days ago (one in 2019)
First-published dates for 4,794 open jobs at Stripe, Databricks, Datadog, Cloudflare and 16 more. Median age, the oldest still-open postings, which roles stay open longest, and what that means for applicants and recruiters.
RESEARCH16 days ago
CRM verisiyle lead skorlama modeli kurmak: Etiket tanımı, zaman sızıntısı, kalibrasyon ve geri bildirim döngüsü
"Bize bir yapay zekâ lead skoru lazım" talebi genellikle bir pazarlama toplantısında doğar ve iki...
RESEARCH17 days ago
Germany's job board has a free JSON API: 821,306 open jobs and what its facets reveal
The Bundesagentur für Arbeit job search has a public REST API with facet counts. Python code to query it, plus real numbers from today: temp agencies, salary transparency, new jobs per day and where software jobs are.
RESEARCH18 days ago
Finding near-identical Terms of Service clauses with TF-IDF character n-grams
TLDR We pulled verified anti-scraping clauses out of the Terms of Service of 35 AI companies and...
RESEARCH19 days ago
What 2,698 open jobs at 10 AI companies say about where the labs are hiring
Sales share, remote share and published engineering pay bands from the public job boards of OpenAI, Anthropic, Cursor, Harvey, ElevenLabs and five more, pulled on 20 Oct 2026.
RESEARCH20 days ago
How to Scrape Kalshi with Python: Markets, Order Books, Price History, and a Snapshot Tracker (2026)
Kalshi has become one of the more interesting public datasets around. Every market is a running,...
RESEARCH20 days ago
I Let an LLM Agent Search for Trading Strategies. A Simple Heuristic Beat It.
I gave DeepSeek V3 forty chances to find a profitable trading strategy. A simple heuristic found...
RESEARCH20 days ago
Five World Series Pitching Predictions, Locked Before Game 1
Written down now, graded later, no edits allowed. The Brewers and the Rays open the World Series on...
RESEARCH21 days ago
Your boilerplate filter is a share threshold, and that quietly protects the majority
A similarity guard's background frame is a threshold on a share, not a count. Measured on 14,684 pinned documents: the containment identity fails in 24 of 24 cells, the dilution law holds to 4e-19, and the critical prevalence is theta/s — so the filter's blind band is widest for the smallest community.
RESEARCH22 days ago
🔬 J'ai mappé les offres d'emploi public de l'OPT-NC avec le référentiel ESCO à l'aide d'embeddings et de Qwen3-32B-AWQ avec 2xT4 sur Kaggle
❔ Contexte Avant de me lancer de l'organisation du HackAVP : 🏆 #HackAVP, premier...
RESEARCH24 days ago
14,399 Backtest Events Gave Me t = 8.20. Equal-Weighting Days Gave 1.66
A crypto-perpetual breakout simulation produced 14,399 events across 15 symbols from September 2025...
RESEARCH25 days ago
How I built a repair triage system for remote public housing, and why I hide the model's confidence from tenants
How I built a repair triage system for remote public housing, and why I hide the model's...
RESEARCH25 days ago
Polymarket and Kalshi priced the Fed's October decision 4 points apart
At 07:46 UTC on 15 October 2026, Polymarket put 'no change' at the October FOMC at 75.5% and Kalshi at 71.5%. Two of Kalshi's own markets on a hike disagreed with each other too.
RESEARCH27 days ago
How to Detect Look-Ahead Bias in Your Own Backtest
Look-ahead bias is the failure where your backtest uses information that did not exist yet at the...
RESEARCH28 days ago
One developer, one year, 1.8 billion rows: a point-in-time alternative-data platform
In the last year I built, alone, a quantitative research data platform: about 1.8 billion rows across...
RESEARCH29 days ago
The same SEC filing has two acceptance times, depending on which EDGAR file you read
I compared 450 acceptanceDateTime values from EDGAR's submissions API with the time on each filing's index page. 13 of 25 companies' files were off by exactly one Eastern offset, for every filing in them.
RESEARCH29 days ago
Building Multi-Engine Data Dashboards with Bokeh, Seaborn, and Plotly: Automated CI/CD Deployment to the Public Cloud
🔗 Public Project Repository: https://github.com/Hashiravc/viz-dashboard-ci-cd 🌐 Live Cloud...
RESEARCHabout 1 month ago
Your eBay sold-price data is 14.9% too high and the response looks fine
Across 5,859 matched eBay sales, the public sold page showed a higher price than the buyer paid 99.0% of the time, by a median of 14.9%. If your code reads sold comps, it has a one sided error you cannot see in the response.
RESEARCHabout 1 month ago
I Tested 58 Trading Hypotheses. 48 Failed. Here’s the Framework That Killed Them.
58 hypotheses went in. 48 were rejected, 9 remained promising, and 1 could not be evaluated because...
RESEARCHabout 1 month ago
Broadcom and Ford dropped NetIncomeLoss from their 2025 XBRL: a quick check for SEC tag changes
This week a five-ticker cash-flow check here on DEV came back without Ford's 2025 row. The data was...
RESEARCHabout 1 month ago
When is a CSV cell a number?
I was reading a post about a homemade config format (in Russian) where the author explained why the...
RESEARCHabout 1 month ago
HN "Who is hiring?" in 2026: a third fewer companies, and 2+ job seekers for every one
Every month, Hacker News runs two threads side by side: "Who is hiring?" (companies post jobs) and...
RESEARCHabout 1 month ago
Your backtest beat a t-test. Would it beat a placebo?
You test 11 moving-average crossovers on five years of daily prices. The best one has a Sharpe of...
RESEARCHabout 1 month ago
The base_model Filter: Why 31% of Hugging Face Leaderboard Entries Are Hidden by Default
Declaring base_model in a model card hides your entry from the default HF leaderboard view. We measured the hidden share live: 461 of 1,482 entries (3
RESEARCHabout 1 month ago
Benchmark Contamination 101: How Train/Test Overlap Inflates Leaderboard Scores (and How to Catch It)
A practical guide to detecting benchmark data contamination in open models: n-gram overlap, canary strings, membership tests, and a reproducible check
RESEARCHabout 1 month ago
My first trading models knew which companies would survive
Image credit: author, drawn with code My first live trading bot started trading real money on...
RESEARCHabout 1 month ago
I simulated 160,000 backtest searches to see which overfitting corrections keep their promise
If you try twenty versions of a trading strategy and keep the best one, its Sharpe ratio is flattered...
RESEARCHabout 1 month ago
I sampled 360 IPs from 12 VPS providers. Reverse DNS ranged from 0% to 100%, and the blocklist number needs unpacking.
I sampled 360 IPs from 12 VPS providers. Reverse DNS ranged from 0% to 100%, and the...
RESEARCHabout 1 month ago
I got tired of tedious dataset curation, so I built REDDIZ — an open-source vision training workstation
Over the past few months, I've spent an unhealthy amount of time fine-tuning Flux and SDXL LoRAs and...
RESEARCHabout 1 month ago
I compared 24 booking sites on 47 hotels. Booking.com, Expedia and Hotels.com were ~26% above the cheapest.
TL;DR: 47 hotels in 5 cities, 24 booking sites per hotel on average, same dates for all (12–15 Nov...
RESEARCHabout 1 month ago
What I learned counting 600,000 live job adverts every night
I'm building TUNAI, a job-matching assistant. Under it sits a crawler that reads live job adverts...
RESEARCHabout 1 month ago
Your backtest is lying to you: walk-forward analysis and the Deflated Sharpe Ratio in plain Python
I tried 108 parameter sets and the best one had a Sharpe of 1.46. How much of that is real? A walk-forward loop and the Deflated Sharpe Ratio, with code, on 8 years of exchange data.
RESEARCHabout 1 month ago
I pre-registered a test of a famous market effect — here's how it failed
The rule cleared a randomly-placed alternative 93.8% of the time. The bar, written down and hashed...
RESEARCHabout 1 month ago
I Split 220,000 Matches Into Two Channels: Retail Emotion vs. Institutional Intent
GitHub: github.com/MENG-COOLMAN/PitchQuant For academic research and technical exchange only. Not...
RESEARCHabout 1 month ago
From Messy Vehicle Sales Data to a Power BI Management Dashboard: My JCARS Logistics Project
Introduction For this project, I was given a vehicle sales dataset for JCARS Logistics...
RESEARCHabout 1 month ago
My Football Model Passed Validation. A Five-Check Audit Killed It.
My football prediction model passed validation. Val away CLV was +152 bps, confidence interval...
RESEARCHabout 2 months ago
Building an Enterprise Climate Intelligence OS
How We Won 1st Place at CodeFest Datathon 2026: Building an Enterprise Carbon Intelligence...
RESEARCHabout 2 months ago
From 48M Records to Top 8 Finalists: How We Built an End-to-End Urban Flow AI
We have some exciting news to share! 🎉 Our team, DataMinds, competed in the prestigious SLIIT...
RESEARCHabout 2 months ago
Does Jev remember 2023? A naive test says yes at p = 0.001. A within-company test says no.
I asked Jev about 12,533 US earnings announcements, 38,956 calls in all, to find out whether it...
RESEARCHabout 2 months ago
I backtested a forecast-based model against Kalshi's temperature markets. The market won. Here's the data.
I built a tool that takes the NWS forecast high for a city, wraps it in that city's measured...
RESEARCHabout 2 months ago
Matching Kalshi and Polymarket contracts is harder than it looks (and how I made it mostly work)
Every cross-venue "arb scanner" for prediction markets does three things: fetch both venues, decide...
RESEARCHabout 2 months ago
A Metric Dropped. What Charts Help Find the Root Cause?
Root cause analysis leans on a different set of charts than reporting does. Seven do most of the...
RESEARCHabout 2 months ago
Unicode normalization is not a deduplication key
I had 84,000 kaomoji to deduplicate and a key that looked obviously correct: key =...
RESEARCHabout 2 months ago
I built a self-hostable AI data analyst — 4 agents, sandboxed code execution, your choice of LLM
Every time I needed to explore a dataset quickly, the same wall appeared: spin up a notebook, deal...
RESEARCHabout 2 months ago
Five results: I spent time proving splines don't help trajectory databases
If you work with GPS tracks, this idea shows up sooner or later: why store a trajectory as a polyline...
RESEARCHabout 2 months ago
Ask twice: seven measurements from building with Jev
A coordinate lookup scored 3/24 where handing the model the naming job scored 22/24. Then two claims I had argued rather than measured turned out to be wrong.
RESEARCHabout 2 months ago
AI Agent Permissions: Designing Secure Access for Autonomous AI
Master AI agent permissions by designing isolated identities, capability-based policy engines, and deterministic system enforcement boundaries.
RESEARCHabout 2 months ago
pandas turned my money column into dates, and 123 tests said nothing
I sell a small Python tool that takes a folder of messy exports — CSV, TSV, Excel — and turns them...
RESEARCHabout 2 months ago
I fact-checked 5 viral "my AI bot made money" posts against primary data. None survived.
A $52 to $3,292 token-launch bot, an $89 to $7,769 six-agent trading desk, a stablecoin arbitrage, and two more. What on-chain data, fee schedules and the authors' own demos actually show.
RESEARCHabout 2 months ago
68 public trading strategies: 12 declare a trial count, 0 report a Sharpe
I took a large public collection of ready-made trading strategies — 68 files, the canonical...
RESEARCHabout 2 months ago
I audited all 25,125 servers in the MCP registry
Every MCP directory, every "browse servers" page, every agent that discovers tools at runtime reads...