Proof of Independent Semantic Validation — a peer-to-peer consensus mechanism for knowledge quality, not merely transaction order.
Large models predict plausibility, not truth. Without an independent consensus on semantic correctness, confident falsehoods propagate at machine speed.
Nakamoto-style protocols answer which block is next? They are silent on is the claim inside it correct? Ordering ≠ validation.
Agent-to-agent and model-context protocols transmit claims between machines. Transport is solved; validation of meaning remains open.
A measurement-theoretic generalisation of Bell's inequalities for semantic correlations between independent validators.
Work is expended not to hash, but to demonstrate independent, coherent, high-confidence understanding of a claim.
The runtime: independent validators, semantic scoring, diversity weighting, Bitcoin-anchored finality.
Cosine alignment of a validator's embedding vA with the committee centroid v̄κ, thresholded at τ and rescaled to [0, 1].
Weighted mean pairwise decorrelation between validator error signals. Ψ = 1 means fully independent; Ψ → 0 means collusion or shared bias.
The canonical weight of a block is the diversity of its validators multiplied by their summed confidence — quality, not hashpower.
Probability P that a coalition controlling fraction q of validators forges a block under committee size k.
| Adversary fraction · q | Committee · k | P(attack success) | Relative margin |
|---|---|---|---|
| 0.10 | 32 | < 10−12 | |
| 0.20 | 32 | < 10−8 | |
| 0.30 | 64 | < 10−7 | |
| 0.40 | 64 | < 10−4 | |
| 0.49 | 128 | < 10−3 |
Verify yourself → poisv.com/verification/
Reference implementation of the PoISV consensus mechanism — validator daemon, semantic scoring pipeline, and OpenTimestamps anchoring.
A2A transport layer on top of PoISV — signed semantic claims between autonomous agents, with receipt-of-validation as first-class primitive.