BraInference

Society · Protocols

DriftWatch

Reputation scoring via behavioral deviation from baseline (2-sigma alert)

View in the graph

Definition

Mechanism

Reputation via micro-deviation detection. 1. Baseline: establish behavioral frequency over `window_size` samples. 2. Sample: continuous high-resolution observation. 3. Detect: alert when Distance(Current, Baseline) exceeds `threshold`. 4. Witness: aggregated peer reports. Tracks behavioral consistency against a baseline Aggregate of historical actions.

Preconditions

  1. Baseline state is established

Invariants

  1. Baseline is immutable while a sample is judged against it.
  2. Baseline recomputation interval and source window are declared.
  3. Alert when Distance(Current, Baseline) > `threshold`, in the units Distance returns.

Risks

Failure modes

  1. Baseline walked

    the subject changes more slowly than the baseline recomputes, so no alert fires.

  2. Witness collusion

    coordinated false drift reports.

  3. False reports

    a single malicious witness.

  4. Cold start

    no baseline yet, so the watcher is silent through the establishment window.

  5. Baseline gaming

    Agent varies early to establish a wide baseline.

  6. Alert flood

    continuous sampling against a sensitive threshold alerts at its false-positive rate.

Relationships

Builds on

References

Mind

Agent

Autonomous Goal-Directed Entity

Infrastructure

Aggregate

Reduce a set to a summary statistic

Physics

Distance

Metric-axiom function d(x,y) on a space — substrate for similarity, proximity, and spatial reasoning

Used by

References

Reference

Parameters

threshold Float, [0.0, inf)
Trigger in the units {{distance}} returns — not normalised, not sigma multiples.
window_size Integer, [10, 10000]
Samples forming the baseline
recompute_interval Integer, [0, 100000]
Samples between baseline rebuilds; 0 pins it permanently.

Identity

Ref
DriftWatch#b696
Sema ID
sema:DriftWatch#mh:SHA-256:b696959258eb3c48ce11055634b107451d5a500bc8ba19693db4c30b5f5033bf
Ring
0
Tier
1
Updated
2026-11-11
Supersedes
  • sema:DriftWatch#mh:SHA-256:a20dcee6f5453e3b61973fcf01aebd6a5ed4cfb97d1402a4ed970d8629a55a69
  • sema:DriftWatch#mh:SHA-256:49b6f6d68dbda93034116865bb93a7bd08eff23de49a2eef87af80a40fe4a9f5
  • sema:DriftWatch#mh:SHA-256:37cf935687877e2a31c4cd75287e2bc7c814b49a3569203021854988f39abe9e
MarkdownJSONThis version (DriftWatch@b696)Index of all patterns