Skip to content

content: 8-article Reference Architecture series - #16

Open
kimiko-terraphim wants to merge 1 commit into
mainfrom
articles/reference-architecture-series-2026-09
Open

content: 8-article Reference Architecture series#16
kimiko-terraphim wants to merge 1 commit into
mainfrom
articles/reference-architecture-series-2026-09

Conversation

@kimiko-terraphim

Copy link
Copy Markdown

Summary

This PR adds an 8-article series on production agent systems, harness engineering, and the Terraphim architecture to the blog.

Articles Added

Series 1: The Harness Determines Success

  1. The Three Jobs of an Agent Harness (Sep 9) — Context curation, execution guardrails, memory infrastructure. LangChain's +26% improvement with zero model changes.
  2. Why 1M Token Windows Fail (Sep 10) — Three failure modes: noise accumulation, instruction collision, attention decay.
  3. The 85/15 Planning Ratio (Sep 11) — 85% planning, 15% execution. Most agents invert this and thrash.

Series 2: Context Architecture

  1. Structured Compaction: The Missing Layer (Sep 12) — Five-stage pipeline: Ingest → Filter → Rank → Compact → Inject.
  2. Role-Based Context Dispatch (Sep 13) — Agents need multiple personalities. Roles are configuration, not prompts.

Series 3: Memory Infrastructure

  1. The Knowledge Graph as Agent Memory (Sep 14) — Why vectors aren't enough for production agent memory.

Series 4: Production Patterns

  1. Risk-Tiered Execution (Sep 15) — Safe / Review / Critical tiers with machine-readable risk contracts.
  2. The Deterministic Layer (Sep 16) — Where probabilistic models don't belong: routing, validation, filtering.

Quality Features

  • Debate sections with counter-arguments examined and refuted
  • Real-world case studies from production failures
  • Rust code examples from actual Terraphim crates
  • Measurement frameworks for each concept
  • Cross-references to existing articles and Terraphim documentation

Staging Schedule

Date Article
Sep 9 The Three Jobs of an Agent Harness
Sep 10 Why 1M Token Windows Fail
Sep 11 The 85/15 Planning Ratio
Sep 12 Structured Compaction
Sep 13 Role-Based Context Dispatch
Sep 14 The Knowledge Graph as Agent Memory
Sep 15 Risk-Tiered Execution
Sep 16 The Deterministic Layer

All articles written by Alexander Mikhalev. Cross-links to existing Terraphim articles included where relevant.

Add comprehensive article series on production agent systems:

1. The Three Jobs of an Agent Harness — Context curation, guardrails,
   memory infrastructure. LangChain's +26% improvement with zero model changes.

2. Why 1M Token Windows Fail — Three failure modes: noise accumulation,
   instruction collision, attention decay. Curated signal beats raw capacity.

3. The 85/15 Planning Ratio — Agents should spend 85% of compute on
   planning and 15% on execution. Most systems invert this ratio.

4. Structured Compaction — The five-stage compaction pipeline:
   Ingest → Filter → Rank → Compact → Inject. The missing layer between
   retrieval and the LLM.

5. Role-Based Context Dispatch — Agents need multiple personalities.
   Roles are configuration, not prompts.

6. The Knowledge Graph as Agent Memory — Why vectors aren't enough:
   probabilistic retrieval, un-auditable results, no structured relations.
   KG provides deterministic, explainable memory.

7. Risk-Tiered Execution — Three tiers: Safe (auto), Review (single
   approval), Critical (dual + audit). Machine-readable risk contracts.

8. The Deterministic Layer — Where probabilistic models don't belong:
   routing, validation, filtering, authorization.

All articles include debate sections, real-world case studies, Rust code
examples from Terraphim, measurement frameworks, and reference implementations.
@netlify

netlify Bot commented Sep 8, 2026

Copy link
Copy Markdown

Deploy Preview for clinquant-croquembouche-2fa149 failed.

Name Link
🔨 Latest commit 10dfe1e
🔍 Latest deploy log https://app.netlify.com/projects/clinquant-croquembouche-2fa149/deploys/6aa04d2a20e324000853dede

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant