WAQAY the safeguarded sovereign memory

WAQAY (Quechua: to keep / guard / store) is our governed, air-gapped, DSSE-signed quantized vector index. We studied the MIT-licensed turbovec and Google Research's TurboQuant data-oblivious quantizer, then built our own pure-Python governed index. Every build and every retrieval emits a signed provenance receipt and passes the Restraint gate. 0 CDN.

Compression · MEASURED

—×

fp32 vs WAQAY bytes

Recall@k · MEASURED

—

vs exact float32 (SAMPLE)

Recall@1 · MODELED bound

—

turbovec profile · never perfect

Train phase

none

data-oblivious · online add

1 · Ingest (SAMPLE)

—

2 · Quantize

—

3 · Retrieve

—

4 · Restraint verdict

—

5 · Signed receipt

—

Top results (approximate · lossy quantized index)

Run a query to see governed retrieval…

Signed retrieval receipt DSSE + Restraint verdict

Run a query to see the DSSE-signed receipt + Restraint verdict…

locked theorems = 8 {F1,F4,F7,F11,F12,F18,F19,F22} @ kernel c7c0ba17 · Λ = Conjecture 1 · Khipu = Conjecture 2 · SLSA L1 honest / L2·L3 roadmap · receipts: DSSE ECDSA-P256-SHA256 · 0 CDN · trust ceiling < 1.0 (recall is a MODELED bound, never perfect).
Honest perf: pure-Python NumPy index INSPIRED by TurboQuant — compression is MEASURED; throughput vs the Rust SIMD original is MODELED/ROADMAP, never claimed to beat FAISS. Attribution: turbovec © 2026 Ryan Codrai (MIT) + Google Research TurboQuant — see NOTICES.md.