Perturbation-Response Map¶

Failure-footprint model — instantiates Neighborhood-Preserving Substrate Mapping

Charts, region by region, what downstream damage follows when each part of the substrate is knocked out — turning 'what if this fails' into a readable footprint map.

A neighborhood-preserving map is supposed to keep local failures local: knock out one region and the damage should stay near it, stay bounded, and be interpretable. Perturbation-Response Map turns that promise into something you can see. It perturbs — or removes — each substrate region in turn and records, as a map, the resulting downstream response: where the deficit shows up, how large it is, how far it spreads. Its defining idea is that it produces a spatial field of failure footprints across the whole substrate — the observable answer to "if this part breaks, what breaks with it?" — so the archetype's locality claim can be read region by region instead of assumed. Where a lesion probe performs one ablation as a test, this is the aggregated, readable footprint the probes add up to.

Example¶

An SRE team runs a service mesh whose dependency map is laid out so tightly-coupled services sit together. To check that the coupling is really local, they build a perturbation-response map: one at a time, they inject failure into each service — latency, then a hard kill — and record which other services degrade and by how much, the blast radius.^[1] The result is a map where each service is shaded by the size and shape of the outage it causes. Most light up only their immediate neighbours — good, locality holds. But one payment adaptor, drawn in a quiet corner, lights up half the mesh: its failure is non-local, a hidden hub the layout concealed. That single far-reaching footprint is the finding — the map surfaced a nonlocal failure before it happened in production.

How it works¶

Enumerate the regions to test — every substrate cell, or a sampled/high-value subset.
Perturb one region at a bounded strength (degrade or remove) and hold everything else fixed.
Measure the response elsewhere — which regions change, by how much, and how far the effect propagates.
Record a footprint per region, then assemble the footprints into one readable field over the substrate, highlighting any region whose footprint is large or reaches far.

It models and displays failure footprints; it consumes a perturbation procedure rather than defining the test plan, and it neither repairs nor reallocates.

Tuning parameters¶

Perturbation strength — a gentle nudge probes local, near-linear response; a full ablation reveals worst-case blast radius but can mask graceful degradation.
Response metric — what counts as "affected" (error rate, latency, lost coverage) and the threshold for inclusion in a footprint.
Spatial vs. list output — a heat-field over the substrate (shows locality at a glance) or a ranked list of worst regions (drives triage).
Region coverage — exhaustive testing of every region (thorough, costly) versus a sampled or targeted subset.
Propagation horizon — how many hops, or how long, to follow the response before a footprint is called closed.

When it helps, and when it misleads¶

Its strength is converting the abstract "failures stay local" promise into a concrete, per-region picture, and surfacing the dangerous exceptions — hidden hubs whose footprint is nonlocal — that a static layout hides from the eye.

Its failure mode is dependence on the perturbation model: a small linear nudge can look reassuringly local while the real-world failure is a nonlinear cascade the nudge never triggers. The classic misuse is running only gentle, cheap perturbations and reading the resulting all-local map as proof of robustness. The discipline is to match perturbation severity to the failures actually feared, and to treat a clean map as valid only up to the strength that was tested.

How it implements the components¶

local_failure_footprint_model — each perturbation yields a footprint; the assembled set is the model of what local failure damages and how far it reaches.
map_observability_and_legend — rendering the footprints as one shaded field makes failure propagation legible, a legend of where the map's locality holds and where it doesn't.

It does not define the ablation or lesion procedure it runs — that is the Local Ablation or Lesion Probe and Canary Region Probe siblings — and it does not enact recovery or re-layout (Adaptive Remeshing and Reallocation).

Instantiates: Neighborhood-Preserving Substrate Mapping — it verifies the archetype's promise that damage to one region stays bounded and interpretable.
Consumes: Local Ablation or Lesion Probe supplies the per-region perturbation this map aggregates.
Sibling mechanisms: Local Ablation or Lesion Probe · Canary Region Probe · Map Fold and Collision Scan · Coverage-Hole Heatmap

Notes¶

The map is descriptive, not prescriptive: a bright, far-reaching footprint flags a design fault — nonlocal coupling the layout failed to contain — but the remedy (re-laying-out the region, adding redundancy, splitting the hub) lives in other mechanisms. Its job is to make the fault visible and located, not to fix it.

References¶

[1] Blast radius — the extent of collateral damage from one component's failure — is the working term in chaos-engineering practice, where failures are injected deliberately to measure it. A layout that preserves locality should keep every region's blast radius small and near; this map is how you check. ↩