Gain-Collapse Test¶

Diagnostic test — instantiates Stock-First Control Restoration

Confirms the lever isn't merely too weak — its marginal effect has collapsed toward zero because the mediating stock has left the range where the lever bites.

When a control stops delivering, the reflex is to pull harder. Gain-Collapse Test is the diagnostic that tells you whether pulling harder could possibly help. It measures the marginal response — the change in outcome per unit change in the lever, the loop's "gain" — rather than the absolute level of the outcome, and asks a single discriminating question: is the lever weak, or is it dead? A weak lever still moves the outcome a little when you push it; a dead one doesn't move it at all, because the stock it acts through has slid outside its responsive window. That distinction is the whole hinge of the archetype: it separates "push harder" from "stop and repair the stock," and it is the only clean way to tell them apart from the outside.

Example¶

A house won't cool. The thermostat already reads 68°F and the occupant keeps nudging it lower — 66, 64 — with no effect. Rather than keep dropping the setpoint (the lever), a Gain-Collapse Test perturbs it by a known step and reads the slope: drop the setpoint 2°F and measure the room's actual response over the next hour against the unit's historical curve. Normally a 2°F setpoint cut buys ≈1.5°F of room cooling in that window; today it buys ≈0.1°F. The gain has collapsed. That number reframes everything — the fault is not "the setpoint isn't low enough," it's that the refrigerant charge (the mediating stock) has fallen below the level at which the compressor can move heat at all. The system has left its responsive window, and the setpoint lever has nothing to act through. The test's verdict: stop chasing the thermostat, recharge the stock.

How it works¶

What sets it apart from just "notice it's not working":

Perturb by a known increment, read the slope. It changes the lever a measured amount and records Δoutcome / Δlever, not the outcome level. A cold-enough reading can coexist with a gain already near zero.
Compare against a responsiveness baseline. The measured slope is judged against how responsive the lever should be — recent history, a healthy peer unit, or a physical model.
Localize a near-zero slope to the stock. When gain has collapsed, it attributes the cause to the stock's position relative to its responsive window, so the finding names what to repair rather than just "it's broken."

The tell it hunts for is counter-intuitive: response that flattens or falls as the lever rises — the opposite fingerprint from a merely underpowered lever, which still yields a little more when pushed a little harder.

Tuning parameters¶

Probe step size — how large a lever change to test with. Bigger steps give a cleaner signal but can shove an already-stressed stock further out of range; smaller steps are safer but hide in the noise.
Baseline reference — what counts as "normal gain" for the comparison. A stale baseline makes a slow drift look like a sudden cliff.
Collapse threshold — how low the slope must fall to call it collapsed rather than merely degraded. Tight and you cry wolf; loose and you keep escalating a dead lever.
Attribution depth — whether the test only flags low gain or goes on to pin it to a named stock and window. Deeper attribution costs instrumentation but tells you where to aim the repair.

When it helps, and when it misleads¶

Its strength is that it kills the single most expensive error in this whole family — pouring more force into a lever that has stopped transmitting — by converting "it's not working, push harder" into a measured verdict about why. It is the gate that authorizes the entire stop-repair-resume sequence, so a system runs it before it spends anything on either escalation or repair.

Its failure modes are symmetric. A too-small probe or a noisy baseline can read a live-but-slow lever as collapsed, so you stop a lever that would have worked; and lag between lever and outcome can mask a real collapse if you read the slope too soon. The classic misuse is running it backwards — reaching for "the lever is dead, nothing could be done" to excuse a decision to stop trying, when the lever was only underpowered. The discipline that guards against both is to fix the baseline curve and the collapse threshold before probing, and to confirm the diagnosis with an independent read of the stock's state rather than inferring it from the flat response alone. The behavior it detects is control-systems saturation — an element run out of its range produces no further output for more input, so its incremental gain goes to zero.^[1]

How it implements the components¶

Gain-Collapse Test fills the detection side of the archetype — the components a test can produce, not the ones that act:

normal_flow_lever — it operates on the lever directly, perturbing it by a known step to read its live marginal effect.
responsive_window_model — it scores the measured gain against the window where the lever is supposed to bite and flags a departure from it.
control_mediating_stock — a near-zero slope is attributed to the stock's position, localizing the fault to the mediating stock rather than the lever.

It does not measure or repair the stock's level (stock_state_observability, stock_repair_plan — [Minimum Stock-Floor Alert] and [Replenishment-or-Refill Protocol]) nor halt the escalation it exposes (lever_escalation_stop_rule — Lever Freeze-or-Cap); a test only diagnoses.

Instantiates: Stock-First Control Restoration — Gain-Collapse Test supplies the discriminating diagnosis the rest of the sequence hangs on.
Sibling mechanisms: Lever Freeze-or-Cap · Stock-Flow Diagnostic Map · Fallback Control Mode · Deleverage-Before-Stimulus Rule · Low-Amplitude Reactivation Probe · Hysteresis-Aware Exit Criterion · Minimum Stock-Floor Alert · Replenishment-or-Refill Protocol · Substrate Repair Protocol · Staged Lever Ramp · Trust-Repair-Before-Persuasion Rule

Notes¶

The test reads a slope, which is why it can fire while the outcome level still looks fine — a stock can glide toward collapse with the room still cool and the alarms still green. That early-warning property is what makes it distinct from the Stock-Flow Diagnostic Map, which maps the structure statically, and from the Minimum Stock-Floor Alert, which watches the level: only the gain test catches the moment the lever quietly stops mattering.

References¶

[1] Saturation — a control element that has reached the limit of its range produces no additional output for additional input, so its incremental gain falls to zero. A saturated (here, stock-starved) lever presents as "too weak" but is really out of range; the remedy is to restore range, not to command harder. ↩