Recovery Is Not Immunity

The measured result the model kept insisting on: a full rescue buys the recovered state nothing against the next trigger

Fourth in the mechanism series. The mechanism post describes the collapse; the risk post asks who is at risk; the rescue post shows what actually gets someone out. This one is the uncomfortable sequel: what the recovery is worth afterwards.

Status. Measured in the frozen model, traceable to the paper (doi:10.5281/zenodo.22943641). Interpretation is marked where it appears.

1. The intuition

After a person comes out of a collapse, there is a natural picture of what the recovery means. They have been through it. They know the territory now. They have a floor that works, people around them, a frame for what happened. The next time something pushes them inward, surely they are harder to collapse — that is what recovering is for.

It is a reasonable picture. The model says no.

2. The measurement

Set up the canonical rescue: the system collapses, is rescued, and settles fully healthy at G = 0.885. Then, later, apply the same second inward episode the first one used — no new mechanism, no retuning, only the schedule.

Read that carefully. The thresholds are the same to within a grid step. The onset timing is identical. And note the starting point is the same in both cases — because a full rescue returns G to where it started. The recovered state is G = 0.885. So was the naïve state.

The recovery bought exactly nothing.¹ Not a smaller margin, not a slower descent — a head start that is not a head start, because it restores the same value the healthy system already had.

3. Why — and it is the S-loop again

There is a specific mechanism, and it is the same one that makes the collapse irreversible in the first place: the setpoint.

Recall that the arming threshold is not fixed. It drifts with the allostatic setpoint S:

Θ_eff = Θ · S / S_rest

During recovery, S climbs back — in the measured run it goes from 0.13 (the drained, collapsed level) up toward 0.76. That rise is what recovery is, mechanically: the threshold returns to a healthy height so the switch stops arming on ordinary fluctuations.

Now apply the second episode. As attention goes inward again, S drains again — 0.76 back to 0.13 — and the threshold falls with it. And here is the point: G’s head start does not help, because the collapse is not decided by G.

The switch arms on the error. The second episode restores the same inward drive, which drains the same setpoint, which lowers the same threshold, and G — however healthy it started — falls at the same rate against the same falling bar. The recovered system is not defending from a better position; it is the same system, and the thing that was restored is also the thing that gets drained.

The S-loop that slowly repaired the system is exactly what re-collapses it.² Restoring the threshold is what recovery is; losing the threshold is what the second episode does. They are the same mechanism run in opposite directions, and the second run is faster than the first.

4. The measured asymmetry, and the honest gap

Two facts sit together and should be stated precisely, because they are different kinds of claim:

Measured: under single-episode schedules, the relapse fraction is exactly 0 — 816 runs in a controlled grid (6 delays × 17 strengths × 8 durations) plus 112 in a deliberately confounded grid, and no run lifts past 0.5 and then re-crosses below 0.1. A rescued state is stable as long as no new trigger arrives. The reason is micro-structural: after a full rescue the error sits at E ≤ 0.15, far below the raised threshold Θ_eff ≈ 0.76, so the switch simply cannot reactivate.

Also measured: the moment a new trigger does arrive, the protection evaporates — 66.0 vs 66.3 — and the system collapses as if it had never been rescued at all.

So the accurate statement is not “recovery is worthless.” It is: recovery restores the condition, and the condition is precisely what the next episode removes. A restored state is safe until it is needed, and then no safer than a naïve one.

The honest gap. The model has no memory term — no content persists between episodes, and consistency across episodes is not represented.³ So it says nothing about what a person learns, and it cannot represent the frame, the knowledge, the community, or the holder. Those are real, and this post does not claim they don’t work.

What it claims is narrower and harder: the part of recovery that is a return to the healthy baseline is not protection. It is a reset. And a reset is not a defence — it is the starting position, which is where the first episode came from too.

5. What it implies

Interpretation, and the same caveat as everywhere in this series:

Being back to normal is not the same as being safe. If the model’s structure holds, the thing that protects someone is not the recovery it is the holding structure around it — which is exactly what the coercion post’s inversion and the safeguards post are about. The baseline is not a defence. The guardrails are.

And it reframes what “recovered” should mean. In this model, a rescued system is one that has had its threshold restored — that is all. Its vulnerability is unchanged; the trigger that arrived before will work again, on schedule, and the number is the same. So the useful question after a collapse is not “are they back to baseline?” — they are — but “what is around them that was not there before?”

That is a much less comfortable question, and it is the one the measurement actually supports.

6. What this is not

• Not a claim that people don’t learn. The model has no memory, so it cannot speak to learning at all. Everything a person gains from an experience is outside what this measures.
• Not a prediction of inevitable relapse. Relapse in the model requires a new trigger — the single-episode relapse fraction is exactly zero. This is a statement about what happens when a trigger arrives, not that one will.
• Not a reason for fatalism, and not a reason for restriction. It argues for holding structure, not for avoiding practice or for treating anyone as fragile. (The series’ standing non-use clause applies: nothing here is a way to assess or restrict a person.)
• Not the memory result. The corpus also has consolidation findings — including that retrieval alone can trigger relapse with no external episode, and that the collapse writes more self-referential content while it happens (M_self 0.778 → 0.998). Those belong to a different, agent-side line of work and are not claimed here.

Notes

Marking: measured — §2’s thresholds and onset timings, §4’s single-episode zero-relapse result and its micro-structural reason. Interpretation — §5, and the reading of the S-loop in §3. Not claimed — anything about learning, memory, or the effect of frame, community and holder.