Weekend void
$3.5B
of tokenized equities on Solana with no principled price, every weekend (Q1 2026 snapshot).
Phase 1 · Devnet + Research Program
Real-world assets
Real-world assets
don't sleep on the weekend.
The promise of tokenized equities is real 24/7, permissionless markets. The missing infrastructure is a defensible closed-market reference with explicit uncertainty. Soothsayer publishes calibration-transparent fair-value bands, audit receipts, and downstream risk evidence grounded in market microstructure, protocol research, and 12 years of public data.
The gap
Every weekend, the incumbents stop thinking.
Weekends are the cleanest proof of the gap. Chainlink's consumed fields either freeze at Friday close or route to an opaque tokenized mark, Pyth + Blue Ocean extend into 24/5 but still hand back the weekend, and continuous off-hours marks remain difficult to audit. The market has activity, but not yet shared closed-market pricing infrastructure.
Provider
MonTueWedThuFriSatSun
ChainlinkLTP frozen / tokenized mark live
Pyth + Blue Ocean24/5 overnight ATS
Soothsayerfair value + calibrated band
● THE 48-HOUR GAP — Fri 8pm → Sun 8pm ET
— Chainlink weekend: frozen LTP, placeholder-derived v11 mid
Market admission
Incumbents understand the risk.
The weekend gap is no longer a hypothetical edge case. Providers themselves are warning that 24/7 tokenized markets can drift away from any defensible reference when the traditional venue is shut and the oracle stack does not adapt.
“We might see a dislocation of the tokenized stock versus the real value on Nasdaq.”
Marcin Kazmierczak · RedStone · quoted in CoinDesk
“If the oracle doesn't update until markets reopen, on-chain protocols could be trading on ghost prices.”
CoinDesk weekend-gap coverage · On-chain stocks mispricing risk
What we build
Primitive, measurement & decisions.
Soothsayer is one stack with three outputs: the calibrated band and receipt, the public measurement layer built from those receipts, and the decision support that turns uncertainty into protocol-relevant evidence.
01 · PRIMITIVE · PAPER 1
Publish a calibrated band and its receipt.
Paper 1 formalizes the coverage-inversion primitive: a public Solana fair-value band with target coverage, claimed-served quantile, and replayable audit receipts instead of a black-box point estimate.
12 years
5,986 panel rows
Reproducible
02 · POLICY · PAPER 3
Map calibrated uncertainty into protocol action.
Paper 3 turns the primitive into protocol-facing policy: reserve-buffer scoring, closed-market decision rules, and evidence for how lending systems should respond when uncertainty is explicit rather than hidden.
Open dataset
Mechanism design
Solana grant
03 · DECISION SUPPORT · PAPER 3
Show how downstream users act on the signal.
Paper 3 asks how protocols, liquidators, and risk teams should use the same band and receipt: which reserve buffers, triggers, and policy defaults look defensible when markets are closed.
Expected loss
Protocol policy
Deployable
Methodology
Five pieces, one auditable contract.
The product is the served band, not the raw forecaster. These five pieces make up the full contract a consumer receives. Each is reproducible from the repo against public data: 12 years of yfinance + per-symbol vol indices, no proprietary feeds. Full evolution log in reports/methodology_history.md.
01
F1 · F0
Raw forecaster stack
F1_emp_regime: Friday close × per-symbol factor return (ES / GC / ZN / BTC), conditional sigma from a log-log regression on the per-symbol vol index (VIX / GVZ / MOVE), empirical quantiles on standardised residuals — no Gaussian assumption, no parametric vol model. F0_stale: Friday close held forward + 20-day Gaussian, the high-volatility hybrid fallback.
02
v1b_bounds.parquet
Empirical calibration surface
For every (symbol, regime, forecaster, claimed-quantile) cell we measure realised coverage on rolling history and persist the table. At serve time, a request (s, r, τ) returns the band at q_served = S⁻¹(s, r, τ_buffered) — the claimed quantile that has historically delivered τ on this bucket.
03
REGIME_QUANTILE_TABLE
Mondrian conformal-by-regime + δ-shifted c(τ)
The deployed v2 / M5 architecture is Mondrian split-conformal by regime: a per-regime conformal quantile q_r(τ) trained on pre-2023 weekends, served after a multiplicative OOS-fit bump c(τ) and a walk-forward-fit shift δ(τ). C_BUMP_SCHEDULE = {0.68→1.498, 0.85→1.455, 0.95→1.300, 0.99→1.076}; DELTA_SHIFT_SCHEDULE = {0.68→0.05, 0.85→0.02, 0.95→0.00, 0.99→0.00}, linearly interpolated. The v1 hybrid forecaster Oracle is preserved as historical evidence in Paper 1 §7.1–§7.5.
04
PricePoint
Audit receipt on every read
target_coverage, calibration_buffer_applied, claimed_coverage_served, forecaster_used, regime — exposed on every PricePoint. The trust primitive, in the data structure itself: any consumer can replay the inversion against the persisted surface.
05
paper §6.4
OOS Kupiec + Christoffersen — full PASS
Held-out 2023+ slice (1,720 rows × 172 weekends). τ=0.95 → realised 0.950, p_uc=1.000, p_ind=0.485. τ=0.85 → realised 0.855, p_uc=0.541, p_ind=0.185. τ=0.68 → realised 0.678, p_uc=0.893, p_ind=0.647. Full surface in
reports/v1b_calibration.md; ablation with bootstrap CIs in reports/v1b_ablation.md.TSLAxxStock · Sat 14:32 UTC · τ = 0.85 default
Regime · normal
$241.07
point · half-width ±2.30% (≈ 230 bps)
band $235.52 — $246.62
claimed_served 0.895
buffer 0.045
Friday close
ES=F return
VIX
SPYxxStock · Sun 03:10 UTC · τ = 0.85
Regime · normal
$604.11
point · half-width ±1.42%
band $595.54 — $612.68
claimed_served 0.891
MSTRxxStock · Sun 18:40 UTC · τ = 0.85
Regime · high vol
$412.66
point · half-width ±3.94%
band $396.39 — $428.93
forecaster F0_stale
Incumbent admission
“Without these capabilities, tokenized products are exposed to potential mispricings, unfair liquidations, regulatory gaps, and other critical risks.”
Chainlink · v10/v11 launch · blog.chain.link
Compare
The gap is not just coverage. It is what gets priced, how, and with what evidence.
Closed-market risk is not just about who publishes a number. It is about whether the closed-market reference, uncertainty band, and calibration evidence are legible enough for downstream protocols and researchers to evaluate.
| Provider | Closed-market handling | Methodology | Confidence interval | Open source |
|---|---|---|---|---|
| Chainlink Data Streams | ~ field-dependent | Frozen LTP or opaque tokenized mark | — none verifiable | — closed |
| Pyth + Blue Ocean | — overnight only | Weighted median | 10.2% at claimed 95% | ~ partial |
| RedStone Live | ~ 24/7 marketed | — underlier only · undisclosed | — none | — closed |
| Soothsayer | ✓ weekend + CI | ✓ published math | ✓ aggregate-feed receipts | ✓ Apache-2.0 |
What it protects against
Three expensive failure modes in closed-market lending.
Protocols still need a number to act on when the reference venue is closed. These are the three places a stale or opaque number turns directly into loss.
01 · LIQUIDATION
Weekend flash crash on a thin book.
WithoutBorrower liquidated on a thin-book print that reverts Monday.
With SoothsayerThe receipt shows the print outside the empirically calibrated range, giving the protocol a defensible basis to pause or tighten policy with an audit trail.
02 · GAP RISK
Adverse Monday open.
WithoutStale hold says Friday close is fine. Monday opens –8%. Bad debt.
With SoothsayerFactor-adjusted point and band widen with the weekend signal set, giving risk teams and policy engines evidence to tighten buffers before the open.
03 · ORACLE OUTAGE
Primary oracle goes stale.
WithoutFreeze the protocol and lose revenue, or trust stale and take bad debt.
With SoothsayerAn independent calibration-aware cross-check and fallback reference, with uncertainty surfaced instead of implied.
Design partners
Build with calibration-transparent closed-market risk infrastructure.
Looking for a small number of lending, perp, and RWA teams that want defensible closed-market references, explicit uncertainty, and public evidence around liquidation and weekend risk.