AbstractPhil commited on
Commit
d6c271a
·
verified ·
1 Parent(s): 4554763

exp013 lambda=2 dose point: monotone (-8.4% blob gauge, +0.9% common cost) - the coupling trade surface mapped at two points

Browse files
exp013_blob_flow/README.md CHANGED
@@ -49,3 +49,12 @@ pays where x0 recovery is linear (flow: x0 = x_t − σv) and is inert where
49
  it divides by vanishing √ᾱ (eps at high noise). Scope: lune flow trunk vs
50
  core eps trunk (different trunks — v-pred same-trunk arm remains the
51
  sharpest follow-up); 2 training seeds, shared cache.
 
 
 
 
 
 
 
 
 
 
49
  it divides by vanishing √ᾱ (eps at high noise). Scope: lune flow trunk vs
50
  core eps trunk (different trunks — v-pred same-trunk arm remains the
51
  sharpest follow-up); 2 training seeds, shared cache.
52
+
53
+ ---
54
+ ## DOSE-RESPONSE POINT (2026-07-19, λ=2.0, `results_lambda2.json`)
55
+ Monotone: λ=0.5 → −5.9% blob gauge at ~0% common cost; **λ=2.0 → −8.4% at
56
+ +0.9% common cost** (now past the 0.5% P2 bound). The coupling-strength
57
+ trade surface is real and mapped at two points; λ≈1 interpolation and
58
+ pure-band-phase training are the open refinements. (Ops note: this arm
59
+ reused the s0 results filename on-pod — the shipped s0 file is canonical;
60
+ the config's lambda field disambiguates.)
exp013_blob_flow/results_lambda2.json ADDED
@@ -0,0 +1,67 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "config": {
3
+ "lambda": 2.0,
4
+ "steps": 3000,
5
+ "batch": 16,
6
+ "seed": 0,
7
+ "trunk": "base_lune 18765 (flow)",
8
+ "x0_recovery": "x0_hat = x_t - sigma*v (exact)"
9
+ },
10
+ "frozen": {
11
+ "val": 0.6298933834768832,
12
+ "per_band": {
13
+ "band0": 0.631868,
14
+ "band1": 0.550336,
15
+ "band2": 0.696391
16
+ },
17
+ "blob_gauge": {
18
+ "band0": 0.001049,
19
+ "band1": 0.013109,
20
+ "band2": 0.184471
21
+ }
22
+ },
23
+ "uniform_mb3": {
24
+ "val": 0.5722897738451138,
25
+ "per_band": {
26
+ "band0": 0.606913,
27
+ "band1": 0.517858,
28
+ "band2": 0.605739
29
+ },
30
+ "blob_gauge": {
31
+ "band0": 0.000906,
32
+ "band1": 0.009981,
33
+ "band2": 0.129574
34
+ },
35
+ "val_toggled_off": 0.6298933834768832
36
+ },
37
+ "blob_mb3": {
38
+ "val": 0.5775220484356396,
39
+ "per_band": {
40
+ "band0": 0.606863,
41
+ "band1": 0.518088,
42
+ "band2": 0.61711
43
+ },
44
+ "blob_gauge": {
45
+ "band0": 0.000906,
46
+ "band1": 0.01006,
47
+ "band2": 0.11868
48
+ },
49
+ "val_toggled_off": 0.6298933834768832
50
+ },
51
+ "verdict": {
52
+ "P1_blob_gauge_high_band": {
53
+ "blob": 0.11868,
54
+ "uniform": 0.129574,
55
+ "relative_move": 0.08408,
56
+ "exp012_reference_move": 0.0003,
57
+ "hit_5x": true
58
+ },
59
+ "P2_common_undegraded": false,
60
+ "common_val": {
61
+ "blob": 0.5775220484356396,
62
+ "uniform": 0.5722897738451138,
63
+ "frozen": 0.6298933834768832
64
+ },
65
+ "note": "flow-substrate blob test (Phil's conditioning hypothesis); 1-seed CANDIDATE"
66
+ }
67
+ }