Accuracy Before Speed

Time savings fall out of the process itself: run an inherently efficient technique accurately and speed arrives without being chased — you would have to deliberately slow down to cancel the gain. Pushing the same process harder has almost nothing to work with. Familiar tasks already run at roughly 80–90% of the fastest pace your quality survives, you carry a reliable internal sense of where errors start to climb, and the claimable headroom is maybe 5–10% — claiming it raises the error rate. Output only counts once it clears a quality bar; motion that produces work you must later redo produces nothing.

That makes accuracy the entire pacing job. Execute the current process correctly, at a calm, brisk pace where focus holds, and put every genuine speed ambition into changing the process — the one lever with real range, since a better process can run several times faster than a worse one.

The arithmetic of one error

• Repair runs near 4x. A mistake must first be noticed, then the surrounding context rebuilt, the work redone, and everything assembled on top of it untangled. Total cost: roughly four times the time of doing it right once.
• Errors cascade. A wrong foundation corrupts what gets built on it: believe 1 + 1 = 3 and you will confidently derive 1 + 1 + 2 = 5, and each downstream misunderstanding adds its own repair bill.
• Foundation errors hide longest. A flaw in a fundamental skill stays invisible for weeks or months, then surfaces everywhere at once and resists tracing back. Repairing one mid-arc can cost four to eight months that two careful weeks at the start would have bought.
• Break-even is out of reach. A 10% speed gain that adds 10% errors costs about 40% more total time; breaking even would need a ~40% speed-up against 5–10% of available headroom. Even a perfect, error-free rush caps out at 5–10% saved — never worth the exposure.
• The long ledger shows no gain. Weeks of rushing deliver the same net output as weeks at a sustainable pace, plus stress and fatigue that degrade focus and feed the error rate.

Where real speed comes from

• Change the process. Switching to a fundamentally better technique can be worth 400%, a gain no amount of rushing reaches. The comparison only holds at a comfortable pace: rushing even an excellent technique drops its output below the same technique run calmly.
• Accuracy converts to efficiency on its own. Executed correctly, an efficient process generates fewer errors to review and repair, so its built-in advantage pays out in full. Execution speed then rises as a by-product of familiarity; the no-effort speed-up is the signal the process has habituated.
• Accurate work may feel slower at first. Per-session coverage drops while depth rises. Reaching the same depth the fast-shallow way costs two to three extra hours of relearning and re-testing later, so judge time efficiency across the whole arc of mastery rather than the single sitting.

The hustle trap

Raw effort competes with the reflection that aims it.

• Grind crowds out process-checking. Consumed by trying hard, you lose the mental space to ask whether the process points at the goal; the further down the tunnel, the harder it becomes to notice the time already sunk into the wrong thing.
• The fly at the window. Maximum effort against the closed pane, an open window a hand-span away: total hustle, zero progress. Effort that is genuinely aligned rarely needs to feel like hustle.
• Productive means goal-ward. Rest, a walk, time with people can be the productive move when the binding constraint is focus or emotional state; the same activity flips between productive and wasteful as your state shifts, so recalibrate priorities weekly — on rough days, daily.
• The grind test. A performance level reachable only through sustained grind is a verdict on the process. Take the short-term loss, free the hours and headroom, and build the process that reaches the level sustainably.

Carrying it into agent work

• Skill building. Practice a new technique for accuracy alone and let pace emerge from habituation; pushing for speed before the technique is correct automates the errors along with the motions, and relearning a mislearned skill runs about three times slower because every step must be checked against the old habit. The gates are asymmetric — moving too slowly is hard, moving too fast is easy — so when unsure, hold a level a week or two longer.
• Agentic engineering. Shipping unverified agent output trades minutes now for the 4x repair when the defect surfaces downstream, and the cascade is steeper because later commits build directly on the bad one. Throughput lives in the harness — sharper prompts, verification steps, reusable patterns — while babysitting more runs per day is the hustle trap in a new costume: it consumes exactly the reflection time that harness improvement needs.

Links into the system

The economics underneath Priority 0+1 System — prioritisation only pays when each unit of effort lands accurately. The no-effort speed-up is the habituation signal in Rapid Skill Acquisition; Pacing Skill Development applies the same load logic to how many processes to run at once; and process change as the one speed lever with range is the daily business of Marginal Gains.