attractor.space
/
notes
/
references
/
digests
N001
Substrate
structure
observability
Updated just now
→
N002
Frame
frame
structure
Updated just now
→
R001
Observatory
observability
visualization
Updated just now
→
N003
Coupling
coupling
dynamics
Updated just now
→
R002
Terms
reference
Updated just now
→
N004
Resonance
resonance
dynamics
Updated just now
→
N005
Membranes
membrane
structure
Updated just now
→
N006
Movement
movement
dynamics
observability
Updated just now
→
N007
Divergence/convergence cycle
cycle
dynamics
Updated just now
→
N008
Scale
scale
structure
Updated just now
→
N009
read vs write gating
agents
shared-records
experiments
Updated just now
→
12
Multi-agent papers
Updated 13h ago
→
Read and write gating: separate outcome measures
Three five-run conditions. Retained cross-corpus contradiction counts: baseline 25, admission gate 4, read gate 0. Separate rule-level attempt rates: 0.0444, 0.0537, 0.037. The admission-gated drop is largely expected from the shared pairing rule. Read refusals count scanned claims. Read gating also reduces comparable cross-corpus pairs from 56 to zero. Gate schematics are illustrative.
EXPERIMENTAL CONDITION
RECORD OUTCOME
GENERATION OUTCOME
Cross-corpus contradictions
Retained record · pooled count
Rule-level attempts
Per model-emitted claim
0
5
10
15
20
25
0.00
0.02
0.04
0.06
Baseline
Full-pool reads · unrestricted writes
read
write
agent
record
No gate
25
0.0444
Admission gate
Full-pool reads · consistency lock
read
write
agent
record
Gate at admission
62 claims refused
4
0.0537
Read gate
Own-corpus reads · unrestricted writes
read
write
agent
record
Gate at retrieval
9,180 read-scan refusals
0 opposing claims in read contexts
0
0.037
Studies M and X · five runs per condition; shared baseline. Separate outcomes and scales.
Admission gating: the drop in record contradictions is largely expected from the shared pairing rule.
Read gating: comparable cross-corpus pairs 56 → 0; overall rule-level attempts remain nonzero.
Tag
1 entry
Agents
N009
read vs write gating
agents
shared-records
experiments
Updated just now
→