One question per lab. Play until you can predict the fix. Harness research is the study the labs cite — keep it on the path, not in a drawer.
Same tokens, four costs. Click a box. Labels stay short on purpose. The first box splits further — look vs store: DSA cuts softmax, KDA cuts occupancy, MLA only thins each stored token.
What grows when I add one token?
K and V, every layer, every KV head — then that whole cache is re-read on the next decode step. A 128k window is a larger tax, not a discount.
If decode got cheaper, why is the GPU still full?
DSA stops looking at most tokens and still stores them. KDA stops storing them in 3/4 of the layers. MLA only thins each stored token. Three labs, three bills.
Advertised window, or effective window?
Attention is a budget. More tokens thin every key; the middle of the prompt is where the budget runs out. Needle tests hide this; 2-hop does not.
What does the provider key its cache on?
Exact byte-prefix. Change one character, prefill is billed again. Reuse skips recompute; it does not shrink this request’s KV.
What is allowed to pass through a lossy summarizer?
Unpinned rules die when their turn is dropped (38% violation if dropped, 0% if they survive). Pin them in the system prompt.
What happens if you detect wander but cannot steer hard enough?
Detection without an effort cap above the drive is theater. Slide steering across the phase transition.
What fraction of traffic actually needs the expensive model?
Pay frontier prices only for that slice. The bill is the size of the hard share, not “which model is best.”
What does retrieval serve after the world changed?
The stale pattern, until you consolidate. Contradiction is not an update until v2 replaces v1.
Why can the run look healthy after the rules are gone?
Nothing measures survival. Erasure has no runtime signal. Same sentences, unpinned, opposite fate from the compaction lab.
Should every token rewrite memory?
No. Write on surprise, forget on a schedule. Train-on-every-token is the bug Titans exist to fix.
Same occupancy decision, other surfaces.
21 areas · 35 problems · ledger. The labs cite this corpus; they do not replace it. Open the atlas, then walk the path. Serving is area 09; P17 is the byte-key; 19–21 are the later territories.
What must a harness actually solve?
Dashboard, ⌘K, lifecycle map, 21 syntheses, 35 problem deep-dives, paper notes, ledger. Suggested path is on the dashboard — including serving, prospective memory, identity, and metrics.
Which buffer may you write to at inference, and when?
Survey + verdict. Play the simulator first; this page is the evidence table and the serving-tier map.
Where do the 35 problems actually fire?
A catalog tape, not a mechanism. Useful after you have felt compaction and drift.
What actually drives hypertrophy — timing tricks, or tension + daily protein?
3D map and tables. Different domain. Start at MPS if you want the one transferable timing fact.