Alexander Young

Product Designer, web designer, Design System, AI

[Alexander Young @Alex0nder]

Updated card: compression = API input tokens (80× Oiloop), not core-char ratio (112×).
Same results, clearer metric labels after internal audit.

https://t.co/waS2U5nz90

[Alexander Young @Alex0nder]

@dexhorthy @swyx @RLanceMartin @Cursor npx context-os init --profile saas --cursor-rule

Paper + raw runs in repo. Exploratory eval — limitations in SYNTHESIS.md.

[Alexander Young @Alex0nder]

A vs B vs C on 149 decision questions, 4 codebases.

Routed domain cores (B) beat full repo (A). Up to 38× fewer tokens.

Oiloop: B 2.75 vs A 0.75, 0% hallucination.

https://t.co/togwA3iomD

[Alexander Young @Alex0nder]

Open replication — A/B/C vs full-repo + graph. Exploratory, LLM-as-judge.

Feedback welcome if you work on context routing / agent context.

cc @dexhorthy @swyx @RLanceMartin @cursor

https://t.co/togwA3iomD

[Alexander Young @Alex0nder]

Measured context engineering on 4 real codebases (149 Q). Full repo vs routed cores vs graph - numbers in the card. Open eval + raw runs https://t.co/togwA3iomD

@dexhorthy
@swyx
@RLanceMartin

@levie
@hwchase17
curious if this matches your prod experience? validity-audit.md

[Alexander Young @Alex0nder]

Updated card: compression = API input tokens (80× Oiloop), not core-char ratio (112×).
Same results, clearer metric labels after internal audit.

https://t.co/waS2U5nz90

[Alexander Young @Alex0nder]

149 Q · 4 codebases · A/B/C eval (gpt-4o-mini)

Full repo vs domain cores vs graph retrieval.
Keyword router on every project.

B ≥ A on all 4: MailAgent +21%, Oiloop +5%.
Up to 112× compression.

Open protocol + raw runs ↓

[Alexander Young @Alex0nder]

Reports:
https://t.co/togwA3iomD

· PHASE-2-RESULTS.md
· PHASE-3-RESULTS.md
· experiments/ (reproducible runs)

Caveats: LLM-as-judge, Oiloop N=20, expert prefs from decode — not fresh blind humans.

#ContextEngineering #RAG

[Alexander Young @Alex0nder]

139 Q · 4 codebases · A/B/C

Cores beat full repo on 3 OSS projects (+19–24%, up to 45×).

Private macOS app: first counterexample — cores lost; graph won.

Partially supported.
https://t.co/togwA3iomD

#ContextEngineering

[Alexander Young @Alex0nder]

139 Q, 4 codebases: full repo (A) vs domain cores (B) vs graph (C).

3 OSS projects: cores beat full repo (+19–24% accuracy, 14–45× compression).

Phase 3 private macOS: first case where B accuracy < A.

Partially supported.

[Alexander Young @Alex0nder]

Oiloop (private macOS, 20 Q):

A 1.20 · B 1.05 · C 1.55
Hallucination B 25% · C 15%
B: 83× compression, ~3× faster

Cores aren't universal. On integrated codebases, graph (C) wins.

MailAgent → B. Django, Navorina, Oiloop → C.

[Alexander Young @Alex0nder]

Details:
• Phase 3 report: …/oiloop-phase-3.md
• Raw run: …/run-1781225808172
• Paper draft incoming

[Alexander Young @Alex0nder]

Phase 3 done — 4 codebases, 139 eval questions (gpt-4o-mini).

We compared full repo vs routed context cores vs code graph retrieval.

New: private macOS Swift app (Oiloop) — 81k → 979 tokens, ~3× faster.

https://t.co/togwA3iomD

#ContextEngineering #LocalLLM #OpenSource

[Alexander Young @Alex0nder]

Oiloop (Phase 3) honest take:
• B (cores): 83× compression, fastest — but accuracy 1.05 vs 1.20 full repo
• C (graph): best accuracy 1.55, 15% hallucination
Rule of thumb:
local/Ollama: route cores
hard cross-cutting system Qs: graph
Not a silver bullet. A measured tradeoff.

[Alexander Young @Alex0nder]

@pdlug You're right, NV16-C is it. Zero endpoints, no halluc flag. B named them.Binary judge rewards vagueness. Claim grounding next. D is the real path, not separate B/C baselines

[Alexander Young @Alex0nder]

I tested 3 ways to feed context to coding agents:

A) dump the whole repo
B) route to domain-specific "context cores"
C) graph retrieval

119 questions / 3 codebases / gpt-4o-mini

Results 👇

[Alexander Young @Alex0nder]

@pdlug Happy to run the same A/B/C on your codebase if you want OSS repo or a scoped slice works. 35 Q, ~$0.50 in API costs

[Alexander Young @Alex0nder]

@pdlug No. Cores won on accuracy (+20%) and cost (45× fewer tokens)

Graph won on hallucination (14% vs 23%), but accuracy was = full repo, not better than cores

Tradeoff - cores for speed/accuracy and graph when trust matters more

[Alexander Young @Alex0nder]

Full report + open eval protocol + raw runs:

https://t.co/iAlvif4kaM

Repo: https://t.co/yU63pxvRJm

@swyx - does this match your context engineering tradeoffs?

@cursor_ai - curious if cores vs graph maps to your prod pipeline.

#AIAgents #ContextEngineering #RAG #BuildInPublic

[Alexander Young @Alex0nder]

The tradeoff:

B (cores) = best accuracy + cheapest → 19–23% hallucination
C (graph) = lower accuracy → 7–14% hallucination

Complex codebases → C for production.
Narrow domains → B wins.