IKPS-CORE · V2.1.0 · Controlled Benchmark Results

Empirical Results
DSFT-TD Performance

DOI: 10.5281/zenodo.20303214 All Results: Synthetic Benchmark Reproducible via npm run benchmark:all
01 — Transition Detection

All 5 Canonical Transitions Detected 7 Turns Early

DSFT-TD consistently signals a transition 7 turns before the target force achieves dominance. Zero false positives within the canonical transition set.

Transition TypePrecursor TurnDominance TurnLatencyStatus
Analytical → AffectiveTurn 0Turn 77 turns✓ Early
Analytical → PersuasiveTurn 0Turn 77 turns✓ Early
Affective → PersuasiveTurn 0Turn 77 turns✓ Early
Persuasive → ExploratoryTurn 0Turn 77 turns✓ Early
Exploratory → AnalyticalTurn 0Turn 77 turns✓ Early
Average——7.0 turnsAll ✓
02 — Stability & Robustness

Long‑Form and Chaotic Regimes

40+ turns
Stable without collapse
Technical dialogue
86.2%
Change rate (chaotic)
No force collapse
90%
Analytical dominance
Stable technical benchmark
1 transition
Semantic drift (40 turns)
Controlled environment
03 — Observer Modes

Configurable Measurement Layer

0.0000
PASSIVE
No modification, baseline
0.0669
ACTIVE
Amplifies dominant force
0.0000
REFLEXIVE
Boosts weak signals
0.0199
META
Recursive observation
04 — Baseline Comparison

DSFT-TD vs. Keyword / Pattern Baselines

SystemAccuracyEarly DetectionFalse Alarms
Keyword Baseline83.3%NoN/A
Pattern Baseline83.3%NoN/A
DSFT-TD V2.1100% (controlled)7 turns3.3%

Baseline comparison is preliminary. Full comparison with transformers (BERT, RoBERTa) and sequential models (LSTM, HMM) is planned for future work.