Loops, Not Streaks

The question

Projects plan. Practices drift.

Your setup at work is already the hard part done: a knowledge base where everything lands, a long-running planning session, and the discovery that cadence can slip without loss because everything is written down and "super-super-recoverable." What felt less tractable was the other kind of goal:

Everything below is how that actually runs in Joe's system, then what you'd build first given what you already have.

The system in one move

Capture → Abstract → Resurface

Joe's system is Claude Code pointed at an Obsidian vault. The vault is long-term memory (the AI doesn't remember between sessions, so anything that matters has to land in a note). A CLAUDE.md operating manual, read at the start of every session, tells the AI who he is, how the vault is organized, and how he likes to work, so no session starts with re-briefing. A separate cross-session memory layer persists smaller facts.

On that substrate, one move repeats everywhere: turn raw experience into durable context, then bring that context back at the moment it's useful. Done continuously, every session starts smarter than the last.

flowchart LR
    CAP["Capture
raw experience
(a meeting, a meal, a thought, a doc)"]
    ABS["Abstract
distill into durable context
(notes, principles, summaries)"]
    RES["Resurface
brought back at the right moment
(kickoff, prep, next session)"]
    ACT["Act smarter
this round starts ahead"]
    CAP --> ABS --> RES --> ACT
    ACT -.->|context compounds| CAP
    

One rule was hard-won from auditing every loop end to end: resurfacing isn't enough. A loop only compounds when something forces the last step. That means a named consumer that actually reads what was captured, and a rule that turns a recurring red signal into a decision instead of wallpaper. Capture nothing that nothing reads, and never let "noticed" substitute for "acted." This rule matters more for practice goals than anything else on this page.

The architecture

One system for work and life, deliberately

Joe doesn't separate work goals from personal ones. The loops for his marriage, his kids, and his health get the same rigor as the ones for the company: real questions, honest tracking, follow-through. Each domain runs its own version of the core loop, and a daily/weekly rhythm is the heartbeat that ties them together. It's where every loop resurfaces, and where compounding actually happens.

flowchart TB
    subgraph SUB["Foundation (the substrate)"]
        F["Vault = memory · Claude = thinking partner · CLAUDE.md = operating manual · cross-session memory"]
    end
    RHYTHM(("Daily / Weekly Rhythm
the heartbeat"))
    FAM["Family / Relationships loop"]
    COACH["Coaching loop"]
    MEET["Meeting loop"]
    KNOW["Knowledge / Strategy loop"]
    HEALTH["Health loop"]
    CONTENT["Content loop"]
    BUILD["Build Mastery loop"]
    FAM --> RHYTHM
    COACH --> RHYTHM
    MEET --> RHYTHM
    KNOW --> RHYTHM
    HEALTH --> RHYTHM
    CONTENT --> RHYTHM
    BUILD --> RHYTHM
    RHYTHM --> COMPOUND(["Compounding
1% better every week"])
    SYS["System-building loop
the system improves itself"]
    SUB --- RHYTHM
    SYS -.->|sharpens every loop| SUB
    

The loops, briefly

For each: the cycle it runs, and what it compounds.

Your question, specifically

Four mechanisms that make practice goals tractable

The personal side of Joe's system rests on four mechanisms. Each is mapped to your three goals.

Two smaller things that help

• Practices as experiments, not streaks. Joe's coaching loop treats practices as experiments with check-ins rather than streak counters. That fits "taiji 3x/week" better than a habit tracker does, because the interesting question isn't the streak; it's "is this practice doing what I wanted, and if not, what's the actual blocker."
• An encroachment tripwire. The family loop includes an explicit flag for when work is eating the space the family practice needs, because practice goals mostly lose to encroachment, not to laziness.