Project Hong Kong

Motivation

A plan for yourself has no compiler and no test suite. Nothing checks the thing against its specification, so a document that looks like a plan is indistinguishable from one that is a plan, at the moment of writing and for as long as you like afterwards.

One signal separates them and it is available immediately: when I missed the deadlines I had set myself, I was not upset about it.

Indifference to a broken commitment is a cheap measurement. A load-bearing commitment produces friction when it breaks. You reschedule, you make it up, or you at least argue with yourself. Nothing at all means nothing rested on it, and a deadline nothing rests on is a date rather than a constraint.

The failure is easy to miss because writing the plan is the reward. It is daydreaming about an imaginary version of yourself who could execute the thing described.

The failure mechanism

It does not feel like self-deception while it is running. Talking about something as though it exists is, after enough repetition, nearly indistinguishable from the thing existing. Your sense of a described system becomes as vivid as your sense of things actually running, and the description does all the work the implementation was supposed to do, faster and more pleasantly.

So the design problem is: how do you stay tethered to reality when your own mind produces such convincing evidence that you are already there? Four answers, which I called reality-checks.

Objective measurement

• One: objective measurement written down on persistent media. A record made at the time by a process that does not care how you feel about it now. Persistent is the operative word: a number you remember has already been through one round of editing by the person who wants it higher.
• Two: something that reads the pile. Tracking on its own produces an archive rather than an instrument. Evaluation turns the log into a statement about progress, so a recording changes a decision instead of just existing.

The tracking half ran inside my daily journal, so it only worked on days I journalled and the instrument measured my journalling habit as much as the thing it pointed at. Either journalling becomes a first-class practice or tracking gets its own mechanism. Naming the dependency is most of the work, and drifting between the two resolutions is the only wrong answer.

Accountability and lenience

Three: accountability. Every commitment I fail to complete gets surfaced, then rescheduled or compensated with something else. Nothing quietly vanishes, because quietly vanishing is how a plan becomes fiction without anyone noticing.

Four looks like a contradiction of three, and the way they fit is the actual design.

The systems before this one were built for a person with unlimited willpower, and real days vary. They failed the same mechanical way every time:

1. consistent for a while
2. hit a day too depleted to continue
3. stop, and feel bad about the missed day
4. abandon the project once the debt is too large to pay down

Debt is the mechanism. Each missed day raises the cost of returning, because returning means facing the pile, until walking away is cheaper than paying it down. The collapse is a debt calculation that came out the wrong way, made inevitable several days earlier by a system with no way to make a small payment.

Degraded modes

Four: the system offers less-demanding forms of the original commitment on days when the full version is not going to happen.

The constraint that keeps this from turning into permission to do nothing is that the reduced version must keep a minimal element of the original. A smaller instance of the same act, not a substitute for it. The gym commitment is a ladder in discrete gradations:

full        60 min
            45 min
            30 min
            15 min
floor       walk through the entrance door and turn around

The floor is almost entirely symbolic and it is the most important rung, because it keeps the debt at zero on the worst day. A ladder with no symbolic floor still has a day it cannot reach, and that day is where the debt starts.

Both have to hold at once:

• accountability alone produces the debt collapse
• degraded modes alone produce a project that has quietly become optional

The time grammar

Different questions need different resolutions, so the units are a set rather than just days.

The spans carry reasons the table does not.

• Projects are sized in an exact number of days, so elapsed over total is a progress figure at a glance. Inside a day the unit is minutes rather than hours, for precision and because a minute figure reads as more significant than the same duration in fractions of an hour.
• A sprint is three days because the work has three phases, one per day. Day one is clarity: research, design, refining the problem statement, planning, prototyping. Day two is execution against that plan, deliberately avoiding refactoring and optimisation. Day three is execution through the first half, then delivery: packaging, tidying the interface, fixing bugs, documenting, and at least thirty minutes writing up what happened, what was hard, what was learned and what is left. Without that half hour a sprint produces work and no knowledge, and the next one starts from the same place.
• A week's order is fixed at the start, sprint-sprint-rest or sprint-rest-sprint, rather than discovered on Wednesday.
• Two weeks is the lag before results become visible. Extrapolated from one domain, gym training showing up about two weeks after the work, then used elsewhere as a rule of thumb. It is also the minimum before a habit reads as established.
• Months are for retrospective summary. Against a three-day sprint or a two-week interval they structure intent worse, which is a comparison drawn from using all three rather than a property derived from anything.
• Phases are three against a deadline and four for ongoing work. Three gives a beginning, a middle and an end; four fits work about improving something rather than arriving somewhere. Two weeks is the floor, so that something can happen inside a phase, and eight the ceiling, so that it does not become the whole project again.

The two-week lag has a consequence larger than the number. A system that reads its own effects sooner than the effects can appear will chase noise, then adjust, then chase the noise its adjustment made. The sampling interval has to be longer than the lag or the feedback loop is measuring itself.

The units nest, which is what makes it a grammar rather than a list of durations. A sprint is a beginning, a middle and an end. A week is two of those plus recovery.

Daily renewal

The last piece addresses a failure mode with no event anywhere in it. A project can stop without ever being stopped: attention moves elsewhere, nothing is decided, and because nothing was decided there is no point at which anything has to be admitted or recorded.

So the project gets renewed every day, out loud, with an explicit clause saying that if I no longer want to continue I will formally resign and terminate it, with reasons. The declaration ends on the line that does the work:

"I understand my decision to continue is only valid for today and my commitment must be again renewed tomorrow."

Expiry is what makes it a mechanism rather than a promise. A commitment with no expiry stays nominally true while nothing is happening, which is exactly the state being designed out.

What it buys is the removal of a state. Every day you are either in, which requires an act, or out, which also requires an act. There is no middle state where the project is technically alive and actually dead, and that middle state is where every previous attempt of mine had gone to die.

Adjacent work: Atomic Habits

One rule in all of this contradicts the source most of it comes from, and I kept both.

James Clear's Atomic Habits argues for gradual intensity and minimal mental exertion: consistency beats intensity, and anything requiring willpower is fragile. I am convinced by that argument and it is the main strategy underneath everything above. This whole document is a response to one line of it, that we do not rise to the level of our expectations, we fall to the level of our systems.

The exception is the start. A burst of effort well out of proportion to the plan, such as a very long first session or a document written in one sitting, functions as a signal that something has started. It works, and it is exactly what the book says not to do.

They answer different questions, so both hold:

• the burst is a tactic for crossing the line from not-started to started
• gradual consistency is the strategy for everything after

The failure mode is confusing the tactic for a requirement. Then a day without the capacity for a grand gesture becomes a day you cannot begin at all, and what was supposed to be momentum turns into another reason to postpone.