Cooperative Survival Network: what 15 mechanisms actually do under stress

Updated on 3 November 2026. Two caveats from version 9 of the paper added to the integration cascade: in the arm without PMA only three of the nine harness mechanisms act, and the nine are stand-ins in the simulation harness.

Added on 9 October 2026. Every figure in this article comes from the paper's ablation simulation. Our correction notice of 18 September 2026 corrects the count: under stress, 10 of 15 components reach the printed criterion (all ten seeds in the same direction), not 9: nine with a loss, the tenth, PriorityMap, with a gain; the article below still lists PriorityMap as cooperatively redundant. The starred p-values are all the same bound of the normal approximation at ten seeds (5.062 × 10⁻³), not a measure of effects of different strength.

In summer 2025 I was developing a memory system with eight algorithms. Sleep consolidation, FSRS-style scheduling, Hebbian update, Bayesian confidence, Ebbinghaus decay, emotional tagging, cross-layer routing, importance boosting. All implemented, all tested.

Then I removed sleep — and the quality metric stayed the same. Removed Hebbian — same. Removed Bayesian — same.

My first reflex: this is dead code. Time to clean up.

Three months later, on the first stress test with 60 days of simulated aging at decay = 0.25/day, seven of these "redundant" algorithms collapsed one after another — the system lost between 78.9 % and 93.7 % of its quality, depending on which was missing.

They were never redundant. They were cooperatively redundant.

What an ablation actually measures

A classical ablation works like this: take your fully-functional system, disable one single algorithm, measure the degradation. If performance stays the same, the algorithm is "redundant." If performance crashes, it is "critical."

That is reasonable methodology — until the system has redundant protection mechanisms. Then single-removal ablations systematically produce false negatives.

Example: imagine a car with a brake pedal and a parking brake. An ablation would test:

• Car drives off, brake pedal removed, drives test track. Stops with parking brake. No performance loss measured. Brake pedal looks redundant.
• Car drives off, parking brake removed, same track. Stops with brake pedal. No performance loss measured. Parking brake looks redundant.
• Conclusion: we can remove both!

The problem: the methodology only tests one failure at a time. It doesn't measure what happens under stress when both mechanisms are demanded simultaneously.

What we did in the paper

Instead of a single-difficulty ablation, the ZenBrain paper has three. They differ only in pressure:

All 15 mechanisms are removed individually at each level, the system runs 10 seeds, and we measure ΔQ = Retention × P@5. Wilcoxon signed-rank test, Bonferroni-corrected.

The result is a four-class taxonomy.

Class 1: Progressive algorithms (5)

These algorithms are fully redundant under moderate conditions — remove one, nothing happens. But the harder the load gets, the more critical they become:

* = Wilcoxon, for every starred value the same bound of the normal approximation at ten seeds (p = 5.062 × 10⁻³).

Read vmPFC-FSRS: under mild conditions you can remove it — the system doesn't notice. Crank aging up to 50 days at 0.20/day, and removing it costs 93.1 % of quality. Steep cliff.

Class 2: Always-critical (2)

The paper's overview (Table 4) groups these two as "always critical". The single rows show that only for sleep: the NeuromodulatorEngine loses −0.1 % without a star under moderate conditions and becomes clearly visible only under challenging and stress conditions.

Sleep is the only component that already provides the largest single contribution under moderate conditions (ΔQ = −34.4 %). NeuromodulatorEngine is just below threshold under moderate (−0.1 %) and one of the survival-critical tier under stress.

Class 3: Stress-only (2)

Two algorithms are redundant at moderate and challenging levels — and only become critical under extreme stress:

These are insurance policies — you never notice them until you need them. StabilityProtector prevents casual rewriting of mature memories under stress; Reconsolidation opens the update window with rollback safety.

Class 4: Cooperatively redundant (6)

Five of the following six components show no significant individual contribution at every level; removing them costs ΔQ ≤ 0.1 % in all three conditions. The sixth, PriorityMap, shows a significant effect only under stress, and it is a gain when it is removed:

iMAD Debate, Spectral KG Health, Compositional Context, HyperAgent, MetacogMonitor, PriorityMap.

Are they redundant?

Not necessarily: group removal shows what single removal hides. Removing all 6 PMA components at once (testing "Foundational-only" against "Full"; of the six above, only PriorityMap and MetacogMonitor are among them) collapses the system by −67.5 % under moderate conditions. Removing all 15 collapses it by −99.0 %.

Whether and where the six Class 4 components contribute individually, single removal does not show; the measured ΔQ combines retention and P@5. Whether they act in end-to-end tests graded by LLM judges, this ablation does not show. On LongMemEval-500, ZenBrain holds three of nine head-to-head comparisons, all against A-Mem; the six against Letta and Mem0 show no significant difference. Per the correction notice, the arm measured there is a Python port, not a measurement of single mechanisms.

The integration cascade

One last table from the paper makes the story particularly clear. Under extreme stress (decay = 0.30/day, 60 days):

Read that again: without PMA, the system falls to the same floor as the bare system. Version 9 of the paper adds two caveats. In the arm without PMA, only three of the nine harness mechanisms act (vmPFC-FSRS, IB Budget and Simulation-Selection sleep), so that floor measures three mechanisms, not nine; in the full system, four of the six PMA components and five of the nine mechanisms act. And the nine are stand-ins in a simulation harness; three of them share a name with the library's foundational algorithms. Within this harness, the PMA components keep memories alive long enough for the other mechanisms to reinforce them; the suite runs no arm with the PMA components alone.

Version 9 of the paper names the effect behind numbers like these cooperative masking: removing one component at a time can hide what the components do together.

What this means for engineering

The intuitive heuristic "if I can remove it without degradation, it's redundant" is wrong for resilient systems. It is right for pipelines with linear single-path dependencies — but not for layered memory architectures where algorithms cooperatively compensate for load.

Practical takeaways for memory engineering:

1. Ablations should test multiple stress levels. A single-difficulty ablation hides 7+ critical algorithms.
2. End-to-end metrics matter more than retention metrics. Whether the six Class 4 components act in end-to-end grading, this ablation does not show; that would need a measurement of its own.
3. Group removal is more informative than single removal. If removing all 6 PMA components under moderate load costs −67.5 % but no individual one costs more than −0.1 %, that's a clear sign of cooperative redundancy, not of obsolescence.
4. Resilient systems look over-engineered under test. That is a feature, not a bug. Fault-tolerant designs intentionally have more redundancy than typical load requires.

Practitioner guidance

The paper suggests a concrete recommendation for practitioners with resource constraints: which algorithms to implement first?

Tier 1: sleep consolidation, already the largest single contribution under moderate load in the simulation; plus the NeuromodulatorEngine, whose single-removal loss becomes clear only under challenging (−34.8 %) and stress (−83.0 %) conditions.

Tier 2 (progressive): vmPFC-FSRS, TripleCopy, Dual-Process CoT, Two-Factor Hebbian, IB Budget. These become critical as load rises — implement them before scaling.

Tier 3 (stress-only): StabilityProtector, Reconsolidation. Implement them for production deployment with long-term memory persistence.

Tier 4 (cooperatively redundant): iMAD, Spectral, Compositional, HyperAgent, MetacogMonitor, PriorityMap. This ablation shows no individual loss for them.

Read more

• Cost ratio: 91 % of the accuracy at 1 % of the tokens
• PMA explained: Predictive Memory Architecture — the 6 components
• Predecessor: 9 neuroscience algorithms behind ZenBrain
• Paper: ZenBrain on Zenodo