Strata

Validate your understanding. Learn at your level. Track your progress — with evidence, not vibes.

✨ Overview

Strata is a standards-grounded knowledge base for software engineering — from system design down to hardware — wrapped in an agent harness. Point any coding agent at this repo and it will validate your mental models against authoritative content, teach you at your measured level, and log your learning journey locally.

Every claim carries an evidence tier (T0 proof → T4 frontier) and a citation record; every topic ships a Bloom-tagged validation kit and a spaced-repetition review schedule; your journey data stays on your machine.

• Industry-grounded, not blog-grade. Content is mapped to SWEBOK v4.0, CS2023, ISO/IEC 25010:2023, ISO/IEC/IEEE 12207, and CMMI V3.0 — audited against them in the coverage report.
• Scientifically proven learning. Retrieval practice, distributed practice, and interleaving (Dunlosky et al. 2013; Cepeda et al. 2006) are baked into every pack.
• Honest provenance. A claim without a record is forbidden (CI-enforced); frontier content is dated, flagged volatile, and expires for re-review.
• Private by default. Profile, skill matrix, and session log live in a local .journey/ — gitignored, schema-validated, never uploaded.

🚀 Features

🎯 Claim Validation State a mental model in plain language; the agent returns a verdict (correct | partial | incorrect) with evidence tiers, S-#### citations, and a corrected model — from 11 copy-paste prompts.	🧠 Evidence-Tiered Knowledge 68 topics on 4 axes (band · track · tier · bloom) form a prerequisite graph; every claim tagged [T0..T4] + record, 200+ verified citations, CI-enforced claim↔record bijection.	📚 Spaced-Retrieval Learning Every topic ships a validation kit (formative/summative/review banks at Bloom levels) and a spaced ladder (1/3/7/14/30/60/120 days) with calibration tracking of predicted vs actual.
🤖 Agent-Agnostic Harness Works with Claude Code, opencode, Cursor, Codex, Gemini CLI and more — any agent that reads AGENTS.md. Progressive-disclosure maps keep agent context budgets flat.	🔍 Standards Coverage Maps standards/ maps the knowledge base against SWEBOK v4.0 (18 KAs), CS2023 (17 KAs), ISO 25010:2023, CMMI V3.0 and ISO 12207 — with a live coverage report as the roadmap.	🛡️ Local-First Journey .journey/ holds your profile, skill matrix, review queue, and event log — gitignored, schema-validated by tools/validate-journey.py. Nothing leaves your machine.

⚡ Quick Start

1. Clone and point your coding agent at the repo (as its working directory)

   git clone https://github.com/trac41799/strata-knowledge.git
   cd strata-knowledge

2. Paste a prompt from harness/prompts/ — e.g. validate a claim:

   I want to validate my understanding of systems-software/http-caching.
   My claim: "Cache-Control: no-cache means the response must not be stored."
   Follow AGENTS.md and give me a verdict with tiers, evidence records, and the
   corrected model. Schedule my spaced review and log the session (ask first).
   

3. Learn a topic — teach-topic.md · quiz-me.md · explain-back.md
4. Plan a curriculum — plan-curriculum.md (topological path from your skill matrix to any target topic)
5. Review your work — review-project.md against standards-mapped rubrics
6. Track it all — every session is logged to .journey/ per the committed schemas

🧩 Supported Agents

Agent	Memory file	Notes
Claude Code	CLAUDE.md	reads AGENTS.md first
opencode	.opencode/memory/	AGENTS.md-aware
Cursor	.cursor/rules	paste any prompt
Codex CLI	AGENTS.md	native
Gemini CLI	GEMINI.md	reads AGENTS.md first
Aider / Goose / Cline	CONVENTIONS.md / .goose/ / .clinerules	paste any prompt

🏗️ Architecture

knowledge/<track>/<topic>/   topic packs: concept.md (tagged claims),
                             validation.md (Bloom-tagged items),
                             teaching.md (examples, misconceptions)
evidence/records/            one file per cited source (S-####)
standards/                   SWEBOK / CS2023 / ISO / CMMI coverage maps
journey/                     committed conventions: schemas, templates, privacy
.journey/                    YOUR local learning data (gitignored, private)
harness/                     AGENTS.md contract + 11 copy-paste prompts
rubrics/                     standards-mapped project-review rubrics
tools/                       stdlib-only validation pipeline (lint, graph, coverage)
docs/                        spec, plan, ADRs, design system, UI spec, reviews
ui/                          journey-interface design system + explorations

Four layers: Knowledge (facts with provenance) → Validation (proving understanding) → Journey (private progress) → Harness (how any agent behaves). Topics form a DAG; learning paths and build waves follow its topological order — knowledge-graph.yml · INDEX.md.

💻 Development

• Docs — full specification · Plan — roadmap · PRINCIPLES.md — axioms (K0–K6, T1)
• Gates (stdlib-only, CI-enforced):

  python tools/lint.py && python tools/check-graph.py && python tools/index.py && python tools/coverage.py

• Contribution flow: see CONTRIBUTING.md

🤝 Contributing

Contributions welcome — this is a knowledge commons. PR flow: draft → CI → human review, per CONTRIBUTING.md. The coverage report is your roadmap. Governed by the Code of Conduct.

🗺️ Roadmap

• 56/68 topics published across 12 tracks; wave 4 = remaining 12 scaffolds
• Journey Interface implementation (design ready: design system, UI spec, TDD plan)
• Agent-context MCP server (optional machinery)

📄 License

• Knowledge, docs, and content: CC BY 4.0 — share and adapt with attribution.
• Tooling and code: MIT.

Copyright (c) 2026 Strata contributors.