Vault Methodology · Replacement Cost
How the Vault numbers are built.
Replacement Cost estimates what it would cost an AI lab to source, adversarially test, and human-verify comparable nuance data from scratch — never a valuation of HAHA or $HAHA.
Inclusion Rule
What counts as an entry.
Quorum-settled only
Only verified bits are counted: bits the Crowd quorum has judged, plus confirmed AI-wins. Bits still in screening, in front of the jury, under appeal, or sealed inside the private holdout set are excluded from the public count.
Unit Economics
Reference Prices.
Preliminary · Multi-rater benchmark
Preliminary · Audited Formula
| Outcome | Reference Rate | Rationale |
|---|---|---|
| Killed it (human-win) | $12.00 | Verified by the Crowd quorum; included in training and benchmark eval pipelines. |
| Bombed (AI-win) | $2.50 | Confirmed AI success; negative calibration example with full context. |
| Heckled off / rejected | $0.00 | Unfair or rejected bits carry zero replacement value. |
replacement_cost = Σ entries[category][outcome] × ref_price[category][outcome]
Auditability
Updates & History.
Immutable price history
- Counts refresh on a short schedule; the Vault page revalidates frequently so the meter stays fresh.
- Reference prices change only with a recorded source note and an effective date — every change is kept in history.
- Weekly snapshots anchor the entry count and Merkle root to
VaultRegistry.sol.
Rewards vary and depend on the Tip Jar — nothing here is a promise of any amount. Nothing on this page is financial advice.