@seanwy / ACC field notes

43 cards. Adding it makes your own copy. The author's later changes won't affect yours, and your progress stays private.

• I keep the learning loop open at every scale — the value is on the process staying revisable, not on any state it reaches.assertion
• When execution is abundant, I spend my scarce effort on aiming better, not doing more — doing more is digging the hole faster; doing better is climbing the tree.assertion
• I command from inside the system I am commanding — so I work on improving my target-forming process instead of declaring the target and defending it.assertion
• I hold the consent floor: friction someone elects grows a learner; suffering imposed by a self-appointed planner is the pedestal — crueler.assertion
• I measure my transmission by divergence: if everyone who learns from me ends up agreeing with me, the practice has failed by its own measure.assertion
• I keep an enforced cadence of stopping to re-aim — intent formation cannot be delegated, made asynchronous, or measured by throughput.assertion
• Look at this week honestly: how much of it was digging the hole faster, and how much was climbing the tree to check it is the right hole?reflection
• Name the last person you taught something real. Are they building on their own foundation, or reproducing yours?reflection
• When did you last stop and re-aim deliberately, on a cadence — rather than because something broke?reflection
• The ACC stack has four layers. Name them in order of how slowly they change — and say where the training happens.qa
• What is the atom of the whole ACC practice?qa
• State Ashby's law of requisite variety — and the bet it becomes in AI-assisted work.qa
• First-order versus second-order cybernetics — what is the difference, in ACC terms?qa
• What is an eigenbehaviour, and why is it the trainable unit of ACC?qa
• The two axes: what runs on the horizontal, what runs on the vertical — and which one is becoming the bottleneck?qa
• Why can the ACC axiom not be proven from inside the practice?qa
• Where does the actor end, in ACC?qa
• If agents keep returning with questions — or a hand-off spawns a clarifying thread — then my intent compression failed: recompress the outcome statement until it can be acted on without me present.if-then
• Intent compression: what is the measure, and what does a bad reading mean?qa
• You are about to park a task with an agent overnight. The brief reads: improve the onboarding flow — you know what I mean. What happens at 3am, and what should the brief have said?scenario
• If I am refusing to park a piece of work because getting back into it costs too much, then context preservation has already failed — the trail, not my memory, is what makes leaving cheap.if-then
• Context preservation: what is the measure, and what does the same behaviour look like across a whole arc of work?qa
• It is Friday, the work is mid-flight, and Monday-you will open it cold. What do you leave behind?scenario
• If I am riding a system — myself included — toward its limit, then I reset before the wall: the recovery tax always costs more than the early reset.if-then
• Capacity and degradation reading: what is the measure, and why do burnout and the agent context wall count as the same failure?qa
• If parallel workstreams keep colliding — or force me back into serial, synchronous execution — then my decomposition was secretly coupled: re-cut the parcels until the interfaces are clean.if-then
• Decomposition with clean interfaces: what is the measure?qa
• Three agents, one feature — and by evening two of them have rewritten the same file to different ends. Where was the failure, and when?scenario
• If I am interrupting the systems I command so often that supervision has crept back — or parking tasks and never returning — then my attention-scheduling is broken: asynchrony has decayed into synchrony, or into neglect.if-then
• Signalling and attention-scheduling spans two orders and carries two measures. Name both measures, and the two orders.qa
• A background job has run for an hour and you have checked it six times — fine every time. What are the checks costing, and what replaces them?scenario
• If I am re-verifying work from a system that has repeatedly earned trust on that task type — or shipping unverified output from one that has not — then my trust model is stale: set verification depth to the reliability actually demonstrated, per task type.if-then
• Trust calibration is the hard one. What is its measure — and what is the signature of getting it right?qa
• An agent that has been reliable for weeks just shipped confident nonsense in a domain you never check. What actually failed?scenario
• If I keep making the same command correction — same misfire, new week — then I am correcting the system but not my correcting: eliminate the recurring error at the source.if-then
• What is correction half-life — and what do a flat one and a shrinking one each mean?qa
• If I optimised a step and the whole system did not get faster, then the step was never binding: find the constraint that, relieved, speeds everything — and act there.if-then
• Bottleneck reading: what is the measure, and why must it be tracked rather than felt?qa
• Reliable command is intent reaching outcome through systems acting unwatched, at a rate and correctness you can predict. What is your honest current rate — and would you have known it without being asked?reflection
• For each system you command: do you know, cheaply and without thinking, how much to check it on which task? Where is that sense still expensive?reflection
• Is your correction half-life shrinking? Name one command error you have eliminated at the source — and one that still recurs.reflection
• How long can you leave a piece of work and return without paying the tax? What would double that horizon?reflection
• Is your identity attached to an occupation, or to the ability to re-aim? What would you still be if the role dissolved this year?reflection