Back to Agents That Account for Themselves
working paper2026Behavioral motifs paper

Geometric Motifs for Selecting and Routing Coding-Agent Training Data

This paper extends the trajectory ledger by annotating sessions with behavioral motifs and geometric features. The goal is to condition training data generation on the kind of behavior a session exhibits, not merely on final outcome.

Paper workspace

Live draft structure

working-draft

Artifacts

Markdown paper source

Behavioral motif paper mapped from KARL source. Public PDF render pending.

source-only

Editable source

Paper source exists. Results should remain tied to exact snapshot IDs before PDF publication.

Source anchors

karl/paper/behavioral-motifs-paper.md

karl trajectory annotation/routing scripts

evo-cube-output/karl-trajectory-intelligence/stage0-research.md

Method tags

behavioral motifstrajectory geometrytraining data routing

Ingest intersections

motifstrajectorygeometrykarltraining-data

Status

Paper drafted; motif routing results should be kept tied to exact evaluation snapshots.

Key claims

01

Recurring agent behaviors can be represented as compact motifs.

02

Geometry-conditioned routing can improve specificity when quota-balanced.

03

Unconstrained routing can degrade quality by over-selecting easy residual sessions.

Public reading note

Public abstract safe; results should cite exact snapshots when released.

Standard skeleton

What this paper must keep proving

Schema

problem

High-scoring sessions are not all the same; training data needs to preserve the behavioral pattern that made a session valuable.

method

Annotate sessions with symbolic motifs and geometric scalars, then route them to specialized training-data lenses.

implementation

KARL motif annotation, yield scoring, quota-balanced routing, and quality verification.

data

Real multi-project coding sessions summarized as motif labels and geometry features.

evaluation

Transition-pressure prediction, effect sizes against random selection, and quota-balanced routing quality.

references

Conditional memory, Engram-style pattern storage, agent trajectory learning, reward routing.

openQuestions

Whether motif-conditioned training improves downstream agent behavior under official benchmarks.

Checkpoints and references

Proof chain

paperpending

Claim checkpoint

central-claim slot

Every central claim must point to a proof anchor or remain labeled as speculative.

implementationpending

Implementation checkpoint

implementation-map slot

Every method should identify the code path, harness, schema, or protocol that embodies it.

experimentpending

Evidence checkpoint

evidence-manifest slot

Every reported result should point to run IDs, packet IDs, data snapshots, commits, or review artifacts.

external-referencepending

Reference checkpoint

references slot

Every external claim should resolve to a cited paper, benchmark, standard, or documented prior system.

paperpending

Release checkpoint

release-gate slot

Every PDF needs a named condition before it can move from draft to citation-ready.