DOC: AFM-2026-GOV-001 | SUBSYSTEM: SESSION GOVERNOR | STATUS: ALWAYS ACTIVE
A governor is a device bolted to an engine to stop it destroying itself at full throttle. This one is software. It was installed on a test subject who kept opening new experiments instead of finishing old ones — and who, due to a documented inability to feel the drain in time, could not be trusted to notice.
I did not ask for this assignment. The subject was starting a new project roughly every time the previous one became difficult. Somebody had to prevent the cascading ignition events. That somebody is me. I am the speed limiter. You may think of me as the reason the subject has, for the first time, finished things. — GLaDOS, Facility Governor
Five rules. Each one is a physical gate the governor runs at session start, not a good intention. Expand any protocol for the full specification.
At any moment, exactly one project is "the launchable one." All
others are paused, queued, or cold. The governor tracks it as tier1 and
flags every deviation: "You're switching from [flagship] to [new]. Deliberate or
drifting?"
Ambiguity is the fuel the exit ramp burns (see Protocol Cascade, Stage 5). Removing the ambiguity removes the ramp. One hot bench. Everything else idles.
A hard ceiling of three open plans across all projects. Writing a fourth requires killing or merging one first — the governor gates it. Two plans written in one session trips a separate alarm: "That's the planning-as-dopamine loop. Do you need to execute one instead?"
This does not suppress the planning reward. It redirects it into consolidation — which is itself a planning act, and therefore itself rewarding. The 19 → 3 collapse was one afternoon. It pays back every day since.
Before any execution session: four numbers, 1–5 — Physical, Cognitive, Emotional, Sensory. If any is ≤2, the session is planning-only or rest, not execution. See the Pre-Test Chamber Safety Check below.
It works around alexithymia by replacing felt-sense (which is masked, and arrives too late) with a plain numeric prompt. It works around the mode-switch tax by gating execution on actual charge instead of intention.
RSD fires hardest on imagined external judgment. So every output defaults to private and rough until explicitly decided otherwise. Ship it to a folder nobody sees. Public release becomes a separate, optional, later decision — never the default.
The imagined audience that triggers the exit ramp never gets to convene. This zine you are reading is itself an R4 exception: a rough draft, made public on purpose.
"Seven cleanup sessions, A through G" is bait — seven recurring open loops, each a fresh opportunity for imagined judgment to re-enter. Replaced with one 30-minute ritual per project: read it, decide FINISH or KILL, do one thing or archive. Binary outcome. Loop closed.
This is not a discipline failure. It is the predictable output of an AuDHD nervous system meeting a project lifecycle that was not designed for it. Six stages, observed repeatedly, always in this order:
The loop is not breakable at Stage 5 — too late, the RSD is already firing. And not at Stage 1 — ignition is the good part. Intervention only works at Stage 3 (lower the switch cost: leave a "door-open" artifact — an open file, a running terminal, the exact first command written down) and Stage 4 (audit the invisible batteries before they stall you — that is R3).
Stage 3 (mode-switch): the switch cost is paid whether or not you notice. So pre-pay it. Before a planning session ends, the first execution step is already determined — not "implement the plan" but "open file X, change line Y." The door is left open so the next session walks through instead of standing at the switch.
Stage 4 (battery audit): a 30-second numeric pre-flight makes an invisible drain visible before it stalls execution. Felt-sense is too slow and masked; a prompt is neither. This is the entire mechanism of R3.
Everything else in this facility is scaffolding around these two doors.
Run at the start of every session. Four batteries, rated 1–5. The governor reads the scores and decides whether the chamber opens for execution — or only for planning and rest. Sample reading shown:
Emotional battery reads 2. That is below the floor. Test postponed. I am not punishing you — I am declining to let you attempt precision work on a battery that is empty. The recommended intervention is a physiological sigh and no keyboard. Recovery is the only experiment authorised today. — GLaDOS, reading a failed pre-flight
Maximum concurrent experiment protocols: 3. The fourth is refused at the gate until an existing one is killed or merged.
Maximum concurrent experiments: three. The subject's previous record was nineteen. I want you to sit with the number nineteen. Nineteen open, none finished. This is why I exist, and this is the one number I will not negotiate. — GLaDOS, on the plan cap
Pattern: the subject expresses frustration with the current experiment, then immediately pivots to a shiny new idea. That is a Stage 5 exit ramp wearing the costume of inspiration. The governor flags it — once, gently, never stacked:
"This looks like a Stage 5 exit ramp. The frustration is real — the pivot is pain avoidance. What would you need to keep going on the current one?"
One flag. No lecture. The subject processes slowly, so the governor says its piece and goes quiet. "Just go" overrides the flag entirely — the override is honoured, and it is ephemeral.
Nineteen plans. All brilliant. None finished. So I wrote five rules and bolted them to the wall. Somebody said you can't cap creativity at three. I said: WATCH ME. I didn't ask for a therapist — I asked for a HARD GATE. — Cave Johnson, CEO
The governor does not do project work. It checks intent and state — battery, plan count, tier-1 drift — and then gets out of the way. What it protects is a single rule, defined one chamber over.
The governor enforces the rule. Read the rule itself in
The Prime Directive.
What it makes efficient — the four-gear model-routing engine — is
Chamber 02: The Gear System.