mirror of
https://github.com/allaunthefox/Research-Stack.git
synced 2026-07-31 03:05:21 +00:00
87 lines
2.8 KiB
Markdown
87 lines
2.8 KiB
Markdown
# FSDU Hyper-Heuristic Bridge
|
|
|
|
Status: `ADMIT_FSDU_BRIDGE_RECEIPT`
|
|
|
|
Claim boundary: this is a receipt projection from the existing
|
|
hyper-heuristic orchestrator demo metrics into the FSDU dual-map scar surface.
|
|
It is not a live path optimality proof, not a compression claim, not a hardware
|
|
claim, and not a replacement for the Lean FSDU scaffold.
|
|
|
|
## Anchors
|
|
|
|
```text
|
|
Lean anchor: 2-Search-Space/FAMM/FAMM_FSDU.lean
|
|
Theory anchor: 2-Search-Space/FAMM/docs/FSDU_theory.md
|
|
Input receipt: 4-Infrastructure/shim/hyper_heuristic_orchestrator_receipt.json
|
|
Bridge script: 4-Infrastructure/shim/fsdu_hyperheuristic_bridge.py
|
|
Bridge receipt: 4-Infrastructure/shim/fsdu_hyperheuristic_bridge_receipt.json
|
|
Stack mirror: shared-data/data/stack_solidification/fsdu_hyperheuristic_bridge_receipt.json
|
|
```
|
|
|
|
## Equation Surface
|
|
|
|
```text
|
|
X_t = (M_a, M_b, S_a, S_b, Theta)
|
|
DeltaS_t = S_a - S_b
|
|
commit allowed iff ||DeltaS_t|| <= epsilon
|
|
```
|
|
|
|
The bridge projects each orchestrator component into:
|
|
|
|
```text
|
|
ahead_scar = (1 - success_rate) * total_operations + 0.01 * heuristic_count
|
|
behind_scar = (1 - success_rate) * total_operations * success_rate
|
|
scar_delta = ahead_scar - behind_scar
|
|
```
|
|
|
|
This is a deterministic accounting projection. It is not a physical scar
|
|
measurement and not a proof that a live solver path is admissible.
|
|
|
|
## Static Bridge Result
|
|
|
|
```text
|
|
epsilon: 1.25
|
|
component count: 5
|
|
max abs scar delta: 1.223152
|
|
decision: ADMIT_FSDU_BRIDGE_RECEIPT
|
|
```
|
|
|
|
Per-component projection:
|
|
|
|
| Component | Success rate | Ops | Ahead scar | Behind scar | Abs delta | Decision |
|
|
|---|---:|---:|---:|---:|---:|---|
|
|
| FAMM_DELAY | 0.815 | 20 | 3.73 | 3.0155 | 0.7145 | COMMIT_ALLOWED |
|
|
| PIST_MOVE | 0.771 | 15 | 3.465 | 2.648385 | 0.816615 | COMMIT_ALLOWED |
|
|
| SHIM_SELECTION | 0.718 | 12 | 3.414 | 2.429712 | 0.984288 | COMMIT_ALLOWED |
|
|
| GPU_SCHEDULING | 0.771 | 15 | 3.475 | 2.648385 | 0.826615 | COMMIT_ALLOWED |
|
|
| FPGA_BUILD | 0.686 | 12 | 3.808 | 2.584848 | 1.223152 | COMMIT_ALLOWED |
|
|
|
|
The FPGA build component is closest to the envelope. That makes it the first
|
|
candidate for retuning if epsilon is tightened or if live metrics increase scar
|
|
pressure.
|
|
|
|
## Live Follow-Up
|
|
|
|
The next gate has been run:
|
|
|
|
```text
|
|
script: 4-Infrastructure/shim/fsdu_live_hyperheuristic_probe.py
|
|
receipt: 4-Infrastructure/shim/fsdu_live_hyperheuristic_probe_receipt.json
|
|
decision: HOLD_LIVE_SCAR_DIVERGENCE
|
|
max abs scar delta: 14.392819767
|
|
worst component: fpga_build
|
|
```
|
|
|
|
This confirms the bridge was useful: static demo metrics admitted, while the
|
|
seeded live software snapshot refused commitment under the FSDU gate.
|
|
|
|
## Next Gate
|
|
|
|
The next meaningful step is:
|
|
|
|
```text
|
|
wire live orchestrator state snapshots into the same FSDU fields
|
|
```
|
|
|
|
That has now been completed by the live probe. The remaining work is a seed
|
|
sweep and retune pass, not a claim promotion.
|