95% interval [+1.0347, +1.4160] across 10 matched pairs — above zero, which is the whole of H1's pass rule. H1 PASS.
UNISWAP V4 · DELAYED EVIDENCE
THETASHIELD
Directional fees that protect liquidity from informed flow—not ordinary market noise.
Directional adaptive fees · Uniswap v4
Protect LPs from
signal, not noise.
ThetaShield waits for delayed signed markout, filters ordinary movement, and raises only the fee direction supported by persistent adverse-selection evidence.
The quiet loss: a trader buys at 100 and a minute later the market prints 101. The trader kept the stale price; the LP financed the gap. ThetaShield prices that gap only when it repeats.
Risk proxy—not exact LVR, individual LP loss, or a profitability claim.
POST-TRADE SIGNAL60s evidence delay
BUY-BASE FEE13.43bps · protected
SELL-BASE FEE5.00bps · baseline
Informed buying stream · 240 events · seed 101end of replay
Research replay from the locked evidence bundle — not live chain state.
What the LP keeps
Two comparisons, both measured.
7.70% of ordinary flow mispriced against 60.54% — while charging 6.28 bps mean fee against 8.16 bps. Signal-blind protection costs the traders who did nothing. Measured on a deterministic synthetic stream, not on an operator-moved live market.
The paired figure is 10 matched pairs; the pooled table across all 75 runs shows a smaller +0.4616 gap over the same baseline. Different populations, both in the record. Absolute LP net is negative for every policy on these synthetic streams — this is a relative comparison, not a profitability claim.
This bundle is not live-market, profitability, deployment, security-audit, Circle-attestation, or Reactive-callback evidence.
Static, volatility-reactive, and directional fee policies compared
Same streams · identical event countsTrailing — a trade cannot widen its own noise bandPersistent — one neutral epoch cannot erase toxic historyDirectional — buy and sell fees evolve independentlyPortable — Circle CCTP V2 carries the finalized evidence
Animated control loop
See the delayed fee travel.
Circle authenticates what crosses chains. Reactive Network drives when bounded processor work advances. The swap path never waits for either.
ACTIVE STEP 01 / 17Swap
Trader calls the v4 pool
240 events replayed · seed 101
UNICHAIN SEPOLIA · Swap path
PROVEN BY RECEIPT01 · Swap observed ↗
CONTROL JOURNEYSwap path
Unichain Sepolia is carrying this stage. The trace advances in execution order without moving the reader between disconnected lanes.
SELECTED PATH
bounded success pathFinalized evidence advances; the next swap can consume a fresh directional fee.
WHAT WOULD THE POOL CHARGE?parameters from the research bundle
- dead band±16.00k 1.00 × sigma
- survives the filter48.00signal
- signed risk7.20magnitude 12.00 × confidence
- premium3.24risk × gain
- fee next swap8.24bps
Computed by the same curve the pool runs — this port is checked against research/datasets/golden_vectors.json, the vectors that also gate the Solidity and the Python. It assumes one observation from rest, because the contract carries an EWMA magnitude between epochs that the bundle does not record. It is what the deployed curve would return for these inputs, not a live quote.
SILENT-LOSS SURFACES MADE VISIBLE
hook try/catch → ObservationTransportFailedqueue bound → DropReason.Capacityepoch bound → DropReason.EpochCapacityLP-benefit replay console
Interrogate the trade-offs.
These cards are simulated—not live chain state. Every displayed result comes from the deterministic evidence bundle. Parameter selectors replay exact one-factor Phase 6 cases; they do not invent untested combinations.
STREAMPersistent Informed Buying
Persistent toxic base buying with favorable future moves.
ACTIVE EXACT SENSITIVITYDead-band k · 1.5 · default
All other Phase 6 parameters remain at that experiment’s locked default. Each dimension’s “· default” is the replay’s trace configuration, which the bundle records separately from the selected research config shown in the registry — the two are allowed to differ.
01 · FEE BY DIRECTION240 events · 30s steps · every 4th plotted
ThetaShield buyThetaShield sellSelected-policy pooled mean
Directional lines are the ThetaShield control replay; the dotted guide is the selected policy’s pooled mean.
02 · WHO PAYSThetaShield
03 · SPECIFICITY / RECALLpooled + sensitivity
92.3% specificity (benign flow passed at baseline) · 30.3% recall · 77 step detection
04 · LP NET OUTCOMEtrue scale + paired zoom
Inventory PnL−731.012 quote
Fee revenue+2.231 quote
Net−728.781 quote
Dead band only−1441.983
Fixed fee−1443.093
Raw positive markout−1441.865
ThetaShield−1441.898
Volatility only−1441.857
Scenario mean −1441.898 quote · 95% interval [−1692.716, −1191.079]. Inventory PnL is shown at its real scale above; the zoom below spans only −1443.093 to −1441.857, so bar length compares policies against each other, not against zero.
05 · TRANSPORT HEALTHnormal · at event 1
100%callbacks applied
- Applied
- 0
- Missing
- 0
- Rejected
- 0
- Expired references · run total
- 0
Healthy loop06 · G1 CLOSED LOOPpass
98.99%benign volume retained
99.45%toxic volume retained
−0.0110fee-revenue delta · quote
The selected controller shifts part of the toxic-risk gain into bounded coverage feedback. It preserves the declared precision and retained-flow gates, but does not improve mean fee revenue in this deterministic synthetic experiment.
EXACT CASE · default
44.3%directional accuracy
69detection steps
+1.225persistent LP improvement
28.49oscillation · bps
Controlled deterministic synthetic evidence only; this is not live-market, profitability, deployment, or security-audit evidence.
Live testnet proof
Don’t trust the demo. Read the contracts.
Read directly from deployed contracts across Unichain Sepolia and Ethereum Sepolia, on every refresh.
ORIGIN · UNICHAIN SEPOLIAreading…
BUY-BASE FEE—bps——
SELL-BASE FEE—bps——
- Hook observations
- —
- Installed sequence
- —
- Read path
- —
- Contract code
- —
- Circle peer
- —
- Pause state
- —
- Pool
- —
- Chain ID
- —
PROCESSOR · ETHEREUM SEPOLIAreading…
—pending
—settled
—expired
- Last observation
- —
- Recommendation sequence
- —
- Dropped observations
- —
- Contract code
- —
- Chain ID
- —
Reading chain state…
Reading both public testnets before stating a fee.
Run the loop · live testnets
Drive the protocol from here
Reading the guards from both chains to see what this deployment will currently permit.
- Start a runUnichain Sepolia
Sends one bounded swap through the protected pool. The hook observes it and dispatches the evidence over Circle, which is what puts an observation into the queue on the other chain.
- ·reading the chain…
- Relay the messageEthereum Sepolia
Delivers the attested Circle message to whichever chain is missing it — the observation outbound, or the recommendation on its way home. Circle holds a message until its source chain finalises, which took about half an hour on the last run, so this step waits rather than fails. Which leg is outstanding is read from both chains' counters, not chosen here.
- ·reading the chain…
- Advance the cycleEthereum Sepolia
Runs one bounded automation cycle: sample the references, sync them, process whatever is due, and dispatch a recommendation if one is warranted. This is the same call the Reactive callback makes, and it is permissionless — the scheduler has no privilege here that a keeper lacks.
- ·reading the chain…
THE PROVEN RUN · LIVE RECEIPT TRAIL6 public transactions · walk the trail, or open any receipt
- interval not read
- interval not read
- interval not read
- interval not read
- interval not read
Read the contracts → all 14 deployed components, with block, verification status and explorer links
Reading this page is read-only: refreshing performs public RPC reads and never connects a wallet, signs a message, or spends testnet funds. The run console above is the one control that broadcasts, and it says so.
State is read through the deployed stateless ThetaShield lenses, with the audited getters as the declared fallback path.
Execution log · DEPLOYED · READ LIVE
Everything the queue has done.
Every observation the processor has queued on Ethereum Sepolia and every automation cycle that ran, rebuilt from the contracts’ own events rather than from a recorded run. CIRCLE CCTP carries the evidence and REACTIVE NETWORK schedules the sweep; neither computes a fee.
Reading the queue’s lifecycle from Ethereum Sepolia…
Verification registry
Every claim opens to a receipt.
Deployment 0x849fd0a1…85707c · profile RESEARCH_V1 · source revision 4b3aff624734 · recorded 2026-08-29. Everything below is generated from that recorded manifest; the deployed-parameter table reads the chain directly.
01Deployment registry
14 components · 3 networks▸
Unichain Sepolia · chain 1301 · Circle domain 10
ThetaShieldCircleTransport
0x4f00e3…092609block 61,140,130unverified · proven by receiptsdeploy tx ↗explorer ↗ThetaShieldController
0x23ae3E…63dFb7block 61,140,137unverified · proven by receiptsdeploy tx ↗explorer ↗ThetaShieldTestToken0
0x29E6D5…023005block 61,140,148unverified · proven by receiptsdeploy tx ↗explorer ↗ThetaShieldTestToken1
0xe97791…591Eb1block 61,140,142unverified · proven by receiptsdeploy tx ↗explorer ↗ThetaShieldHookFactory
0xb3c880…2b11b8block 61,140,153unverified · proven by receiptsdeploy tx ↗explorer ↗Ethereum Sepolia · chain 11155111 · Circle domain 0
ThetaShieldReferenceToken0
0x29E6D5…023005block 11,590,720unverified · proven by receiptsdeploy tx ↗explorer ↗ThetaShieldReferenceToken1
0xb3C880…2B11b8block 11,590,721unverified · proven by receiptsdeploy tx ↗explorer ↗PoolMedianReferenceSampler
0xEF9C63…593D5dblock 11,590,734unverified · proven by receiptsdeploy tx ↗explorer ↗ThetaShieldCircleProcessor
0x7bdF95…a4BBF2block 11,590,735unverified · proven by receiptsdeploy tx ↗explorer ↗ThetaShieldProcessorLens
0x4a1B45…2F1EaBblock 11,590,736unverified · proven by receiptsdeploy tx ↗explorer ↗ThetaShieldAutomationExecutor
0x1A3a27…ab9707block 11,590,737unverified · proven by receiptsdeploy tx ↗explorer ↗Reactive Lasna · chain 5318007
ThetaShieldAutomationRSC
0x4f00e3…092609block 4,777,133unverified · proven by receiptsdeploy tx ↗explorer ↗02Circle message evidence
2 messages▸
Generic CCTP V2 messages — no USDC burn. Delivery is permissionless; the receiving contract authenticates transmitter, source domain, sealed peer, and finality before any state changes.
03Deployed parameters
28 research keys · reading deployed values▸
DEPLOYED · READ LIVE
Reading the scheduler and fee-curve configuration from the deployed processor…
RESEARCH BUNDLE
alpha_wad- 0.25
base_fee_pips- 500 pips (5.00 bps)
callback_delay_steps- 1 steps
confidence_cap_wad- 0.6
confidence_floor_wad- 0.5
dead_band_k_wad- 1
epoch_observation_count- 4
fast_path_confidence_floor_wad- 0.5
fast_path_enabled- enabled
fast_path_hold_epochs- 0
fast_path_toxic_threshold_wad- 0.00075 (7.50 bps)
markout_delay_steps- 2 steps
markout_horizon_steps- 1 steps
maximum_decrease_pips- 100 pips (1.00 bps)
maximum_fee_pips- 10,000 pips (100.00 bps)
maximum_increase_pips- 500 pips (5.00 bps)
maximum_reference_dispersion_wad- 0.02
maximum_trade_notional_wad- 100
minimum_epoch_notional_wad- 8
minimum_fee_pips- 500 pips (5.00 bps)
minimum_observation_notional_wad- 1
minimum_trailing_observations- 16
persistence_window- 5
recommendation_ttl_steps- 12 steps
required_toxic_epochs- 3
target_observation_count- 4
toxic_threshold_wad- 0.00075 (7.50 bps)
trailing_window- 16
The research bundle locks the simulated candidate; the deployed pool answers live. The two columns are allowed to differ (the deployed confidence cap and fee gain do) — provenance is labeled so neither is mistaken for the other.
04Reference
cost · gas · judge Q&A · glossary▸
Deployment cost
Unichain Sepoliagas limit 11,406,580owner approved
actual 0.000006 approved max 0.00005 Unichain Sepolia ETH
Ethereum Sepoliagas limit 18,061,398owner approved
actual 0.02087 approved max 0.044042 Ethereum Sepolia ETH
Preflight fingerprints
0xb7fe46a4648b39fa370aedffc5fb5eb7e26f988bdf79869509c1d02d8e118a4f0x4a4ffae821da31bc073500fb08885b0024e82f5066e001f8a2bab560f0fe72450x80d9fa12d5c1a6f09d214ec70c03afc12f17222e74c5e90c3429a805efb4dc3f0x76f6a39250423b9dfc1a9048dd9f944f8b37524f6c6efa4da01bccf14194ae91Measured gas
MEASURED GAS · ISOLATED LOCAL EVMhook total 199,973 per swap
hook · beforeSwap + afterSwap (warm)33,192 + 166,781
controller · apply recommendation cold / warm142,543 / 15,175
controller · feeForSwap (warm read)3,771
Isolated local EVM call measurements under the pinned compiler profile. They exclude the PoolManager and router transaction and are not a live-chain cost quote.
Judge Q&A
Why split the system across chains?
The latency-sensitive hook stays small and deterministic on Unichain Sepolia, while delayed histories, bounded queues, confidence, persistence, and reference selection live on Ethereum Sepolia. Reactive Lasna runs only the automation plane, which schedules bounded work and holds no fee authority. Cross-chain delivery is never required for the current swap — it updates later swaps.
Why would a trader use it?
Routers choose pools by the all-in quote. ThetaShield aims to let LPs keep a low baseline for ordinary flow instead of charging every trader a permanent volatility premium. The commercial claim still requires real-market routing, depth, and elasticity evidence.
Can an operator label a wallet toxic?
No. The mechanism evaluates delayed directional price outcomes and never uses a wallet blacklist or identity score.
What if Circle or Reactive stops?
The current swap continues. Existing recommendations expire and the controller returns to the bounded baseline. Reactive is outside the critical path.
Is the live reference price real market data?
No. The RESEARCH_V1 reference market is 3 fee tiers of a project-issued pair on Ethereum Sepolia, separate from the protected Unichain Sepolia pair, seeded and moved by the operator. It exercises the full multi-source path — liquidity floors, robust median, dispersion, confidence — against a controlled market. Independent evidence needs reference tiers co-located with the protected pair; that is the next architectural step, not a claim made today.
Why did the live fee not increase?
The completed samples are deliberately cold start and carry zero shared confidence, so both directions hold the configured baseline. A safety mechanism that moved the fee off one observation would be easier to demo but easier to manipulate. A non-baseline directional fee needs a mature trailing window and shared confidence above the deployed floor; the local end-to-end lifecycle proves that transition, and the live panel above reports the current cycle count and confidence.
Glossary
- signed markout
- direction × (reference price − execution price) / execution price — the post-trade outcome with its sign preserved.
- dead band
- k × trailing volatility. Movement inside the band is treated as noise; the current sample is excluded from its own band.
- n-of-k persistence
- activation requires at least n toxic epochs inside a k-epoch window, so one noisy sample cannot move the fee.
- fee pips
- 1,000,000 pips = 100%, so 100 pips is 1 bp. The deployed baseline and cap for this pool are read live in the deployed-parameters table above.
- WAD
- fixed-point unit where 1e18 represents 1.0 — used for prices, markouts, risk, and confidence.
- coverage premium
- a bounded fee component that responds to estimated-loss coverage deficits, separate from the toxic premium.
- RESEARCH_V1
- the live release profile: a permissionless pool-median sampler over 3 liquidity-qualified pools replaces the owner-published demo feed used by the historical Phase 8D proof.
Falsifiable evidence
The failures stayed in the record.
Phase 6 failed H4 and H5. Phase 6.1 locked a versioned change on training streams, then scored it once on reserved holdout seeds.
POLICY SEPARATIONSignal-blind fees tax everyone.
Five policies on identical streams; the four dynamic policies' fee budgets were calibrated within 0.49 bps of each other, with the fixed baseline pinned at 5.00 bps. Whiskers are 95% intervals across seeds.
RESERVED HOLDOUTFailures remediated, then re-scored once.
H4 · Detection trade-offFAIL · historical→PASS · holdout
Gate: rank correlation is at most -0.35 with at least three Pareto points spanning 5 percentage points and 5 steps.
- 6 Pareto points (≥ 3 required)
- 22.75% false-positive span (≥ 5.00% required)
- 29 step latency span (≥ 5 required)
rank correlation · versioned remediation locked on training streams, then scored once on reserved holdout seeds.
H5 · Manipulation resistanceFAIL · historical→PASS · holdout
Gate: ThetaShield reduces raw-markout false positives and oscillation with lower intervals above zero while retaining at least 50% toxic-premium coverage.
- false-positive reduction 13.43%–28.15% (95% interval above zero required)
- oscillation reduction 809–3,623 fee pips (95% interval above zero required)
toxic coverage retained · versioned remediation locked on training streams, then scored once on reserved holdout seeds.
H1LP protectionPASS▸
Gate: paired 95% interval for ThetaShield minus fixed-fee LP net PnL is above zero.
measured as the paired LP figure at the top of the page
H2Benign-flow fairnessPASS▸
Gate: upper intervals stay within 10% of baseline mean fee and 5% false positives.
0.00% benign false positives
H3Noise robustnessPASS▸
Gate: paired 95% interval for raw-markout minus ThetaShield false-positive rate is above zero.
23.88% fewer false positives
H4Detection trade-offFAIL · historical / PASS · holdout▸
Gate: rank correlation is at most -0.35 with at least three Pareto points spanning 5 percentage points and 5 steps.
re-scored on reserved holdout — see the chart
H5Manipulation resistanceFAIL · historical / PASS · holdout▸
Gate: ThetaShield reduces raw-markout false positives and oscillation with lower intervals above zero while retaining at least 50% toxic-premium coverage.
re-scored on reserved holdout — see the chart
H6Directional discriminationPASS▸
Gate: ThetaShield's paired correct-direction advantage is at least 20 percentage points and mean volatility-fee correlation is at most 0.80.
44.33% directional advantage
Benign Flow Noninferiority
elastic benign retained-volume rate >= historical ThetaShield
PASSFeedback Is Exercised
coverage policy must observe both eligible and deficit epochs
PASSPrecision Preserved
coverage policy inelastic mean FPR <= historical ThetaShield mean FPR
PASSTotal Flow Noninferiority
elastic total retained-volume rate >= historical ThetaShield
PASSPOLICY COMPARISONselect a row to replay it
| Policy | Mean fee · bps | False positives | Detection · steps | Behavior |
|---|---|---|---|---|
| 6.44 | 13.19% | 39 | Filters, but reacts once | |
| 5.00 | 0.00% | — | Reference floor | |
| 7.33 | 33.85% | 5 | Noise-biased | |
| 6.28 | 7.70% | 77 | Directional + persistent | |
| 8.16 | 60.54% | 8 | Moves with market state |
ΣFull sensitivity sweep
41 locked cases · 11 dimensions▸
Every point is one locked one-factor Phase 6 case replayed across all scenarios and seeds — never an interpolated combination. Highlighted points sit on the false-positive / latency Pareto frontier.
Trust surface: This bundle is not live-market, profitability, deployment, security-audit, Circle-attestation, or Reactive-callback evidence. Evidence bundle thetashield-dashboard-g9-v2 · 10 content-addressed sources.