09 — Summary & Reset Protocol

“The roadmap serves you. You do not serve the roadmap.”

You’re Off Track. Now What?

This page is for the moment when you’ve stopped, slipped, stalled, or derailed — and you want to find your way back without having to rebuild the entire mental model of where you were and what to do next.

The protocol is four steps. It takes one focused hour to complete. Do it once, then act.

Step 1: Assess

Which failure mode is this? How far off am I?

Read the table. Match your situation to the closest failure mode. If it’s more than one, pick the primary one.

What’s happening

Failure Mode

Protocol File

Haven’t built anything in weeks, just watching content

Tutorial Hell

01

Using frameworks without understanding why

Math Phobia

02

One project that never ships

Scope Creep

03

Paralyzed by what others have built

Comparison Trap

04

Exhausted, dreading study sessions

Burnout

05

Job change or rejection derailed momentum

Job Pressure

06

Chased a trending topic, lost the thread

Hype Chasing

07

Life genuinely required you to step back

Life & Family

08

Measure the gap honestly:

Current sprint: ___
Sprint I should be on (based on start date): ___
Weeks behind: ___
Buffer sprints remaining: ___
Committed GitHub projects: ___
Last lab notebook entry: ___

Fill this out before proceeding. Numbers only. No narrative yet.

Step 2: Scope

What is salvageable? What must be cut?

Based on the gap measurement above:

Gap

Scope Decision

< 2 weeks behind

Absorb. Continue current sprint. No change.

2–4 weeks behind

Activate one buffer sprint. Consolidate, don’t advance.

4–8 weeks behind

Phase review. Identify which phase deliverables must be cut to maintain the core arc.

> 8 weeks behind

Minimum viable path (see below). Recalibrate the timeline.

Phase triage — if you’re 4+ weeks behind:

Not all phases are equal. Some are load-bearing for what comes after. Some can be compressed or combined.

Phase

Load-bearing?

Can compress?

Math prerequisites (M1)

✅ Yes

Partial — core derivations only

Classical ML (M2-3)

✅ Yes

Partial — implement 3 algorithms, not 8

Deep learning fundamentals (M4-6)

✅ Yes

No — this is the core

NLP + Transformers (M7-8)

✅ Yes

Compress to one phase: implement attention + fine-tune one model

Generative models (M9-10)

⚠️ Moderate

Can defer diffusion; keep VAE or GAN basics

LLMs + Alignment (M11)

⚠️ Moderate

Read + critique, reduce implementations

MLOps + Production (M12)

✅ Yes for portfolio

Deploy one model. One. That’s enough.

Portfolio + capstone (M13)

✅ Yes

Non-negotiable. Finish.

The rule: Deep learning fundamentals (M4-6) cannot be cut. Everything else can be compressed if life requires it.

Step 3: Recommit

One deliverable this week. Not the full plan. One thing.

Write it here:

This week's single deliverable: ___________________________

Definition of done: ___________________________

Time allocated (hours): ___

Committed to (accountability partner): ___

Do not try to catch up to where you “should” be this week. That attempt will fail and produce a second demoralization. Do one thing. Complete it. Commit it to GitHub. That act re-establishes the behavioral pattern. Momentum comes from completion, not from intensity.

The sprint plan resumes in full next week. This week is proof of life only.

Step 4: Track

Return to the lab notebook. Log the reset.

Write one entry:

Date: ___
Reset type: [pause / drift / burnout / life / other]
Gap: ___ weeks
Failure mode identified: ___
Scope decision: ___
This week's recommit: ___
How I'm feeling about it (honest): ___

The lab notebook is the record of the whole journey — including the hard parts. A reset entry is not evidence of failure. It is evidence of self-awareness and recovery. That is harder and rarer than uninterrupted progress, and it is more honest.

The Mermaid Decision Tree

The Minimum Viable 13-Month Goal

Life intervened. You got 70% of the way through. The capstone is unfinished. Some phases were compressed. You are not where you imagined you’d be.

Here is the honest truth: 70% completion of this roadmap still changes your trajectory.

The minimum viable version of this plan — the version that still has real value even if life compressed it — requires three things:

1. Solid deep learning fundamentals (non-negotiable) You can implement a neural network from scratch, explain the backward pass, train a language model, and debug a training loop without a framework crutch. This is the core. If you have this, you have something.

2. One end-to-end project, deployed and documented One project that works. That takes an input and produces a meaningful output. That has a README someone else could follow. That is on GitHub and (ideally) accessible via a URL. One. That’s the portfolio rung that matters most.

3. The ability to read a paper and evaluate it You can open a research paper in your domain, understand the method section, identify the claims being made, and form a reasoned opinion about whether the evidence supports them. This is the PhD-adjacent skill. If you have this, the credential doesn’t matter.

If you complete the roadmap at 70% and have those three things, you are a different practitioner than when you started. That is not consolation. That is the actual outcome.

The Closing

Here is what the plan expects of you: not perfection. Not uninterrupted forward motion. Not the absence of hard months.

It expects you to keep coming back.

Every time you return after a pause — every time you open the lab notebook after a difficult week, write one honest entry, and start the next sprint — you are doing the thing that most people who start ambitious plans do not do. You are continuing. Not heroically. Just steadily.

The frontier is not a destination you reach and then stop. It is a moving thing. You grow toward it, and it moves, and you grow again. That relationship doesn’t end at Month 13. Month 13 is just the point where you’ve built enough foundation to navigate it independently.

Get back to the work. The work is good. You are capable of it.