- decision
Light mode only, no dark theme, no toggle
- assumption: the belief that made me confident
Warm paper reads as editorial confidence; dark mode reads as dev-tool default, and a toggle is a decision pushed onto the visitor
- confidence at the time
- 70% · high
- risks I named
- Developers skimming at night may find it bright
- If tastes shift, every screenshot of the site dates it
- reality: what actually happened
not yet evaluated, this one is still open
Colophon
This site keeps its own decision log, built on the same structure Lore imposes on a team's reasoning, applied to itself. Every design decision below is recorded with the assumption it rests on, how confident I was, and the risks I named at the time. If one of these assumptions breaks, this page is where you'll see it first.
built with Claude Code · Next.js · deployed on Vercel
- decision
Present the career as a reasoning graph, not a timeline or an About page
- assumption: the belief that made me confident
Showing decisions and the assumptions they rested on proves more than narrating roles. Structure is the argument
- confidence at the time
- 45% · medium
- risks I named
- Could read as a gimmick if the records aren't genuinely load-bearing
- Graphs are harder to skim than lists
- how I'd know it worked
A founder or hiring manager mentions the site unprompted
- reality: what actually happened
not yet evaluated, this one is still open
- decision
2.5D depth instead of full 3D navigation
- assumption: the belief that made me confident
Orbit controls fight usability. Depth should be felt, not steered
- confidence at the time
- 95% · very high
- risks I named
none named
- reality: what actually happened
Kept the alive, dimensional feel at zero navigation cost; reduced-motion and mobile visitors get a timeline instead
- decision
Case studies are structured decision records, not case-study prose
- assumption: the belief that made me confident
The record format (assumption, confidence, risks, success criteria, outcome) makes claims falsifiable, which is exactly what a claims-document portfolio can't do
- confidence at the time
- 70% · high
- risks I named
- Structure could feel cold without editorial prose around it
- reality: what actually happened
not yet evaluated, this one is still open
- decision
Build the site with Claude Code, and say so
- assumption: the belief that made me confident
For AI-builder audiences, how you build is itself a credential. Hiding the tooling would be hiding evidence
- confidence at the time
- 95% · very high
- risks I named
- Some readers still discount AI-assisted work
- reality: what actually happened
not yet evaluated, this one is still open
- decision
Serif prose with monospace record labels
- assumption: the belief that made me confident
The tension between editorial warmth and systematic structure is the brand: journalism roots on the surface, reasoning layer underneath
- confidence at the time
- 70% · high
- risks I named
none named
- reality: what actually happened
not yet evaluated, this one is still open
- decision
All content is schema-validated data; the UI is a pure renderer
- assumption: the belief that made me confident
Adding a project must never require touching UI code. A portfolio that's expensive to update is a portfolio that rots
- confidence at the time
- 95% · very high
- risks I named
none named
- how I'd know it worked
A new project ships end-to-end by editing only the content directory
- reality: what actually happened
not yet evaluated, this one is still open
- decision
Make the graph the hero: 60% graph, 40% temporal panel that narrates whatever you click
- assumption: the belief that made me confident
If the graph is genuinely the site's argument, hiding it below a text hero undercuts the claim; visitors explore before they read
- confidence at the time
- 45% · medium
- risks I named
- A canvas as the first paint is riskier than text for slow connections
- Visitors who don't click see less narrative
- how I'd know it worked
Node clicks per session beat scroll-past rate; the fallback stack keeps mobile bounce unchanged
- reality: what actually happened
not yet evaluated, this one is still open
- decision
The agent view is dark, the one exception to light-mode-only
- assumption: the belief that made me confident
The light-only rule exists for human readers; a machine-readable console signals its different audience best by breaking the palette deliberately
- confidence at the time
- 70% · high
- risks I named
- An exception weakens the original rule's clarity
- Two palettes to keep coherent instead of one
- reality: what actually happened
not yet evaluated, this one is still open
- decision
Removed search entirely. Filters and the map replace it
- assumption: the belief that made me confident
On a graph of ~20 nodes, search adds a text-input detour to things the eye can already find; grouping chips and hover connections do the same job with less interface
- confidence at the time
- 70% · high
- risks I named
- Visitors trained to expect a search box may briefly look for one
- reality: what actually happened
not yet evaluated, this one is still open
- decision
No About page
- assumption: the belief that made me confident
The reasoning graph answers 'who is this person' better than a bio paragraph could, and the absence forces the graph to carry that weight
- confidence at the time
- 45% · medium
- risks I named
- Visitors who expect a conventional About may look for it
- reality: what actually happened
not yet evaluated, this one is still open