from wu

notes

Reading the data

Good data display is not decoration; it is an argument you can check with your own eyes.

May 24, 2026 · 4 min

This is the science-data half of the kit. Same brand, same one accent, no rainbow: comparison tables, distributions, a single-hue heatmap, small multiples, error bars, and labeled figures.

A comparison table

The TM-style pattern: column groups, row groups, the best value in each row tinted (our oxblood wash, not amber), a featured column, and footnotes.

InstantThinking
TML-smallGemini-flashGPT-realtimeQwen-omni
Streaming
FD-bench V1latency (s) 0.40.571.632.14
FD-bench V1.5average 77.854.347.839
Turn-based
Audio MultiChallengeAPR % 43.426.848.5-
IFEvalaccuracy % 89.785.895.283.4
A comparison table in the From Wu style.
  1. Illustrative sample data, not a real benchmark.

A distribution

The real BTC monthly returns, as a histogram and a box plot:

-33% 49.9%
Distribution of BTC monthly returns. Most months are unremarkable; the tails are where the story is.
-33% 1.5% 49.9%
The same returns as a box plot: median, the middle 50 percent, and how far the worst and best months reach.

A single-hue heatmap

Correlation reads fine without a rainbow; intensity is just the accent mixed into the ground:

Illustrative asset correlations.

Small multiples

The same series, split so the calm years and the violent ones are obvious at a glance:

2021
2022
2023
2024
2025
2026
BTC monthly returns by year, same scale. The calm years and the violent ones are obvious at a glance.

Risk against return, with error bars

Return vs. riskannualized, 2021 to 2026
5 10 15 20 10 20 30 40 50 60 BTC: BTC, annual volatility % 56, annual return % 13.3 BTC S&P 500: S&P 500, annual volatility % 15, annual return % 10 S&P 500 Gold: Gold, annual volatility % 15, annual return % 6 Gold Bonds: Bonds, annual volatility % 5, annual return % 3 Bonds annual return % annual volatility %
BTC S&P 500 Gold Bonds
BTC is real-computed; the rest illustrative. Shape encodes the series (no rainbow); error bars show a confidence band.

Bulleted paragraphs

When each point needs a sentence, not a phrase, the lead-in bolds and the rest explains. Square markers, hanging indent, generous spacing:

  • Reversibility first. The cheapest information is finding out you were wrong while it is still easy to change your mind. A weekend test beats a year of planning, almost every time.
  • Speed of feedback. A decision you cannot evaluate for a decade is mostly a bet on your own story. Prefer the ones reality grades quickly, and let it grade them.
  • Survivability over optimality. Size the choice so that being wrong is a lesson, not a wound. Then you can afford to be wrong often, which is the only way to ever be right a lot.

A labeled figure

A wide atmospheric landscape 12
  1. Pins sit on the image; the legend explains them in order.
  2. Numbered, accent-filled, with a ring so they read on any photo.
An annotated figure.Unsplash

A framed diagram

CaptureDraftPublish
The lifecycle, as a framed inline SVG.

What I know now is that the brand survives data just fine, as long as the data obeys the same rules as the prose: one accent, sans labels, tabular figures, and no decoration that is not also information.

Wu · Da Nang · May 2026

Scaffold sample: the science-data kit, exercised. Comparison table, distribution, heatmap, small multiples, error bars, annotated figure, framed diagram.

all letters