Same Rate, Different Weather / design diagrams

Decision flows, supporting boards and model evidence. All shapes, labels and connected arrows are editable in the native Excalidraw files. Open a downloaded file in Excalidraw to revise it. SVG and PNG exports are ready to share.

From a matching rate to different evidence

Each layer answers a different question about the quoted rate.

Editable ExcalidrawSVGPNG
From a matching rate to different evidence
Read the diagram as text
  1. A quoted average begins the comparison
  2. Raw observations expose the range and sample count
  3. Box summaries locate central spread
  4. Density exposes shape under a chosen bandwidth
  5. Chronology tests whether the pattern persists

An annualized rate is not a cash return

The current instrument compares observations of a rate.

Editable ExcalidrawSVGPNG
An annualized rate is not a cash return
Read the diagram as text
  1. Daily observations are annualized APY quotes
  2. A 30-day window averages those quotes
  3. Overlapping windows reuse observations
  4. Cash returns require a compounding and fee model
  5. Insufficient history remains an explicit result

Read a rate through four evidence layers

Schematic marks illustrate the encoding, not additional rate observations

Editable ExcalidrawSVGPNG
Read a rate through four evidence layers
Read the diagram as text
  1. Raw observations
  2. Count and range stay visible.
  3. Center + spread
  4. A box summary condenses the observations.
  5. Density + bandwidth
  6. Smoothing changes the visible shape.
  7. Chronology + persistence
  8. Order reveals whether a pattern persists.
  9. 30-day mean = arithmetic average of annualized APY quotes.
    Cash returns require compounding, fees and flows.

Which evidence layer changes the first impression?

Five pools share the displayed 12.0% rate; different evidence exposes each one

Editable ExcalidrawSVGPNG
Which evidence layer changes the first impression?
Read the diagram as text
  1. Layer
  2. Pool / evidence
  3. Reading implication
  4. Box
  5. A / median 5.5%
  6. Campaign average masks the typical day
  7. Box
  8. D / lower whisker −26.0%
  9. Wide downside variation
  10. Cloud
  11. E / humps near 7.3% and 16.6%
  12. A mixed distribution needs chronology
  13. Rain
  14. C / 21 days of history
  15. Too little history for 30-day windows
  16. 30-day quote average
  17. B / range 11.4%–12.4%
  18. Steady control within this history
  19. Window averages summarize annualized APY quotes; they are not realized cash returns.

Commit → inspect → choose again

The interaction exposes what changed the research choice

Editable ExcalidrawSVGPNG
Commit → inspect → choose again
Read the diagram as text
  1. Rate / five 12.0% lines
  2. First pick / A–E or same
  3. Box / inspect A and D
  4. Cloud / inspect E
  5. Compare first and final
    pick
  6. Choose again
  7. 30-day quote averages
  8. Rain / inspect C
  9. Drawing controls reopen the 2025 chart choices for box, cloud and rain.
  10. The comparison records a research choice, not an observed investment return.

Data → instrument → captures → checks

Build provenance retains the generator, template and real browser captures

Editable ExcalidrawSVGPNG
Data → instrument → captures → checks
Read the diagram as text
  1. Stage
  2. Input / output
  3. Check
  4. Generate
  5. gen_data.py / seed 20251001
  6. values, summaries and catch rules
  7. Build
  8. data.json + instrument template
  9. Generated interactive instrument
  10. Capture
  11. capture.js + install_img.py
  12. Real screenshots of browser states
  13. Showcase
  14. Same Rate, Different Weather
  15. Figures retain source provenance
  16. Verify
  17. Instrument + page + data
  18. Behavior and numbers reconcile
  19. The current model scope is annualized quote observations and arithmetic windows.