product-dev-os-OPEN/PROJECTRULESPROCESSES.md

Laws of Engineering

via https://lawsofsoftwareengineering.com/

Amdahl’s Law — The maximum performance improvement of a system is constrained by the portion that cannot be parallelized.

Beck’s Law — Optimism is an occupational hazard of programming, and feedback is the treatment.

Benford’s Law — In many real-world datasets, lower digits appear more frequently as leading digits.

Betteridge’s Law of Headlines — Any headline that ends in a question mark can usually be answered “no.”

Brooks’s Law — Adding manpower to a late software project makes it later.

Chesterton’s Fence — Do not remove a system or feature until you understand why it exists.

Conway’s Law — System design reflects the communication structure of the organization that built it.

Cunningham’s Law — The best way to get the right answer online is to post the wrong one.

Dijkstra’s Law — Program testing can show the presence of bugs, not their absence.

Eagleson’s Law — Any code you write will eventually be maintained by someone less experienced than you.

Gall’s Law — A working complex system evolves from a working simple system.

Gilb’s Law — Anything you need to quantify can be measured in a way that is superior to not measuring it at all.

Goodhart’s Law — When a measure becomes a target, it ceases to be a good measure.

Greenspun’s Tenth Rule — Any sufficiently complex program contains an ad hoc implementation of Lisp.

Hick’s Law — Decision time increases with the number and complexity of choices.

Hofstadter’s Law — It always takes longer than expected, even when accounting for this.

Hyrum’s Law — All observable behaviors of a system will be depended on by someone.

Ivory Tower Law — Theoretical designs often fail when applied to real-world constraints.

Kernighan’s Law — Debugging is harder than writing code, so overly clever code is risky.

Knuth’s Optimization Principle — Premature optimization is the root of most inefficiency.

KISS Principle — Simplicity should be a key design goal in systems. Law of Demeter — Components should only interact with their immediate neighbors.

Leaky Abstractions — All non-trivial abstractions eventually expose underlying complexities.

Lehman’s Laws — Software must continuously evolve or it becomes less useful over time.

Linus’s Law — Given enough reviewers, all bugs become easier to find.

Liskov Substitution Principle — Subtypes must be usable in place of their base types without altering correctness.

Murphy’s Law — Anything that can go wrong will go wrong. Occam’s Razor — The simplest explanation or solution is usually best.

Parkinson’s Law — Work expands to fill the available time.

Pareto Principle — A small number of causes often produce the majority of effects.

Peter Principle — People are promoted until they reach their level of incompetence.

Postel’s Law — Be strict in what you send and tolerant in what you accept.

Reed’s Law — The value of a network grows exponentially with the number of possible subgroups.

Schelling Point — People tend to coordinate around obvious focal points without communication.

Second-System Effect — Engineers overcomplicate systems when designing their second attempt.

Shirky Principle — Institutions try to preserve problems to maintain their relevance.

Tesler’s Law — Every application has an irreducible amount of inherent complexity.

The Law of the Instrument — If all you have is a hammer, everything looks like a nail.

The Unix Philosophy — Build simple, modular tools that do one thing well.

Wadler’s Law — Developers disproportionately focus on minor syntax issues over important architecture concerns.

Wirth’s Law — Software slows down more quickly than hardware speeds up.

Zawinski’s Law — Every program attempts to expand until it can read email.

Atwood’s Law — Any application that can be written in JavaScript will eventually be written in JavaScript.

Bell’s Law of Computer Classes — New computer classes emerge as technology advances and costs decrease.

Brandolini’s Law — Refuting misinformation takes significantly more effort than creating it.

Campbell’s Law — Metrics used for decision-making become corrupted and distort processes.

Clarke’s Third Law — Advanced technology is indistinguishable from magic.

Gallagher’s Law — Inefficiencies grow unnoticed until they become critical problems.

Gresham’s Law (applied) — Lower-quality practices tend to drive out higher-quality ones if unchecked.

Hanlon’s Razor — Do not attribute to malice what can be explained by incompetence.

Metcalfe’s Law — The value of a network grows proportionally to the square of its users.

Moore’s Law — The number of transistors on a chip doubles roughly every two years.

Ninety-Ninety Rule — The final 10% of a project takes another 90% of the time.

Sturgeon’s Law — Ninety percent of everything is of low quality.

The Law of Large Numbers (applied) — Outcomes become more predictable as the number of trials increases.

Zipf’s Law — In many systems, a few items are extremely common while most are rare.

Project Rules

Dump in rules that I come across - could be anything, specific or general.

1. Human is the architect (what + why), AI is the operator (how + execution)
2. In planning phase of projects, the goal is to implement solutions that scale, that fit well int he bigger picture, are easy to maintain, and respect good software engineering paradigms. AI agents are to think like a founder, not a part-time engineer. Model this expression: Prototype for 1X; Build for 10X; Engineer for 100X.
3. Break down projects and problems into many different sub-tasks that you can put a box around. Think components that are connected together once, not overly interconnected systems with functions consolidated into same script files. THink folders for components and putting them into discrete folders. Big sytems are daunting, but not if you break them down into discrete components that also means discrete tasks to work on them. THe bigger the component or task, the more context memory is needed, which creates problems. Respect and be good to AI agents by keeping context and memory as small as possible. See Architcture Philosophy below.
4. Components are evoling fast - a better one could come out tomorrow. Think about developing systems where you can swap out components and pieces with better ones that emerge.
5. Never mutate data without reviewing plan, understanding impact, vertifying results
6. Before deleting, backup, then review, then confirm
7. If working in windows workstation, handle windows + Linux paths for testing
8. Never modify production database schemas without migration scripts.
9. Update documentation when architecture changes.
10. Follow directory structure strictly.
11. Do not place business logic in API controllers.
12. All new features require tests.
13. AI wil work in isolated environments that give them safe, parallel workspaces. Never work on production environments.
14. Feed AI agents currated context that provides the right info without overwhelming them.
15. Repos might contain contradictory information. AI agents will flag that and ask before proceeding to get clarification.
16. Build components cheap so they can be thrown away when a better one comes along.
17. Skills: build using this evolving tempalte: C:\projects\productivity\skill-template

Dev. Vs Production

All projects worked on and tested in dev VM, then pushed to producton via Git.

1. Dev work

• build features
• test

2. Push to GitHub

• commit
• push

3. Deploy to prod

• git pull
• restart

4. Validate
5. Tag release

• v1.2.0

Or, create branch, spin up VM and work on it, test, open PR, merge to main.

Architecture Philosophy

Core Philosophy

Project is designed as a modular AI-native platform composed of small, isolated services with clearly defined responsibilities. The system prioritizes simplicity, replaceability, transparency, and operational control over excessive abstraction and orchestration complexity.

The architecture intentionally leans away from heavy containerization (Docker/Kubernetes) unless a specific application materially benefits from it.

The preferred model is:

• Proxmox VM-level isolation
• App-level isolation via folders, virtual environments, and systemd services
• API-driven communication between services
• Git-based deployment and version control
• Clear schema and contract boundaries
• Minimal inter-service coupling

Key Architectural Principles

1. Single Responsibility Repositories

Each repo should own one major responsibility only.

Examples:

• paddlenet-front-website
• paddlenet-pid-ingestion
• paddlenet-profile-store
• paddlenet-search
• paddlenet-match-engine
• paddlenet-agent-runtime
• paddlenet-notifications
• paddlenet-shared-schemas

This enables:

• easier maintenance
• independent upgrades
• lower coupling
• easier AI-assisted development
• easier future replacement/refactoring

2. API Contracts Over Shared Logic

Services communicate through APIs, not through shared internal code or direct database manipulation.

Think:

• controlled interfaces
• stable contracts
• modular boundaries

Not:

• tightly interconnected systems
• shared assumptions
• direct cross-service file manipulation

Internal API calls usually remain entirely inside the server or VM and do not traverse the public internet.

Examples:

http://localhost:8001/api/pid/submit
http://search-service:8002/api/search

3. Avoid Shared Runtime Chaos

Without Docker, runtime isolation becomes critically important.

Every application should have:

• its own folder
• its own virtual environment
• its own environment variables
• its own logs
• its own systemd service
• its own dependencies

Recommended structure:

/srv/apps/paddlenet-pid/
  .venv/
  .env
  logs/
  src/

Never share:

• Python virtual environments
• Node modules
• application configs
• runtime dependencies

between services.

4. One systemd Service Per App

Each service should run independently under systemd.

Examples:

paddlenet-front.service
paddlenet-pid.service
paddlenet-search.service
paddlenet-match.service

Benefits:

• easier restarts
• cleaner logs
• better monitoring
• fault isolation
• simpler deployments

Typical deployment flow:

git pull
systemctl restart paddlenet-pid

This is intentionally simpler than container rebuild workflows.

5. Shared Schema Repository

A dedicated schema repo should define canonical object structures.

Example:

paddlenet-shared-schemas/
  pid.schema.json
  profile.schema.json
  match.schema.json

These schemas become:

• the contract layer
• the validation layer
• the interoperability layer

All services should validate against these schemas.

This is especially important for AI-native systems and agent interoperability.

6. Folder Structure Strategy

Use a master workspace folder locally:

PaddleNet/

Containing multiple independent repos:

PaddleNet/
  paddlenet-front-website/
  paddlenet-pid-ingestion/
  paddlenet-search/

On servers:

/srv/apps/

Each app remains operationally isolated.

7. Reverse Proxy As Public Entry Point

Only the reverse proxy and public frontend should be internet-facing.

Recommended flow:

Internet
   ↓
Caddy / NGINX
   ↓
Frontend
   ↓
Internal APIs
   ↓
Backend services

Backend services should remain private/internal whenever possible.

8. Prefer Simplicity Over Premature Orchestration

Avoid introducing:

• Kubernetes
• service mesh
• distributed orchestration
• event buses
• Kafka
• excessive microservices

until scale genuinely requires them.

PaddleNet v1 should optimize for:

• development velocity
• transparency
• maintainability
• operational simplicity
• AI-assisted coding workflows

9. Docker Is Allowed, But Not Default

Docker is not banned.

Use Docker only when it meaningfully reduces complexity or is operationally required.

Examples:

• third-party apps
• complex dependency isolation
• vendor-supported deployments
• multi-runtime conflicts

Default approach:

• native services
• folders
• venvs
• systemd
• APIs

10. Long-Term Goal

Build a modular, AI-native, antifragile platform where:

• services are replaceable
• schemas are stable
• APIs are clean
• infrastructure remains understandable
• deployment remains simple
• AI agents can easily reason about the system

AI Processes

0. Prep: Cleans the working tree by analyzing any uncommitted work and doing the right thing with it (stash or commit). Also runs the entire current test suite and fixes any failures it encounters.
1. Picks a task from bugs first, or if bugs are complete, a feature that I've completed a spec for
2. Loads up the spec, and then analyzes it
3. Loads relevant docs, then looks at relevant code
4. Develops a testing plan (absolutely critical)
5. Writes extensive tests for this, then runs them, expecting failures
6. Develops an extensive plan of its own (I NEVER read this, I do not care)
7. Runs sub-agents as critical reviewers (review agents) based on 6 personas I've detailed in REVIEW_PERSONAS.md: Designer, Architect, Domain Expert, Code Expert, Performance Expert, Human Advocate. Each of these "owns" a portion of the docs, and reviews against their own documentation, including suggesting where their own docs need to be adapted.
8. Adapts plan based on review agent reviews, and loops to 7 until green light from all review agents
9. Implements the plan, including documentation adjustments (docs live in the same code base under Docs)
10. Runs type checking, linting, compiler, other static analysis tools such as bundle size reporter, as many things as possible, and of course the relevant tests themselves, and verifies that it works, iterating as it goes. Be as strict as possible with your type checking and linting system. I used to be anti strictness, but that was when I was a wetware dev. For agents, I want the most strictness possible.
11. Run the entire test suite to protect against regressions, fix any new issues
12. Runs the review agents again on the implementation diff, and loops back to step 10 until getting a green light from all review agents.
13. Add any encountered unrelated TODOs for human review that they've noticed along the way to the TODO doc
14. Wrap-up: write a CHANGELOG entry, commit with a detailed commit message meant for human context when reviewing the code. (More on commits later)
15. Loop back to the beginning (step 1), and select the next task or spec.
16. When completely done, write up a report for human review. Extremely concise. Details live in commit messages.

Only give an AI a few rounds to fix a problem because if not fixed, then repeated rounds offer diminishing returns of fixing, and more rounds cost more tokens without proportional improvements. Knowing when to stop is important

Human Processe After AI Processes

1. Review: Check the changelog and agent recap. Then go commit by commit, reviewing each commit message, implementation diff, tests, and docs changes.
2. Stacked commits: I keep all commits in the same branch so they stack on each other (a good use for stacked PRs, btw). Improvements carry forward to each subsequent commit. Fewer conflicts, better results, less duplicative work.
3. Quick fixes: If I need to correct something quickly, I do it using an interactive agent session or by hand. But before I do, I ALWAYS analyze and correct the docs, workflow, and validations/testing first.
4. Postmortem: If the agent misbehaved, don't just fix the code. Don't tell it to fix the code either. Use that valuable context to figure out why it did the wrong thing. Have it analyze its own context and tell you what docs, skills, or workflow led it astray, and what improvements would make it make the right decision next time. Have it fix those issues first — be diligent, because you can amortize the improvement over the rest of your project. Only after that, have it fix the original issue. Use that feedback cycle and continuous improvement to get to a point where it is making the right decisions more often than not.
5. Manual testing: I check almost every change manually and thoroughly. Not just to catch bugs, but to catch gaps in my docs, skills, specs, validations/tests, and my own understanding of the system. And fix them!
6. Spec writing: Then I get back to the first part — gathering requirements, writing specs, doing architecture, and thinking a lot.

Notes

Automated testing is incredibly important. This WILL NOT WORK if you don't have a super robust end-to-end testing harness in place and excellent docs so the agents can create their own tests.