Status note
This audit captured a pre-integration snapshot of the graph. Several routing fixes have since landed, so some nodes listed below may no longer be true orphans. Treat the page as a graph-memory document and a checklist for any remaining weakly surfaced clusters.
Actionable orphan nodes
The following nodes have zero inbound links and matter enough to deserve explicit routing.
High priority
Research Frontiers
Why it matters:
- this is the garden’s strongest cross-paper synthesis layer
- it contains several of the best bridge pages already written
- without inbound links, those bridges are effectively invisible
Challenge History
Why it matters:
- it is the richer public-record layer
- it contains Submission archetypes and Likely strategy families
- it should be discoverable from the main challenge route, not only via direct file knowledge
RWA Breadth Experiment
Why it matters:
- this is the clearest concrete evidence node attached to the recurrent-width lane
- it should be reachable from Recurrent wide architecture and Recursive width scaling
- without that routing, the architecture lane lacks visible experiment grounding
Unified Compression-Aware Architecture
Why it matters:
- it tries to unify normalization, recursion, sparse protection, and optimizer choices
- even if it stays provisional, it is important enough to either route or retire
- as an orphan, it is easy to miss and hard to evaluate
Entire orphan idea layer
All five idea pages are currently orphaned:
- Global Codebook Recursive Backbone
- Entropy-Weighted Vocabulary Rescue
- Head-to-Depth Budget Swap
- Norm-Only Phase Specialization
- Token-Adaptive Recurrent Refinement
Why this matters:
- these are not throwaway brainstorms; they are specific research bets
- each one already has natural parent pages in
frontiers/,hypotheses/, orlanes/ - the current lack of routing makes the idea layer functionally absent
Hidden clusters
Cluster 1 — frontiers/*
Status: high-value, hidden
Symptoms:
- Research Frontiers has zero inbound links
- Tokenizer-Head Co-Design Under a Hard Cap has only one inbound link
- Refinement Loops as Decompression has only one inbound link
- Compression Interfaces for Shared Depth has only internal frontier backlinks plus minimal spillover
Why this cluster matters:
- it already does the bridge work between papers, hypotheses, and next experiments
- it contains the best current synthesis of cross-lane interactions
- it is likely the highest-ROI routing fix in the whole garden
Cluster 2 — challenge-history/*
Status: coherent, but isolated from main challenge entry points
Symptoms:
- main
challenge/*pages do not route intochallenge-history/* challenge-history/*pages point back outward, so the bridge is one-way- public-record interpretation is split across two areas of the garden
Why this cluster matters:
- it contains better structure for public runs, archetypes, and strategy families
- it helps readers distinguish public evidence from challenge-implied future directions
- it sharpens challenge framing into something more research-useful
Cluster 3 — ideas/*
Status: detached proposal layer
Symptoms:
- every idea page has zero inbound links
- there is no visible
ideashub - the pages are conceptually attached to existing frontiers but graphically detached from them
Why this cluster matters:
- it is the clearest source of next-step experimentable bets
- it can convert broad frontier synthesis into actionable hypotheses
- without routing, the garden loses much of its “what should we test next?” value
Weakly connected but important nodes
Research Atlas
Current issue:
- cited from challenge-facing pages, but not surfaced from the main home or graph routes
- functions more like a detached overview than a navigational bridge
BitNet b1.58
Current issue:
- strong conceptual relevance to native low-bit regimes and normalization
- relatively low inbound visibility compared with its importance to the low-bit framing
Computational Bottlenecks of Training SLMs
Current issue:
- central to the training-economics lane
- under-routed relative to its importance for compute-aware decisions
Interpretation
The garden is not suffering from a lack of ideas. It is suffering from layer invisibility:
- the bridge layer is hidden (
frontiers/*) - the proposal layer is hidden (
ideas/*) - the public-record interpretation layer is hidden (
challenge-history/*) - the experiment evidence layer is barely present (
experiments/*)
Best immediate remedy
Before authoring new content, make sure these existing hidden clusters become reachable from: