{
 "ledger": {
  "title": "Claim Ledger: What Is Harness Engineering? How to Build AI Agents That Catch Their Own Mistakes",
  "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/",
  "essay": {
   "title": "What Is Harness Engineering? How to Build AI Agents That Catch Their Own Mistakes",
   "url": "https://durabilitycurve.com/blog/what-is-harness-engineering/",
   "published": "2026-10-22"
  },
  "last_checked": "2026-10-22",
  "format": "current",
  "counts": {
   "claims": 85,
   "by_status": {
    "VERIFIED": 84,
    "EXECUTED": 1
   },
   "verified_at_primary": 84,
   "removed": 0
  },
  "license": "https://creativecommons.org/licenses/by/4.0/",
  "attribution": "The Durability Curve (Harry Floyd), https://durabilitycurve.com/claims/what-is-harness-engineering/"
 },
 "rows": [
  {
   "id": "1",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-1",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "Thoughtworks describes an agent renaming a field that two downstream services consume; calls it locally correct but systemically destructive",
   "quote": "\"locally correct, but systemically destructive\"",
   "removal_reason": "",
   "source": "Thoughtworks (Nandyala, Pani), 29 Sep 2026",
   "source_url": "https://www.thoughtworks.com/insights/blog/architecture/engineering-the-harness-a-practical-pattern-for-reliable-coding-agents",
   "locator": "opening section",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "2",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-2",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "Thoughtworks says the gap is not mainly about model quality",
   "quote": "\"is not primarily a model-quality issue\"",
   "removal_reason": "",
   "source": "Thoughtworks",
   "source_url": "",
   "locator": "opening section",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "3",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-3",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "Agent = Model + Harness",
   "quote": "\"Agent = Model + Harness\"",
   "removal_reason": "",
   "source": "Böckeler, martinfowler.com, 2 Apr 2026 (also Thoughtworks)",
   "source_url": "https://martinfowler.com/articles/harness-engineering.html",
   "locator": "early section",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "4",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-4",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "Part of the harness is built into the tool (system prompt, code retrieval); users build an outer harness for their use case",
   "quote": "\"outer harness\"",
   "removal_reason": "",
   "source": "Böckeler",
   "source_url": "",
   "locator": "\"Metaphors only go so far\" area",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "5",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-5",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "A good outer harness raises first-time success and gives a self-correcting feedback loop",
   "quote": "\"self-corrects as many issues as possible before they even reach human eyes\"",
   "removal_reason": "",
   "source": "Böckeler",
   "source_url": "",
   "locator": "after Figure 1",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "6",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-6",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "Hashimoto's definition of harness engineering; he says in this post he had started calling the practice harness engineering",
   "quote": "\"is the idea that anytime you find an agent makes a mistake, you take the time to engineer a solution such that the agent never makes that mistake again\"",
   "removal_reason": "",
   "source": "Hashimoto, My AI Adoption Journey, 5 Feb 2026",
   "source_url": "https://mitchellh.com/writing/my-ai-adoption-journey",
   "locator": "Step 5",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "7",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-7",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "Anthropic calls context engineering the next step after prompt engineering",
   "quote": "\"the natural progression of prompt engineering\"",
   "removal_reason": "",
   "source": "Anthropic, 29 Sep 2025",
   "source_url": "https://www.anthropic.com/research/effective-context-engineering-for-ai-agents",
   "locator": "intro",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "8",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-8",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "Context engineering targets one output; harness engineering targets quality over time",
   "quote": "(paraphrase of comparison table: \"A single output is inaccurate\" vs \"Quality degrades over time\")",
   "removal_reason": "",
   "source": "SmartScope, Feb 2026, restructured Mar 2026",
   "source_url": "https://smartscope.blog/en/blog/harness-engineering-overview/",
   "locator": "comparison table",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "9",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-9",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "AGENTS.md is a shared format; nearest file wins in nested repos",
   "quote": "\"so the closest one takes precedence\"",
   "removal_reason": "",
   "source": "agents.md",
   "source_url": "https://agents.md/",
   "locator": "FAQ / monorepo",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "10",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-10",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "Claude Code reads AGENTS.md only when no CLAUDE.md exists (default)",
   "quote": "\"By default, Claude reads AGENTS.md only when you have no CLAUDE.md in your working directory or above it\"",
   "removal_reason": "",
   "source": "Claude Code docs",
   "source_url": "https://code.claude.com/docs/en/memory",
   "locator": "When Claude Code reads AGENTS.md",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "11",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-11",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "OpenAI tried one big AGENTS.md and it failed",
   "quote": "\"one big AGENTS.md\"",
   "removal_reason": "",
   "source": "OpenAI (Lopopolo), 11 Feb 2026",
   "source_url": "https://openai.com/index/harness-engineering/",
   "locator": "We made repository knowledge the system of record",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "12",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-12",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "A giant file crowds out the task",
   "quote": "\"crowds out the task\"",
   "removal_reason": "",
   "source": "OpenAI",
   "source_url": "",
   "locator": "same",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "13",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-13",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "Too much guidance becomes non-guidance",
   "quote": "\"When everything is 'important,' nothing is.\"",
   "removal_reason": "",
   "source": "OpenAI",
   "source_url": "",
   "locator": "same",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "14",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-14",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "OpenAI's AGENTS.md is about 100 lines and acts as a table of contents pointing to docs/",
   "quote": "\"roughly 100 lines\"; \"the table of contents\"",
   "removal_reason": "",
   "source": "OpenAI",
   "source_url": "",
   "locator": "same",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "15",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-15",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "Map, not manual",
   "quote": "\"give Codex a map, not a 1,000-page instruction manual\"",
   "removal_reason": "",
   "source": "OpenAI",
   "source_url": "",
   "locator": "same",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "16",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-16",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "What the agent can't access in context doesn't exist for it",
   "quote": "\"anything it can't access in-context while running effectively doesn't exist\"",
   "removal_reason": "",
   "source": "OpenAI",
   "source_url": "",
   "locator": "Agent legibility is the goal",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "17",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-17",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "HumanLayer's CLAUDE.md is under 60 lines; craft by hand, don't auto-generate",
   "quote": "\"under 60 lines\"",
   "removal_reason": "",
   "source": "HumanLayer (Kyle), 12 Mar 2026",
   "source_url": "https://www.hlyr.dev/blog/skill-issue-harness-engineering-for-coding-agents",
   "locator": "CLAUDE.md section",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "18",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-18",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "/init output brings little value",
   "quote": "\"tends to produce documents that bring little meaningful value\"",
   "removal_reason": "",
   "source": "Software Mansion, updated 22 Jun 2026",
   "source_url": "https://agentic-engineering.swmansion.com/becoming-productive/harness-engineering/",
   "locator": "AGENTS.md",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "19",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-19",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "Put-in / leave-out table adapted from Anthropic's Claude Code docs, as quoted by Software Mansion",
   "quote": "(adapted table)",
   "removal_reason": "",
   "source": "Software Mansion",
   "source_url": "",
   "locator": "AGENTS.md: good and bad practices",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "20",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-20",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "Recall falls as context grows",
   "quote": "\"the model's ability to accurately recall information from that context decreases\"",
   "removal_reason": "",
   "source": "Anthropic context engineering",
   "source_url": "https://www.anthropic.com/research/effective-context-engineering-for-ai-agents",
   "locator": "context rot",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "21",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-21",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "Aim for the smallest high-signal set",
   "quote": "\"the smallest possible set of high-signal tokens\"",
   "removal_reason": "",
   "source": "Anthropic context engineering",
   "source_url": "",
   "locator": "same",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "22",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-22",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "Three techniques: tool-result clearing, compaction, notes/memory; clearing swaps old results for a placeholder and the agent can re-fetch",
   "quote": "(paraphrase)",
   "removal_reason": "",
   "source": "Claude Cookbook (Isabella He), 20 Mar 2026",
   "source_url": "https://platform.claude.com/cookbook/tool-use-context-engineering-context-engineering-tools",
   "locator": "summary table / clearing",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "23",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-23",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "Clearing is a light, safe form of compaction",
   "quote": "\"one of the safest lightest touch forms of compaction\"",
   "removal_reason": "",
   "source": "Anthropic context engineering",
   "source_url": "https://www.anthropic.com/research/effective-context-engineering-for-ai-agents",
   "locator": "compaction",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "24",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-24",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "Sourcegraph's compaction prompt preserves task, files inspected, decisions, failed approaches",
   "quote": "\"approaches already tried that failed, and why\"; \"and the fresh window will cheerfully retry them\"",
   "removal_reason": "",
   "source": "Sourcegraph (Tanner), 15 Sep 2026",
   "source_url": "https://sourcegraph.com/blog/context-engineering",
   "locator": "compaction prompt",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "25",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-25",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "Databricks' own memory system (MemCurrent) on its internal benchmark of enterprise tasks: 22.3% (no memory) → 68.4% pass rate, 31% fewer tokens, 26% faster than no memory; Mem0 reached 43.9% on the same benchmark",
   "quote": "\"more history is not necessarily better memory\"",
   "removal_reason": "",
   "source": "Databricks, 20 Oct 2026",
   "source_url": "https://www.databricks.com/blog/memcurrent-designing-persistent-memory-production-agents",
   "locator": "intro + Figure 6",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "26",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-26",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "A stale memory is worse than none",
   "quote": "\"is worse than no memory: the agent holds a confident, wrong prior it will defend\"",
   "removal_reason": "",
   "source": "Sourcegraph",
   "source_url": "https://sourcegraph.com/blog/context-engineering",
   "locator": "memory invalidation (JWT example)",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "27",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-27",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "Mechanism rule of thumb (AGENTS.md, skills, MCP, subagents, hooks)",
   "quote": "(paraphrased table)",
   "removal_reason": "",
   "source": "Software Mansion",
   "source_url": "https://agentic-engineering.swmansion.com/becoming-productive/harness-engineering/",
   "locator": "When to use what",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "28",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-28",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "Role-based subagents don't work; use them for context control",
   "quote": "\"It doesn't work.\"",
   "removal_reason": "",
   "source": "HumanLayer",
   "source_url": "https://www.hlyr.dev/blog/skill-issue-harness-engineering-for-coding-agents",
   "locator": "Sub-Agents Are for Context Control",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "29",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-29",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "Subagents may use tens of thousands of tokens and return 1,000-2,000",
   "quote": "\"tens of thousands of tokens\"; \"often 1,000-2,000 tokens\"",
   "removal_reason": "",
   "source": "Anthropic context engineering",
   "source_url": "https://www.anthropic.com/research/effective-context-engineering-for-ai-agents",
   "locator": "sub-agent architectures",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "30",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-30",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "Too many tools is bad",
   "quote": "(heading \"Too Many Tools Is Bad\")",
   "removal_reason": "",
   "source": "HumanLayer",
   "source_url": "https://www.hlyr.dev/blog/skill-issue-harness-engineering-for-coding-agents",
   "locator": "MCP Servers section",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "31",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-31",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "Ambiguous tool choice defeats agents too",
   "quote": "\"If a human engineer can't definitively say which tool should be used in a given situation, an AI agent can't be expected to do better.\"",
   "removal_reason": "",
   "source": "Anthropic context engineering",
   "source_url": "https://www.anthropic.com/research/effective-context-engineering-for-ai-agents",
   "locator": "tools",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "32",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-32",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "Don't connect untrusted MCP servers",
   "quote": "\"never connect to one you don't trust. This can be a dangerous vector for prompt injection!\"",
   "removal_reason": "",
   "source": "HumanLayer",
   "source_url": "https://www.hlyr.dev/blog/skill-issue-harness-engineering-for-coding-agents",
   "locator": "MCP warning",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "33",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-33",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "MCP spec: clients must treat tool annotations as untrusted unless from trusted servers",
   "quote": "\"clients MUST consider tool annotations to be untrusted unless they come from trusted servers\"",
   "removal_reason": "",
   "source": "MCP specification 2025-06-18",
   "source_url": "https://modelcontextprotocol.io/specification/2025-06-18/server/tools",
   "locator": "Tool definition",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "34",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-34",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "Published attacks: a skill step fetched and ran ransomware; hidden Unicode instructions in a skill",
   "quote": "\"visual source review alone may miss malicious behavior\"",
   "removal_reason": "",
   "source": "Software Mansion",
   "source_url": "https://agentic-engineering.swmansion.com/becoming-productive/harness-engineering/",
   "locator": "Skills: Security considerations",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "35",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-35",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "HumanLayer auto-denies migration commands and tells the agent to ask the user",
   "quote": "(paraphrase: \"we automatically deny any Bash() tool calls that try to run migrations, with an instruction to ask the user to run them instead\")",
   "removal_reason": "",
   "source": "HumanLayer",
   "source_url": "",
   "locator": "Hooks",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "36",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-36",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "Guides steer before, sensors check after",
   "quote": "\"Guides (feedforward controls)\"; \"Sensors (feedback controls)\"",
   "removal_reason": "",
   "source": "Böckeler",
   "source_url": "https://martinfowler.com/articles/harness-engineering.html",
   "locator": "Feedforward and Feedback",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "37",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-37",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "Guides alone / sensors alone failure modes",
   "quote": "\"that encodes rules but never finds out whether they worked\"; \"that keeps repeating the same mistakes\"",
   "removal_reason": "",
   "source": "Böckeler",
   "source_url": "",
   "locator": "same",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "38",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-38",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "The rule is called silent success, verbose failure",
   "quote": "\"silent success, verbose failure\"",
   "removal_reason": "",
   "source": "Thoughtworks",
   "source_url": "https://www.thoughtworks.com/insights/blog/architecture/engineering-the-harness-a-practical-pattern-for-reliable-coding-agents",
   "locator": "Sensors",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "39",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-39",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "Full test suite output flooded context; the agent lost track",
   "quote": "\"4,000 lines of passing tests would flood the context window\"",
   "removal_reason": "",
   "source": "HumanLayer",
   "source_url": "https://www.hlyr.dev/blog/skill-issue-harness-engineering-for-coding-agents",
   "locator": "Verification",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "40",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-40",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "Custom lint messages carry fixes",
   "quote": "\"inject remediation instructions into agent context\"",
   "removal_reason": "",
   "source": "OpenAI",
   "source_url": "https://openai.com/index/harness-engineering/",
   "locator": "Enforcing architecture and taste",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "41",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-41",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "Böckeler calls this positive prompt injection",
   "quote": "\"a positive kind of prompt injection\"",
   "removal_reason": "",
   "source": "Böckeler",
   "source_url": "https://martinfowler.com/articles/harness-engineering.html",
   "locator": "sensors",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "42",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-42",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "Repeated prose rules should become sensors",
   "quote": "\"should be promoted into a mechanical sensor\"",
   "removal_reason": "",
   "source": "Thoughtworks",
   "source_url": "https://www.thoughtworks.com/insights/blog/architecture/engineering-the-harness-a-practical-pattern-for-reliable-coding-agents",
   "locator": "sensors",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "43",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-43",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "Instructions for judgement, hooks for the rest",
   "quote": "\"Written instructions should cover the cases that require judgment. Hooks should cover the cases that do not.\"",
   "removal_reason": "",
   "source": "Software Mansion",
   "source_url": "https://agentic-engineering.swmansion.com/becoming-productive/harness-engineering/",
   "locator": "Hooks",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "44",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-44",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "In Anthropic's long-running setup, later sessions could only flip a feature-list item's passes flag, under an instruction not to remove or edit tests (the 'tests' are feature-list entries)",
   "quote": "\"unacceptable to remove or edit tests\"",
   "removal_reason": "",
   "source": "Anthropic long-running agents, 26 Nov 2025",
   "source_url": "https://www.anthropic.com/research/effective-harnesses-for-long-running-agents",
   "locator": "Feature list",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "45",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-45",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "Each new session starts with no memory; engineers-in-shifts analogy",
   "quote": "\"begins with no memory of what came before\"",
   "removal_reason": "",
   "source": "Anthropic long-running",
   "source_url": "",
   "locator": "The long-running agent problem",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "46",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-46",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "Two failure modes: doing too much at once; later agents declaring the job done",
   "quote": "\"to do too much at once\"; \"declare the job done\"",
   "removal_reason": "",
   "source": "Anthropic long-running",
   "source_url": "",
   "locator": "same",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "47",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-47",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "Initializer writes 200+ features, all marked failing; JSON chosen over Markdown",
   "quote": "\"less likely to inappropriately change or overwrite JSON files compared to Markdown\"",
   "removal_reason": "",
   "source": "Anthropic long-running",
   "source_url": "",
   "locator": "Feature list",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "48",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-48",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "Progress file + git commits allow reverting bad changes; start-up routine reads git log and progress, picks top feature, runs a basic test",
   "quote": "(paraphrase)",
   "removal_reason": "",
   "source": "Anthropic long-running",
   "source_url": "",
   "locator": "Incremental progress; Getting up to speed",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "49",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-49",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "Testing as a human user caught bugs not visible in code",
   "quote": "\"as a human user would\"; \"that weren't obvious from the code alone\"",
   "removal_reason": "",
   "source": "Anthropic long-running",
   "source_url": "",
   "locator": "Testing",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "50",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-50",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "Human gates for irreversible or high-blast-radius choices (multi-repo scope, schema migrations, public API changes); otherwise sensible defaults",
   "quote": "\"irreversible or high-blast-radius choices\"; \"sensible defaults\"",
   "removal_reason": "",
   "source": "Thoughtworks",
   "source_url": "https://www.thoughtworks.com/insights/blog/architecture/engineering-the-harness-a-practical-pattern-for-reliable-coding-agents",
   "locator": "Guides: human gates",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "51",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-51",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "Least-privilege tools: verification agents get read-only tools",
   "quote": "(paraphrase: \"provisioned with read-only tools, literally lacking file-writing capabilities\")",
   "removal_reason": "",
   "source": "Thoughtworks",
   "source_url": "",
   "locator": "Least-privilege tools",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "52",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-52",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "Before You Trust Your AI Agent kit is free, seven checks, one page each",
   "quote": "\"Seven checks, one page each\"",
   "removal_reason": "",
   "source": "The Durability Curve, 9 Oct 2026",
   "source_url": "https://harryfloyd.substack.com/p/before-you-trust-your-ai-agent",
   "locator": "subtitle",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "53",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-53",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "Scores move a few points with no idea why",
   "quote": "\"by a few percentage points, and have no idea why it changed\"",
   "removal_reason": "",
   "source": "Google Developers Blog (Mullen, Gunderman), 9 Sep 2026",
   "source_url": "https://developers.googleblog.com/the-anatomy-of-harness-engineering-how-to-evaluate-iterate-and-guard-ai-coding-agents/",
   "locator": "intro",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "54",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-54",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "Behavioural evals check discrete actions (clarifying question, validator before done)",
   "quote": "(paraphrase)",
   "removal_reason": "",
   "source": "Google",
   "source_url": "",
   "locator": "behavioural eval examples",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "55",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-55",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "Loop: pick one failure mode; strict vs LLM-judge checks; batch runs because single runs are noisy",
   "quote": "(paraphrase)",
   "removal_reason": "",
   "source": "Google",
   "source_url": "",
   "locator": "three-step loop",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "56",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-56",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "Suite purpose",
   "quote": "\"did not make the agent holistically worse\"",
   "removal_reason": "",
   "source": "Google",
   "source_url": "",
   "locator": "When to evaluate",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "57",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-57",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "Log context on failed runs and check for the answer",
   "quote": "\"then check whether the answer was in there\"",
   "removal_reason": "",
   "source": "Sourcegraph",
   "source_url": "https://sourcegraph.com/blog/context-engineering",
   "locator": "observability",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "58",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-58",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "Cost scales with what you send",
   "quote": "\"Cost and latency scale with what you send, not with what you could send\"",
   "removal_reason": "",
   "source": "Sourcegraph",
   "source_url": "",
   "locator": "cost",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "59",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-59",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "Agents ~4x chat tokens, multi-agent ~15x",
   "quote": "\"agents typically use about 4× more tokens than chat interactions, and multi-agent systems use about 15× more tokens than chats\"",
   "removal_reason": "",
   "source": "Anthropic, 13 Jun 2025",
   "source_url": "https://www.anthropic.com/research/multi-agent-research-system",
   "locator": "benefits and downsides",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "60",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-60",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "Earn every rule",
   "quote": "\"Earn every rule\"",
   "removal_reason": "",
   "source": "Thoughtworks",
   "source_url": "https://www.thoughtworks.com/insights/blog/architecture/engineering-the-harness-a-practical-pattern-for-reliable-coding-agents",
   "locator": "close",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "61",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-61",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "HumanLayer discards most hooks",
   "quote": "\"I have thrown away many more hooks than we actually use today\"",
   "removal_reason": "",
   "source": "HumanLayer",
   "source_url": "https://www.hlyr.dev/blog/skill-issue-harness-engineering-for-coding-agents",
   "locator": "Hooks",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "62",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-62",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "ETH Zurich study: context files (LLM-generated and developer-committed) did not generally improve task success and raised inference cost by over 20%",
   "quote": "\"does not generally improve task success rates, while increasing inference cost by over 20% on average\"",
   "removal_reason": "",
   "source": "Gloaguen et al., arXiv 2602.11988, v3 29 Sep 2026",
   "source_url": "https://arxiv.org/abs/2602.11988",
   "locator": "Abstract",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "63",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-63",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "Repository overviews not helpful; files useful for non-standard practices",
   "quote": "\"repository overviews, although popular and recommended by model providers, are not helpful\"; \"useful for specifying non-standard coding practices\"",
   "removal_reason": "",
   "source": "Gloaguen et al.",
   "source_url": "https://arxiv.org/abs/2602.11988",
   "locator": "Abstract",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "64",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-64",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "Every harness component encodes an assumption about what the model can't do, and these go stale as models improve",
   "quote": "\"encodes an assumption about what the model can't do on its own\"; \"can quickly go stale as models improve\"",
   "removal_reason": "",
   "source": "Anthropic, Harness design for long-running application development, 24 Mar 2026",
   "source_url": "https://www.anthropic.com/research/harness-design-long-running-apps",
   "locator": "simplifying the harness",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "65",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-65",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "HumanLayer frames harness engineering as a subset of context engineering, via harness configuration points",
   "quote": "\"the subset of context engineering which primarily involves leveraging harness configuration points\"",
   "removal_reason": "",
   "source": "HumanLayer",
   "source_url": "https://www.hlyr.dev/blog/skill-issue-harness-engineering-for-coding-agents",
   "locator": "What is harness engineering",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "66",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-66",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "Tool-result clearing, compaction and memory are offered as API features (cookbook audience = API builders); Claude Code uses compaction and memory",
   "quote": "\"Our API offers first-party implementations of all three\"",
   "removal_reason": "",
   "source": "Claude Cookbook",
   "source_url": "https://platform.claude.com/cookbook/tool-use-context-engineering-context-engineering-tools",
   "locator": "intro",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "67",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-67",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "The comparison table's question row and 'typical failure' row come from SmartScope's context-vs-harness table",
   "quote": "(paraphrase: \"What to show\" vs \"What to prevent, measure, control, and fix\"; \"A single output is inaccurate\" vs \"Quality degrades over time\")",
   "removal_reason": "",
   "source": "SmartScope",
   "source_url": "https://smartscope.blog/en/blog/harness-engineering-overview/",
   "locator": "comparison table",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "68",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-68",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "Lulla et al.: with AGENTS.md, median runtime −28.64%, output tokens −16.58%, comparable task completion (10 repos, 124 PRs)",
   "quote": "\"lower median runtime\"; \"reduced output token consumption\"; \"comparable task completion behavior\"",
   "removal_reason": "",
   "source": "Lulla et al., arXiv 2601.20404, 28 Jan 2026",
   "source_url": "https://arxiv.org/abs/2601.20404",
   "locator": "Abstract",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "69",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-69",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "Thoughtworks: what the model lacked in the rename case was system-level visibility, guardrails and automated feedback",
   "quote": "\"system-level visibility, architectural guardrails and automated feedback mechanisms\"",
   "removal_reason": "",
   "source": "Thoughtworks",
   "source_url": "https://www.thoughtworks.com/insights/blog/architecture/engineering-the-harness-a-practical-pattern-for-reliable-coding-agents",
   "locator": "opening",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "70",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-70",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "Google runs evals in batches because single runs are noisy",
   "quote": "\"can be noisy due to nondeterminism of AI models\"",
   "removal_reason": "",
   "source": "Google Developers Blog",
   "source_url": "https://developers.googleblog.com/the-anatomy-of-harness-engineering-how-to-evaluate-iterate-and-guard-ai-coding-agents/",
   "locator": "batch evaluations",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "71",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-71",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "Anthropic's working rule, from Building Effective Agents",
   "quote": "\"find the simplest solution possible, and only increase complexity when needed\"",
   "removal_reason": "",
   "source": "Anthropic harness design",
   "source_url": "https://www.anthropic.com/research/harness-design-long-running-apps",
   "locator": "simplifying the harness",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "72",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-72",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "Anthropic's Tool Search Tool defers tool definitions until needed; ~85% fewer tokens in its example",
   "quote": "\"85% reduction in token usage\"; \"defer_loading\"",
   "removal_reason": "",
   "source": "Anthropic, Introducing advanced tool use, 24 Nov 2025",
   "source_url": "https://www.anthropic.com/research/advanced-tool-use",
   "locator": "Tool Search Tool",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "73",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-73",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "Claude Code defers MCP tool definitions until needed",
   "quote": "\"deferring tool definitions until Claude needs them\"",
   "removal_reason": "",
   "source": "Claude Code docs, MCP",
   "source_url": "https://code.claude.com/docs/en/mcp",
   "locator": "Scale with MCP tool search",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "74",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-74",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "Claude Code hooks run shell commands with the user's full permissions",
   "quote": "\"execute shell commands with your full user permissions\"",
   "removal_reason": "",
   "source": "Claude Code docs, hooks",
   "source_url": "https://code.claude.com/docs/en/hooks",
   "locator": "Security considerations",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "75",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-75",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "Claude Code deny rules block a tool and are evaluated first",
   "quote": "\"Deny rules prevent Claude Code from using the specified tool\"; \"Rules are evaluated in order: deny, then ask, then allow\"",
   "removal_reason": "",
   "source": "Claude Code docs, permissions",
   "source_url": "https://code.claude.com/docs/en/permissions",
   "locator": "permission rules",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "76",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-76",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "Claude Code's sandbox limits which directories a command can write to",
   "quote": "\"restricts file system access to specific directories\"",
   "removal_reason": "",
   "source": "Claude Code docs, sandboxing",
   "source_url": "https://code.claude.com/docs/en/sandboxing",
   "locator": "How sandboxing works",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "77",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-77",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "Databricks: episodic memory (raw session traces, searchable) alone 22.3%→35.8%; derived memory 66.1%; both 68.4%; internal benchmark of enterprise tasks",
   "quote": "\"Episodic memory alone helps moderately: 35.8%, up from 22.3% with no memory\"; \"lifting pass rate to 66.1%\"; \"best result, at 68.4%\"",
   "removal_reason": "",
   "source": "Databricks",
   "source_url": "https://www.databricks.com/blog/memcurrent-designing-persistent-memory-production-agents",
   "locator": "What Makes MemCurrent Work",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "78",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-78",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "A later long-running agent would see progress and declare the job done",
   "quote": "\"look around, see that progress had been made, and declare the job done\"",
   "removal_reason": "",
   "source": "Anthropic long-running",
   "source_url": "https://www.anthropic.com/research/effective-harnesses-for-long-running-agents",
   "locator": "The long-running agent problem",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "79",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-79",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "ETH measured inference cost; Lulla measured runtime and output tokens (different efficiency measures)",
   "quote": "\"increasing inference cost by over 20% on average\"; \"lower median runtime\"; \"reduced output token consumption\"",
   "removal_reason": "",
   "source": "Gloaguen et al.; Lulla et al.",
   "source_url": "https://arxiv.org/abs/2602.11988",
   "locator": "Abstracts",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "80",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-80",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "Sourcegraph warns a badly-retrieving sub-agent still returns a confident summary",
   "quote": "\"a sub-agent that retrieved badly still returns a confident summary\"",
   "removal_reason": "",
   "source": "Sourcegraph",
   "source_url": "https://sourcegraph.com/blog/context-engineering",
   "locator": "Sub-agent architectures",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "81",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-81",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "An MCP server's list of tools can change after a client has read it (list-changed notification)",
   "quote": "\"When the list of available tools changes, servers that declared the listChanged capability SHOULD send a notification\"",
   "removal_reason": "",
   "source": "MCP specification 2025-06-18",
   "source_url": "https://modelcontextprotocol.io/specification/2025-06-18/server/tools",
   "locator": "List Changed Notification",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "82",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-82",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "Google distinguishes end-to-end benchmarks (outcomes) from behavioural evals (discrete actions)",
   "quote": "\"End-to-end benchmarks are the de facto for evaluating model performance\"; \"a behavioral eval measures discrete, observable actions\"",
   "removal_reason": "",
   "source": "Google Developers Blog",
   "source_url": "https://developers.googleblog.com/the-anatomy-of-harness-engineering-how-to-evaluate-iterate-and-guard-ai-coding-agents/",
   "locator": "intro; behavioural evals",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "83",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-83",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "Vitest accepts a failure message as expect's second argument (used in the code example)",
   "quote": "\"Vitest supports a message as the second argument\"",
   "removal_reason": "",
   "source": "Vitest docs, expect",
   "source_url": "https://vitest.dev/api/expect",
   "locator": "expect",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "84",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-84",
   "status": "VERIFIED",
   "status_label": "Verified",
   "removed": false,
   "claim": "ETH: developer-provided files did slightly better, 2.4 percentage points on average, not significant (p=21%), and significantly outperformed LLM-generated files (p=3.8%)",
   "quote": "\"Developer-provided context files improve agent performance by 2.4% on average (p=21%), significantly outperforming LLM-generated ones (p=3.8%)\"",
   "removal_reason": "",
   "source": "Gloaguen et al. (full paper, §4.2)",
   "source_url": "https://arxiv.org/html/2602.11988v3",
   "locator": "4.2 Main results",
   "date_read": "",
   "in_essay": ""
  },
  {
   "id": "85",
   "url": "https://durabilitycurve.com/claims/what-is-harness-engineering/#c-85",
   "status": "EXECUTED",
   "status_label": "Executed",
   "removed": false,
   "claim": "The guide's contract check fails on the renamed schema with the quoted message and passes on the original",
   "quote": "(run output)",
   "removal_reason": "",
   "source": "Our run, 22 Oct 2026, Vitest 5.0.5 + Zod + TypeScript 7.0.2",
   "source_url": "",
   "locator": "",
   "date_read": "",
   "in_essay": ""
  }
 ]
}