This audit is for the Parameter Golf research garden as a research system, not as a generic blog.
The central question is:
does the garden help a serious reader find the best research paths quickly, or does it mostly expose the repository’s authoring structure?
Status note
This audit was written before the latest round of routing fixes that surfaced frontiers, challenge history, ideas, and local experiment history more explicitly. Treat it as a structural baseline and proposal source rather than a guarantee that every cited weakness is still live.
Scope and method
This audit looked at:
- Quartz layout and navigation behavior from
quartz.config.tsandquartz.layout.ts - the current content topology under
quartz-kb/content - hub pages such as home, Map of content, Research lanes, Paper index, Hypothesis ledger, Research frontiers, Challenge history, and Meta layer
- link density, missing-link targets, and tag spread
This is an editorial / structural audit, not a live usability study.
Executive judgment
The garden is already much better than a generic Quartz notebook because it has a clear research spine:
But its public discovery model still leans too hard on content type and folder layout rather than research intent.
The biggest current problem is not lack of material. It is that some of the highest-value layers for serious readers — especially frontiers and challenge history — are structurally under-routed compared with older or broader hubs.
Current-state snapshot
At audit time, the garden contains 74 markdown pages across these top-level sections:
| Section | Pages | What it currently means |
|---|---|---|
papers/ | 18 | literature shelf |
hypotheses/ | 9 | active claims / bets |
notes/ | 8 | mechanism notes / synthesis fragments |
challenge-history/ | 7 | public-record interpretation |
challenge/ | 6 | challenge framing |
lanes/ | 6 | major research buckets |
meta/ | 6 | operations and harness |
frontiers/ | 5 | cross-paper synthesis |
ideas/ | 5 | seed ideas |
experiments/ | 1 | experiment record |
| root pages | 3 | home, atlas, map |
That is enough material that discovery has become an IA problem, not a content-seeding problem.
What is working already
1. The garden has a serious conceptual center
Home correctly frames the site as a research layer rather than a personal blog. The separation between main research and meta is one of the strongest structural choices in the garden.
2. The best hubs are already research-oriented, not chronological
Research lanes, Hypothesis ledger, and Paper index are all more useful than a default “recent notes” model would be.
3. The folder vocabulary is mostly sane
The current top-level sections mostly reflect real intellectual roles: challenge, lanes, hypotheses, papers, frontiers, ideas, experiments, meta.
That is a good base for a research garden.
Main structural problems
1. Navigation is still author-first more than reader-first
The current left rail uses a fully open Quartz Explorer over the whole folder tree. That makes the main navigation object a filesystem view.
For maintainers, that is fine.
For serious readers, it means the default question becomes:
- “which folder should I open?”
instead of:
- “where should I begin if I care about public record, frontier bets, tokenizer strategy, or recursive architectures?”
Why this matters
A strong research garden should route by intent before it routes by document type.
Right now the most important routes are split across multiple folders:
- public challenge understanding lives in both challenge and challenge history
- current high-upside ideas live in frontiers, but the main spine emphasizes papers and hypotheses first
- mechanism synthesis lives partly in notes, partly in frontiers, and partly in lanes
That is intellectually defensible, but weak as first-contact routing.
2. Some of the strongest sections are hidden from the main routing spine
The biggest IA miss is that two very useful layers are not treated as first-class routes from the main entrance pages:
These pages exist, but they are under-signaled relative to their value.
Evidence
- Home emphasizes challenge, lanes, hypotheses, papers, graph, and meta, but not frontiers or challenge history.
- Map of content likewise routes into challenge, lanes, hypotheses, papers, and meta, but not frontiers or challenge history.
- Link analysis shows both frontiers and challenge history are effectively low-discoverability compared with the main hubs.
Why this matters
For a serious research reader, these are exactly the kinds of pages that should answer:
- what is publicly established?
- what still looks open?
- where should I place my next hour of reading?
Those are not secondary questions.
3. There are too many “map-like” concepts without a single authoritative routing layer
The garden currently has overlapping map functions:
These are all defensible pages individually, but together they create mild routing ambiguity.
The main issue
A serious reader should not have to infer the difference between:
- a top-level entry page
- a routing map
- a frontier-synthesis layer
- a research-history layer
- an older atlas / synthesis page
Today, that difference is not always obvious from the main spine.
Specific concern
Map of content is named like a visualization route, but the actual interactive graph is the right-rail Quartz graph component. That naming collision hurts discoverability and expectation-setting.
4. Explorer usefulness is limited by being fully open, alphabetic, and folder-centric
The current layout sets the explorer title to Research Map and opens folders by default.
That helps visibility, but it also means the left rail behaves like an always-expanded directory tree rather than a curated research navigator.
What readers gain
- direct visibility into all sections
- low click cost for maintainers and repeat readers
What readers lose
- prioritization
- a sense of recommended entry order
- compression of low-value branches
- separation between “core public reading path” and “working-note substrate”
Why it matters for this garden specifically
In a research garden, the explorer is most useful when it answers one of two questions:
- what are the major conceptual regions?
- where am I inside the research program?
The current explorer mostly answers:
- what folders exist on disk?
That is not enough.
5. Tag strategy is too flat and too noisy to carry discovery well
The current tag system appears broad but shallow.
Evidence
- there are 61 distinct tags across 74 pages
- 33 tags are singletons
- many tags are document-type or generic tags such as
paper,note,hub,report,map,leaderboard,protocol
Quartz is configured to generate tag pages and the graph is configured with showTags: true.
Why this matters
A flat tag vocabulary with many singletons produces weak discovery:
- tag pages become sparse or redundant
- graph nodes for tags add clutter rather than meaning
- tags repeat folder semantics instead of adding orthogonal structure
The best discovery tags in this garden would answer questions like:
- which pages are about the artifact cap?
- which pages are about recursive sharing?
- which pages belong to public record vs local research protocol?
- which pages are active bets versus archival literature?
The current flat tags do not yet do that consistently.
6. Graph discoverability is conceptually confused
Quartz already places a graph in the right rail, including a global graph modal. But the garden also has a page named Map of content that explicitly says it is not a graph visualization.
Result
There are effectively two “graph-ish” affordances:
- a true interactive graph hidden in the right rail
- a text routing page living at the most graph-like slug
That makes the graph harder to discover and harder to trust as a research-navigation tool.
Additional problem
Because graph tags are enabled while the tag system is flat and noisy, the graph is likely to over-emphasize:
- generic tags like
paper,note,hub - singleton tags with little navigational value
- central hub pages acting as supernodes
That reduces the chance that the graph reveals the deeper structure serious readers actually want.
7. High-value hubs still contain missing targets
A research garden loses credibility fast when central index pages point at pages that do not exist.
Missing targets currently referenced from live content
papers/awqpapers/quarotpapers/universal-transformerspapers/albertfrontiers/refinement-loops-as-decompression
These appear in places like Paper index, FiPS, and Research frontiers.
Why this matters
These are not obscure leaves. They sit in core routing pages. For serious readers, dead-end routes signal that the map is ahead of the shelf.
8. Deep research paths exist, but they are not declared strongly enough
The garden already contains enough material to support excellent deep-reading paths. The problem is that those paths are still implicit.
Where serious research paths should begin
Path A: “I need to understand the challenge correctly before I read papers.”
Best starting route:
- Challenge overview
- Constraints and scoring
- Challenge history
- Public runs
- Research lanes
- Research frontiers
This should be the default route for serious newcomers.
Path B: “I want the highest-upside open bets, not a broad intro.”
Best starting route:
- Research frontiers
- relevant lane page such as quantization and outliers or recursive sharing
- Hypothesis ledger
- supporting paper cluster in Paper index
This is the right route for researchers looking for next-step leverage.
Path C: “I want the current public record and strategic openings.”
Best starting route:
This is the route for readers trying to distinguish saturated from open directions.
Path D: “I already know the challenge; show me mechanism clusters.”
Best starting route:
- Research lanes
- a lane page
- linked hypotheses
- mechanism notes
- paper cluster in papers
This is the route for advanced readers and builders.
Priority recommendations
P0 — highest-value structural fixes
1. Make frontiers a first-class hub from the home spine
If the garden wants to help readers discover serious research, frontiers should sit beside lanes, hypotheses, and papers in the public route.
2. Make challenge history part of the challenge spine
The garden currently separates challenge framing from public-record interpretation too weakly. Serious readers need both.
3. Resolve missing targets in hub pages before expanding the shelf further
Dead links inside major hubs are a higher-priority UX problem than adding more notes.
4. Clarify the role split among home, map, and atlas
One page should be the authoritative entry router. The others should have narrower identities.
P1 — strong improvements after the spine is fixed
5. Recast the explorer as a conceptual navigator, not a full open filesystem tree
The explorer should foreground major hubs and reduce working-note noise.
6. Treat tags as secondary faceting, not primary IA
Tag pages can help once the taxonomy is coherent, but they should not be the main discovery backbone yet.
7. Make the graph a reinforcement tool, not the primary entry point
The graph is most useful after a reader has entered through a lane, frontier, or challenge-history path.
P2 — garden maturity improvements
8. Distinguish more clearly between ideas, hypotheses, frontiers, and experiments
Those are four different research states. The garden has all four, which is great, but the transition logic is still mostly implicit.
9. Connect experiment records back into the conceptual graph
RWA breadth experiment is currently too isolated to teach much as navigation.
Bottom line
The Parameter Golf garden already has the content needed for a strong research UX.
What it needs now is routing discipline:
- promote frontiers and challenge history
- reduce the gap between reader intent and folder structure
- stop letting flat tags and generic graph affordances pretend to be the main IA
- make the best deep-reading paths explicit
This should remain a research garden with a public research program, not drift into generic blog navigation.