KPI Dashboard¶

What You Actually Measure — 7 Weekly, 4 Monthly, 3 Quarterly¶

The hardest discipline problem in self-directed learning is measuring the right things. The wrong metrics — hours logged, videos watched, notes taken — create the illusion of progress while hiding the absence of it. The metrics below measure capability, not activity. If your numbers look good but you can’t implement the week’s material from memory, the numbers are lying.

Anti-KPIs are listed at the bottom. These are the things you will be tempted to track because they feel like progress. They are not.

7 Weekly Metrics¶

Track these every Sunday in your lab notebook. 5 minutes maximum.

#	Metric	How to Measure	Target
W1	Primary deep work blocks completed	Count (0, 1, or 2)	2/2
W2	Implementation tests passing	Did the week’s code exercise run and produce correct output?	Yes
W3	Lab notebook entries written	Count (target 2/week)	2
W4	Concepts I can explain without notes	Pick 3 concepts from the week. Close the laptop. Explain each in 60 seconds. Count how many you complete.	2/3
W5	GitHub commits	Count (any push to a project repo)	≥3
W6	Prediction log: did I review last week’s prediction?	Yes/No	Yes
W7	Energy rating (1-5)	Subjective: how sustainable did this week feel?	≥3

Weekly retro template (Sunday, 10 minutes):

## Sprint Retro — Week [N], [Date]

W1 Deep work blocks: [0/1/2]
W2 Implementation working: [Yes/Partial/No]
W3 Lab entries: [N]
W4 Can explain without notes: [N/3]
W5 GitHub commits: [N]
W6 Prediction reviewed: [Yes/No]
W7 Energy (1-5): [N]

One thing that worked:
One thing to change next week:
Next week's 3 implementation tasks:
  1.
  2.
  3.

4 Monthly Metrics¶

Check these at the end of each month. 20 minutes.

#	Metric	How to Measure	Target
M1	Phase file completion rate	Files with all exercises complete / total files in current phase	≥80%
M2	Portfolio rung status	Is the current rung on track? (Green/Yellow/Red)	Green
M3	Explain-to-a-stranger test	Pick the hardest concept from the month. Explain it to someone non-technical (in person or written). Did they understand?	Yes
M4	Sprint slippage	Weeks marked Yellow or Red in the past 4 weeks	≤1

Monthly review template:

## Monthly Review — Month [N], [Date]

M1 Phase file completion: [N/N] files
M2 Rung status: [Rung N — Green/Yellow/Red]
M3 Explain test: [Concept explained] — [understood/not understood]
M4 Slippage weeks: [N]

Am I on track for the month's exit check? [Yes/No]
If no — what specifically is incomplete:
Recovery plan for next month:
Zoho alignment note:

3 Quarterly Metrics¶

These are the high-signal checkpoints. They measure whether the roadmap is actually working.

#	Metric	Measurement	Target
Q1	Portfolio rungs shipped	Count of rungs that are public and complete	Per schedule (see table below)
Q2	ML system design fluency	Can you answer “design an ML system for X” for 3 problems without notes?	3/3 by Q3
Q3	Technical writing output	Total published posts	3 by Q4

Rung tracking table (update quarterly):

Rung	Target Month	Status
1 — Math From Scratch	M2	⬜ Not started
2 — Classical ML Battle	M4	⬜ Not started
3 — Deep Learning From Scratch	M6	⬜ Not started
4 — Transformer Lab	M8	⬜ Not started
5 — LLM Engineering ⛔	M10	⬜ Not started
6 — Production ML System ⛔	M12	⬜ Not started
7 — Paper Reproduction	M12	⬜ Not started
8 — Technical Writing	M13	⬜ Not started

Update status to: ⬜ Not started → 🔄 In progress → ✅ Complete → 🔗 [URL]

⛔ The M9 Hard Gate Checkpoint¶

At the end of Month 9, before entering the hard gate sprints of Month 10, run this audit. If you fail more than 2 items, do not proceed. Use the reset protocol.

## M9 Hard Gate Checkpoint — [Date]

FOUNDATIONAL CAPABILITY:
[ ] Can implement backpropagation from scratch without looking it up
[ ] Can train a transformer from scratch on a small corpus
[ ] Can explain attention mechanism with correct tensor shapes
[ ] Rung 1, 2, 3, 4 all shipped and public

READINESS FOR HARD GATE:
[ ] QLoRA fine-tune has been attempted (even if rough)
[ ] A specific target task for Rung 5 fine-tune is defined
[ ] Compute budget is confirmed (GPU access confirmed, cost calculated)
[ ] RAGAS library is installed and you have run it on at least one toy example

HEALTH CHECK:
[ ] Energy rating average for last 4 weeks is ≥3
[ ] Zero consecutive red weeks in past 2 months
[ ] No major life disruptions expected in next 6 weeks

VERDICT: [GO / NO-GO]
If NO-GO — specific items to fix before retry:

Anti-KPIs (What NOT to Track)¶

These feel like progress. They are not. Every minute spent on these is a minute not spent on implementation.

Anti-KPI	Why It’s Dangerous
Hours logged (total)	Activity metric. Measures seat time, not learning.
Videos watched	Passive consumption. No correlation with capability.
Notes pages filled	More notes ≠ more understanding. Compress your lab notebook, not expand it.
Courses started	Starting is not progress. Completing with exercises is progress.
Twitter/X followers	Vanity metric. Irrelevant until Month 11 at earliest.
Kaggle leaderboard rank	Optimizing for rank teaches overfitting tactics, not ML understanding.
Books purchased	Books on a shelf are not books in your head.

The correct loop is: implement something → it fails → understand why → fix it → verify → write it in your lab notebook. Everything that is not this loop is cost, not investment.

Yearly Review Template (Month 13)¶

## Year-End Review — Month 13, [Date]

PORTFOLIO AUDIT:
[ ] All 8 rungs public and accessible
[ ] Hard Gate 1 (Rung 5) — [URL]:
[ ] Hard Gate 2 (Rung 6) — [URL]:

CAPABILITY AUDIT:
[ ] Can explain gradient descent from first principles (3 min, no notes)
[ ] Can implement attention mechanism from scratch
[ ] Can design a RAG system with evaluation for a novel domain
[ ] Can write a monitoring setup for a production ML model
[ ] Can read a new ML paper and explain the core claim in 5 minutes

PITCH READINESS:
[ ] 30-second pitch rehearsed and smooth
[ ] Portfolio URL memorized
[ ] 3 specific examples ready for "tell me about a project you shipped"

WHAT WAS HARDER THAN EXPECTED:
WHAT WAS EASIER THAN EXPECTED:
WHAT WOULD I DO DIFFERENTLY:
WHERE I AM GOING NEXT:

Return to: README.md · Next: 06_background_alignment.md