State — The Progress Ledger

The one file that spans all 130 weeks. RESUME.md lives inside each project and tells you where you stopped yesterday. This tells you where you are in the journey.

Update it at every project boundary and every stage review — roughly twenty times across the journey, not weekly. A ledger you must maintain constantly gets abandoned; one you touch every six weeks does not.

Everything below is the template with the starting state filled in. Overwrite as you go. The <!-- comments --> explain each field; delete them once the habit is formed.


Right Now

DATE            : not started
PHASE           : pre-calibration
CURRENT PROJECT : none
CURRENT WEEK    : 0 of 130
PACE            : undecided  (see README "The Honest Duration")
MODE            : —          (full 11 h/wk | maintenance 3 h/wk | zero, planned)

Next action: take the 3-hour calibration battery, cold, then run python3 tools/calibrate.py C1 C2 C3 C4 C5 and record the adjusted schedule below.


Calibration

TaskScoreAdjustment applied
C1 attention_/5
C2 nearest neighbour_/5
C3 systems arithmetic_/5
C4 ownership / Rust_/5
C5 measurement judgement_/5
Total_/25net _ weeks → _ active weeks

Re-takes (C3 / C5 only):

Stage reviewC3C5Note
M7
M15
M22
M26
M31

Projects

Gate = the eight-item completion gate. A project is done when the gate passes, not when the code works.

#ProjectWeeksStatusPlanned hActual hGateReportWeakest category
P01Transformer1–8not started88
P02ANN index9–15not started77
P11-ITree-walk interpreter16–20not started55
P13-IAutodiff core21–26not started66
P03Vector database27–34not started88
P04LSM engine35–43not started99
P11-IIBytecode VM + GC44–50not started77
P13-IIExecution optimisation51–54not started44
P05Distributed KV55–67not started143
P06MapReduce68–75not started88
P07Streaming76–83not started88
P08Recommender84–89not started66
P09Simulator90–95not started66
P10A/B platform96–99not started44
P12Kernel100–110not started121
P14Hardware-aware111–117not started77
P15Integrated118–130not started143
Total1300/171,4300

Status values: not started · in progress · complete · cut (reason) · abandoned (postmortem written).


Estimation Calibration

ProjectPlanned hActual hRatioWhat ran over, and why

Running multiplier:

If the multiplier exceeds 1.3 after three projects, your real pace is lower than planned. Re-derive the end date from the duration table and move it. Do not plan to catch up; nobody catches up.


Scorecard Trend

Recorded by tools/scorecard.py into scores/scorecard.json. Summarise here at each stage review.

StageWeakest categoryChosen focus for next stageDid it move?
1 (M7)
2 (M15)
3 (M22)
4 (M26)
5 (M31)
cd tools && python3 scorecard.py trend

Retention

WeekRebuild drillOutcomeNote
15Multi-head attention
26Beam search + distance counter
40Bloom filter
54SSTable reader
67Raft election logic
83Watermark tracker
99EMA profile + metric suite
117Blocked matmul

Outcomes: working in 3h (you own it) · bugs findable (mostly) · cannot start (you never owned it — schedule a re-read, add queue items).

Review queue: _ items · last run: —


External Feedback

StageAdversarial self-reviewPublishedReproduction exchangeReviewer conversation
1☐ W20
2☐ W34, W43
3☐ W67, W83
4☐ W99
5☐ W110, W117
6☐ W126, W130

Corrections received that changed a score:


Decisions and Cuts

DateWeekDecisionReasonCost

Blocked On

ItemSinceNeeded byNotes
Choose a pace (11 h/wk default)Week 1Duration table
Take the calibration batteryWeek 13 h 10 m, cold
Decide P12 target (RISC-V vs user-space)Week 100Re-decide at week 3 of P12
Register for MINDMonth 20Datasets
Find one reproduction partnerStage 1 reviewFeedback
Decide the P15 research questionWeek 118Six candidates
What is this ultimately for?Career, research, curiosity. Changes P15's question and several priorities

Parking List

  • (empty)

Log

[not started]  Track built. Calibration is the next action.