Fixed-Gain Degraded Mode¶
Fallback mode — instantiates Adaptive Gain Retuning
When the adaptation signal can no longer be trusted or the stakes rise too high, it suspends retuning and pins the gain at a safe fixed value so the fast pathway keeps running predictably.
Fixed-Gain Degraded Mode is the archetype's off-switch for adaptation. When the inputs feeding the slow retuning loop become untrustworthy — data quality collapses, the system moves out of its known distribution, an adversary may be manipulating what the loop reads, or the stakes rise past what an automatic adjustment should decide — this mechanism stops adapting, pins the gain at a pre-agreed safe value, and lets the fast pathway keep operating on that fixed setting. Its defining premise is that a predictable gain is safer than a correct-looking one computed from signals you can no longer believe. It is not another way to tune the gain; it is the disciplined decision to quit tuning and hold, so a compromised adaptation loop can't drive the system somewhere worse.
Example¶
An automated market-making system continuously retunes how aggressively it quotes, scaling its sizing gain to recent volatility and liquidity. Then a flash dislocation hits: prices gap, the order book thins, and the volatility and liquidity signals the adaptive loop depends on go haywire — exactly the moment a gain computed from those signals would be most confidently wrong and most dangerous. Fixed-Gain Degraded Mode trips. It freezes the sizing gain at a conservative preset, stops the adaptive loop from reacting to the corrupted market signals, and holds the quoting pathway running at that safe fixed level while flagging for human attention.
The system keeps functioning — it still quotes, just cautiously and predictably — instead of either amplifying into the chaos or halting entirely. When conditions normalize and the signals become trustworthy again, adaptation is re-enabled deliberately, not automatically. The safety came from refusing to retune on bad data, not from retuning better.
How it works¶
Its logic is a guarded switch out of adaptation, not a smarter adjustment:
- Watch a trust condition, not the gain. The trigger is a judgement that the adaptation signal is no longer believable — data-quality collapse, out-of-distribution operation, suspected manipulation, or a stakes threshold — distinct from the ordinary in-range signals the loop uses.
- Freeze at a safe preset. On trip, it detaches the gain from the loop and pins it to a conservative fixed value chosen in advance to be survivable across a wide range of conditions.
- Keep the fast path alive. The pathway keeps running on the frozen gain, so the system degrades to predictable rather than off — availability without adaptation.
- Require a deliberate exit. Adaptation resumes only when the trust condition clears and re-entry is authorized, so the mode can't flicker on and off with the disturbance.
Tuning parameters¶
- Trip condition — what evidence of untrustworthiness triggers the freeze. Sensitive triggers protect aggressively but drop out of adaptation too often; lax ones leave the loop exposed to bad data.
- Safe preset value — the fixed gain held while degraded. Conservative presets are survivable but sacrifice performance; the whole bet is that this value is "wrong but safe" everywhere it might be needed.
- Exit / re-entry criterion — how much the trust condition must recover, and who authorizes resuming adaptation, before the loop is trusted again.
- Hysteresis on the trip — how much separation between trip and reset points, so the system doesn't oscillate between adaptive and degraded at the boundary.
When it helps, and when it misleads¶
Its strength is bounding the worst case. An adaptive loop is only as trustworthy as the signal it reads; when that signal fails, continuing to adapt is the dangerous choice, and a fixed safe gain converts an unbounded failure into a known, survivable one while keeping the pathway available — graceful degradation rather than collapse.[n1]
Its hazards are the mirror of its virtues. Trip too eagerly and the system spends its life in a numb, conservative mode, forfeiting the adaptation it was built for; trip too reluctantly and the fallback arrives after the damage. The subtler failure is the stuck degraded mode: the "temporary" freeze quietly becomes the permanent operating point because nobody owns re-entry, and the safe preset slowly drifts from safe to merely stale as the world moves on. The discipline is to make the exit criterion and its owner as explicit as the trip, and to review the safe preset on a cadence so the fallback stays genuinely safe rather than just familiar.
How it implements the components¶
This mode fills the fallback-and-continuity components of the archetype — the safe-state subset, not the adapting subset:
degraded_or_fixed_gain_fallback— it is this component: the defined safe state the system drops into when adaptation can't be trusted.gain_parameter— which it detaches from the loop and pins to a conservative fixed value.fast_forward_pathway— whose continued, predictable operation it preserves through the disturbance, trading adaptivity for availability.
It does no sensing or retuning of its own — the monitors and loops it overrides (Automatic Gain Control Loop, Adaptive Normalization Layer) hold those components — it sets no soft operating limits (Gain Floor/Ceiling Rule) and runs no validation of its own trip logic (High-Load Clipping Test, Weak-Signal Recovery Test).
Related¶
- Instantiates: Adaptive Gain Retuning — the safe fallback the whole loop reverts to when its adaptation signal fails.
- Consumes: Saturation Occupancy Dashboard — a health/quality signal that can trip the freeze; a trust condition on the adaptation inputs.
- Sibling mechanisms: Automatic Gain Control Loop · Adaptive Normalization Layer · Gain Floor/Ceiling Rule · Contextual Gain-Scheduling Table · Exposure or Alarm Sensitivity Adjuster · Weak-Signal Recovery Test · High-Load Clipping Test
Editorial Notes¶
Form Classification¶
Form family: Control, Automation & Runtime
Rationale: Fixed-Gain Degraded Mode operates as a live operational control that automatically routes, enforces, adapts, or responds during execution because it when the adaptation signal can no longer be trusted or the stakes rise too high, it suspends retuning and pins the gain at a safe fixed value so the fast pathway keeps running predictably.
Independent corroboration: The frozen evidence defines Fixed-Gain Degraded Mode as 'When the adaptation signal can no longer be trusted or the stakes rise too high, it suspends retuning and pins the gain at a safe fixed value so the fast pathway keeps running predictably', so its operative form is Control, Automation & Runtime.
Review outcome: Independent reviewer agreement; high confidence.
Origin Attribution¶
Primary origin: Engineering & Design
Origin pattern: Cross-disciplinary synthesis
Present-day reach: Multi-domain
Rationale: Control engineering is primary because fixed-gain fallback preserves bounded behavior when adaptation or estimation becomes unreliable. Systems theory and fault-tolerant software materially shape the transition and recovery logic; the generalized degraded-mode mechanism is encyclopedia synthesis with multi-domain use.
Related originating lineages:
- Computer Science & Software Engineering — Production systems implement fail-static modes when online-learning signals become unsafe.
- Systems Thinking & Cybernetics — Adaptive regulation and requisite stability supply the broader cybernetic rationale.
Review resolution: Control engineering is primary because fixed-gain fallback preserves bounded behavior when adaptation or estimation becomes unreliable. Systems theory and fault-tolerant software materially shape the transition and recovery logic; the generalized degraded-mode mechanism is encyclopedia synthesis with multi-domain use.
Encyclopedia synthesis: The exact catalogued form synthesizes established practice rather than reproducing a single standard historical label.
Review outcome: Researched adjudication after independent review; high confidence.
Sources consulted:
Notes¶
A floor/ceiling rule and this degraded mode are easy to conflate but differ in kind: a Gain Floor/Ceiling Rule bounds the range the loop may adapt within while adaptation continues, whereas Fixed-Gain Degraded Mode stops adaptation and holds a single value because the loop's inputs can no longer be believed. One clamps a running loop; the other switches it off.
[n1] Graceful degradation — the design principle that when part of a system fails, it should drop to a reduced but safe and predictable level of function rather than fail catastrophically (its engineering cousins are "fail-safe" and the automotive "limp-home mode"). Pinning a safe fixed gain when the adaptation signal fails is exactly this principle applied to a retuning loop. ↩