"Harness Appendix E2 — Source Map and Fact-Checking: How to Separate Primary Sources, Working Notes, and Cases"

In a fast-moving topic like harness engineering, source discipline matters more than polished phrasing. If reading notes, official docs, public case studies, and local repository examples are blended without boundaries, the article may sound persuasive while becoming hard to verify. This appendix is a practical source protocol for that problem.

Key Takeaways

  • The first step is to separate claim type from source tier.
  • Anything about product capabilities, policies, dates, or official structure should be rechecked against primary sources.
  • Files in sources/ are excellent starting points, but they are not automatically final evidence.
  • Public company cases should be used only within the scope the company itself has disclosed.
  • Good fact-checking is not about piling up links. It is about showing which class of source supports which class of sentence.

1. Why a source map matters

docs/blog_series_ํ•˜๋„ค์Šค์—”์ง€๋‹ˆ์–ด๋ง_์ด๊ด„_design.md already establishes a concise verification protocol for the series.

  • paraphrase chapter arguments rather than copying them
  • re-verify product capabilities and documentation paths at publication time
  • use company cases only from what those companies publicly disclose
  • keep community references as trend context, not as the base for structural claims

The missing step is operationalization. That is what a source map provides.

2. Four source tiers

Tier 1. Primary sources

  • official documentation
  • product-owner feature descriptions
  • official blogs, release notes, or specifications
  • research or case material published by the original author or institution

Use for:

  • existence of a capability
  • policy, permissions, API structure
  • dates, versions, official terminology
  • company-case factual claims

Tier 2. Working notes

  • sources/260518_ํ•˜๋„ค์Šค์—”์ง€๋‹ˆ์–ด๋ง_15์žฅ_๋ธ”๋กœ๊ทธํ™œ์šฉ๋…ธํŠธ.md
  • reading notes, synthesis notes, chapter breakdown notes

Use for:

  • article structure
  • argument compression
  • comparative framing

Do not use as:

  • the final proof of a product claim
  • the final basis for numbers, policy statements, or latest-state assertions

Tier 3. Public case explanations

  • engineering blogs
  • public product case writeups
  • talks, demos, interviews, or conference presentations

Use for:

  • operational patterns
  • context around design choices
  • practical framing of how a setup is used

Warning:

  • do not infer undisclosed internals
  • do not blur "appears to work this way" into "is confirmed to work this way"

Tier 4. Local repository evidence

  • AGENTS.md
  • docs/memory-map.md
  • tasks/handoffs/, tasks/sessions/
  • existing series drafts and operating docs

Use for:

  • repository-native mapping
  • turning abstract concepts into concrete operating examples
  • practical application sections backed by actual paths

Warning:

  • do not present local working habits as universal industry rules
  • keep "example" distinct from "general law"

3. What kind of source does each sentence need

Sentence type Default source requirement
"this feature exists" primary source
"this company publicly uses this structure" that company's public primary material
"the main chapter argument is this" working notes plus source recheck when needed
"our repository maps the idea this way" local files and actual paths
"the field seems to be moving this way" multiple cases, clearly framed as interpretation

The core discipline is simple: interpretive sentences are allowed, but they should stay visibly interpretive.

4. A fast fact-checking workflow for harness articles

Step 1. Split the draft into three sentence classes

  • factual sentences
  • interpretive sentences
  • application sentences

Step 2. Mark the factual ones separately

Ask:

  • could this change over time
  • does this depend on a vendor or company claim
  • should a reader be able to verify this in the original source

If any answer is yes, it belongs in primary-source review.

Step 3. Keep working notes in the structure lane

sources/260518_ํ•˜๋„ค์Šค์—”์ง€๋‹ˆ์–ด๋ง_15์žฅ_๋ธ”๋กœ๊ทธํ™œ์šฉ๋…ธํŠธ.md is valuable for framing and chapter synthesis. It should remain strongest in that role.

  • valid use: chapter message compression, series mapping, comparison angles
  • invalid use: proving current feature support, proving current policy, proving official terminology

Step 4. Leave concrete paths for local examples

When using repository-native examples, avoid abstract claims without reproducible anchors.

Examples:

  • AGENTS.md
  • docs/memory-map.md
  • tasks/handoffs/

Step 5. Narrow uncertain statements instead of inflating them

When certainty is weak, reduce scope.

  • "industry standard" becomes "a recurring pattern across several public cases"
  • "always" becomes "in this repository and in the cited cases"
  • "officially supports" becomes "is documented in the official docs"

5. A source-map example for this harness series

Material Best role Safe to quote as direct factual support
sources/260518_ํ•˜๋„ค์Šค์—”์ง€๋‹ˆ์–ด๋ง_15์žฅ_๋ธ”๋กœ๊ทธํ™œ์šฉ๋…ธํŠธ.md chapter reinterpretation and framing limited
docs/blog_series_ํ•˜๋„ค์Šค์—”์ง€๋‹ˆ์–ด๋ง_์ด๊ด„_design.md series scope, mapping, internal protocol yes, for internal planning claims
AGENTS.md workspace boundary and operating rules yes, for local repository examples
docs/memory-map.md memory-index example yes, for local memory organization examples
public company docs capability, policy, and structure verification yes, preferred for structural claims

6. Common failure modes

Treating working notes as primary evidence

Notes are excellent for composition but weak as final proof.

Overstating case explanations

Even a company blog is only evidence for what it explicitly discloses.

Presenting local practice as industry law

Repository examples are powerful, but they are still examples.

Forgetting dates and recency

Product and policy claims decay. Verification time matters.

7. Minimal pre-publication checklist

  1. Did all capability, policy, and date claims get rechecked against primary sources?
  2. Are sources/ documents being used as structure notes rather than final evidence?
  3. Are local examples paired with actual repository paths?
  4. Are case interpretation and factual reporting separated?
  5. Were uncertain claims narrowed instead of overstated?

8. How this continues E1

E1 organized term boundaries. E2 organizes evidence boundaries.

  • E1 answers: what does this term mean
  • E2 answers: what kind of source can support this sentence

Reliable harness writing needs both.

References

  • AGENTS.md
  • docs/blog_series_ํ•˜๋„ค์Šค์—”์ง€๋‹ˆ์–ด๋ง_์ด๊ด„_design.md
  • docs/memory-map.md
  • sources/260518_ํ•˜๋„ค์Šค์—”์ง€๋‹ˆ์–ด๋ง_15์žฅ_๋ธ”๋กœ๊ทธํ™œ์šฉ๋…ธํŠธ.md
  • drafts/blog/260519_ํ•˜๋„ค์Šค๋ถ€๋กE01_์šฉ์–ด์ง‘๊ณผ์น˜ํŠธ์‹œํŠธ_๋ธ”๋กœ๊ทธ.md

This is Appendix Companion E2. Next: a practical harness workbook for breaking a real task into instructions, tools, verification, and handoff artifacts.

Series overview: Harness Engineering Series Guide

๋Œ“๊ธ€

๋Œ“๊ธ€ ์“ฐ๊ธฐ

์ด ๋ธ”๋กœ๊ทธ์˜ ์ธ๊ธฐ ๊ฒŒ์‹œ๋ฌผ

Agent Memory Engine (2/10) — Building an AI Agent Memory System with SQLite Alone

← 1/10 SQLite Memory Engine f… ๐Ÿ“š Series Index 3/10 memcore Completed → A memory engine that runs without daemons, without dependencies, anywhere ํ•ต์‹ฌ ์š”์•ฝ This post covers how to design an AI agent memory engine (memcore) on a single-file SQLite backend. - Structure: 25 modules, ~6,464 lines, 22 DB tables - Principles: no daemon, single-file portability, no required external dependencies, host-agnostic - Core techniques: 4-layer hybrid search (topic · vector · FTS5 · LIKE), U-tag dialectic-based confidence evolution, ingestion-layer bias rejection rules What This Post Covers One approach to building AI agent memory without a server or cloud vector DB. Specifically: (1) the limitations of text-file-based memory, (2) vector DB alternatives compared and selection criteria, (3) how to layer search tiers, (4) how to evolve confidence scores over time, and (5) how to block biased memory accumulation at the ingestion stage. Design Principles: Why Single-File SQLite Text-file-based ...
Read more »

"ML Foundations (9/9) — PyTorch vs TensorFlow, and the Road to Local LLMs"

์ด๋ฏธ์ง€
← 8/9 Deep Learning Architec… ๐Ÿ“š Series Index (series end) The final part. So far we've focused on models. Now we focus on the tools that actually run them . We'll lay out the philosophical difference between PyTorch and TensorFlow, and trace how Hugging Face transformers , llama.cpp , MLX , and Ollama built the bridge to running large language models on your own machine. By the end you should have the full mental model of "download a pretrained LLM and serve it locally." 0. Learning Objectives Explain the eager-vs-graph difference between PyTorch and TensorFlow. Explain, in graph terms, how autograd automates the backward pass. Use Hugging Face transformers ' AutoModel , AutoTokenizer , and pipeline abstractions. Describe llama.cpp's GGUF quantization, INT4 inference, and CPU-first flow. Describe Apple MLX's use of unified memory and how it differs from PyTorch. Run a local LLM with Ollama and call it through an OpenAI-compatible API. ...
Read more »

"RAG Core Study (14/26) — Evaluation Sets with RAGAS & DeepEval"

์ด๋ฏธ์ง€
Series overview: Series index ๐Ÿ“š Series Index Without an evaluation set, every RAG improvement is a story, not evidence. RAG systems fail in more than one place: retrieval, context selection, answer generation, and grounding. That means "looks better" is not a metric. Part 14 explains how to build a golden dataset , how RAGAS and DeepEval fit into the workflow, and why teams must separate retrieval quality from answer quality if they want tuning decisions to hold up. 0. Prerequisites Part 13 reranking — system quality now depends on multi-stage ranking. Part 1 grounding — a correct-looking answer is not enough. Part 12 Hybrid — multiple retrieval settings need comparison on fixed data. 1. Learning Objectives Build a small but useful golden dataset for RAG. Distinguish retrieval metrics from answer metrics . Use RAGAS and DeepEval in the right roles. Avoid the common traps in judge-model-based evaluation. 2. ํ•ต์‹ฌ ์š”์•ฝ An evaluation set for RAG is ...
Read more »

"ML Foundations (8/9) — Deep Learning Architectures: CNN, RNN, Attention"

์ด๋ฏธ์ง€
← 7/9 Deep Learning Training ๐Ÿ“š Series Index 9/9 PyTorch vs TensorFlow → What did we keep adding on top of an MLP? Vision went CNN, sequences went RNN/LSTM, and eventually both converged on Attention and the Transformer. Each architecture is easier to remember if you read it as what it gave up to gain something else . This part is the one-line summary of the decisive inflection points: LeNet → AlexNet → ResNet → LSTM → Attention → Transformer. 0. Learning Objectives Explain why CNN's convolution, pooling, and weight sharing match images so well. Trace what got unblocked from LeNet → AlexNet → VGG → ResNet as depth grew. State BPTT for RNNs and the vanishing/exploding gradient problem. Write the LSTM gate equations (forget, input, output) with intuition. Write the attention formula in query/key/value form and the scaled dot-product variant. State the precise reasons Transformers replaced RNNs (parallelism + long-range dependencies). 1. ํ•ต์‹ฌ ์š”์•ฝ CNN : weight sharin...
Read more »

"ML Foundations (7/9) — Deep Learning Training: Optimizers, Regularization, Initialization"

← 6/9 Neural Networks ๐Ÿ“š Series Index 8/9 Deep Learning Architec… → Building an MLP, as we did in Part 6, is not the same as getting it to train well . The right optimizer, regularization, initialization, and learning rate — when these line up, deep networks converge. When they don't, the network refuses to learn at all. This part is the catalog of those crafts, with formulas, paper citations, and code in one place. 0. Learning Objectives Compare and write the update rules for SGD, Momentum, Nesterov, and Adam. Explain how Dropout, BatchNorm, and LayerNorm work and where they belong in a model. Derive the variance formulas for Xavier and He initialization and match them to activations. Implement step, cosine, and warmup learning-rate schedules in PyTorch. Explain why gradient clipping is effectively required for RNNs and Transformers. Diagnose the most common training failures (NaN, plateau, overfitting) and apply first-line fixes. 1. ํ•ต์‹ฌ ์š”์•ฝ SGD : \(w \leftarro...
Read more »

OpenClaw to Hermes Migration (2/13) — What to Preserve, Partially Port, or Discard

← 1/13 Current Structure Inve… ๐Ÿ“š Series Index 3/13 What Moves to Hermes → A code review record classifying operational scripts into three tiers: preserve, partial-preserve, and discard. What This Post Covers An asset classification framework (preserve / partial-preserve / discard) for migrating a legacy agent system to a successor A concrete case applying the Strangler Fig + Subset Migration strategy to a memory pipeline Classification results evaluated across LOC, external dependencies, and replaceability axes A method for slimming down a bloated single file (1,310 LOC) using the migration as a natural entry point Problem Definition: Port Everything or Cut Deep? The OpenClaw inventory comprises multiple agents, multiple scheduler/daemon processes, and multiple operational scripts. When moving to the successor system, Hermes, the choice is not binary. A full port carries accumulated technical debt along with it; a minimal port discards hard-won domain logic. Code-revie...
Read more »

AI Agents I Built (5/7) — Building an Automated Blogger API Publishing System

← 4/7 My Life Organizer Agent ๐Ÿ“š Series Index 6/7 Dual → Designing a Markdown-Based Auto-Publish Pipeline with Blogger API v3 + OAuth 2.0 Key Summary Blogger API v3 + OAuth 2.0 enables a pipeline that converts Markdown files into publishable HTML posts blogger_publish.py handles the full flow — frontmatter parsing → body cleanup → HTML conversion → CSS injection → API post — in a single script For bulk publishing, quota management (delay, retry, TOC suppression) is the critical design concern Platform Selection Criteria API access is mandatory for AI agent-driven content publishing. A comparison of platforms: Platform API Cost AdSense Decision WordPress.com REST API Paid plan required Paid only Cost overhead Ghost REST API Self-hosted or paid Manual setup Ops overhead Blogger v3 API Free Native integration Adopted Blogger is free, exposes a REST API, and integrates natively with AdSense. For personal blog automation, this combination i...
Read more »