Inspect inputs before action

Make every signal earn the right to decide.

Price and weather can drive one automated action only after each signal independently clears freshness, agreement, and source quorum.

Observation bundle CSV0 observations

Bundled observations are synthetic. No live feed is queried.

Pre-action decision

Evaluate every required signal before proceeding.

WAIT

ETH / USD

NOT TESTED

Center — · accepted — · quorum 2

0 plotted observations

Rainfall / mm

NOT TESTED

Center — · accepted — · quorum 2

0 plotted observations

The action is conjunctive: every required signal must pass its own gate.

Intelligence cannot repair a missing chain of evidence.

Fresh first

Age limits belong to each signal. A market quote and a rainfall observation can have different useful lifetimes.

Agreement next

The fresh median anchors a disclosed tolerance band, keeping visibly inconsistent observations out of the decision.

Quorum last

Every required signal must retain enough accepted sources. One passing channel cannot compensate for another failing.

Agent Input Trust Gate Verification & Gating Logic

How does the agent input gate evaluate multi-source observation bundles before allowing an action?

This interactive tool processes multi-source observations across multiple signals (such as ETH/USD price and rainfall) through a sequential three-stage gating pipeline: freshness, median agreement, and quorum. First, observations older than the signal's maximum allowable age are classified as stale. Second, the median of the fresh values establishes an anchor center, and a relative percentage tolerance band filters out deviant outliers. Finally, an individual signal passes only if the count of accepted observations equals or exceeds its quorum requirement. The overall decision is conjunctive: every required signal must pass its quorum for the gate to permit execution.

The model applies deterministic, heuristic in-browser arithmetic to synthetic observation bundles. It does not check cryptographic digital signatures, verify oracle node identity, enforce transport layer security, or validate Byzantine fault tolerance across decentralized networks. The relative median band heuristic assumes honest majority among fresh reporting sources; if compromised feeds form the median, genuine readings could be flagged as deviant.

Try a worked example

Click 'Evaluate trust' with the default bundle. The engine evaluates two signals: ETH/USD has three observations with maximum age 60s, yielding a fresh median center of 3510.00 and ±1% band (3474.90 to 3545.10); source_A (3500) and source_B (3510) are accepted while source_C (4200) is rejected as deviant, meeting the quorum of 2. Rainfall has three observations where station_C is 400s old (exceeding the 300s limit), leaving station_A (22) and station_B (23) to establish a fresh median of 22.50 within ±10%, satisfying its quorum of 2. Because both signals meet quorum, the gate displays ALLOW. Clicking 'Corrupt ETH source' modifies source_B to 4000, leaving only one observation within the tolerance band, causing ETH to fail quorum and flipping the gate to BLOCK.

Three-Stage Evaluation Pipeline

For each signal group in the parsed CSV, the gate first discards rows where age_seconds exceeds max_age_seconds. The remaining observations form the fresh cohort.

The median of the fresh values is calculated as the signal center. A tolerance threshold is calculated as the absolute center multiplied by tolerance_pct / 100. Observations deviating from the center by more than this band are marked as deviant, while those within tolerance are accepted.

The signal passes if the accepted count meets or exceeds the specified quorum. The system requires all designated signals to pass for an overall ALLOW decision.

Fresh medians, tolerance and per-signal quorum

Your browser does not support video.

Read the explanation

This tool uses synthetic observations and queries no live feeds. Freshness is inclusive: age must be at most maximum age. Rainfall station C is four hundred seconds old against a three hundred second limit, so it is excluded from the fresh median. The two fresh values, twenty two and twenty three, average to twenty two point five. No signatures or source independence are verified. The three fresh Ether prices are thirty five hundred, thirty five ten and forty two hundred. Their median is thirty five ten. One percent tolerance gives a band of thirty five point one. Source A differs by ten and source B by zero; both are accepted. Source C differs by six hundred ninety and is rejected. Equality at the tolerance boundary is accepted because only strictly greater deviation is rejected. Corrupt ETH source changes source B to four thousand and immediately evaluates. The fresh median moves to four thousand with tolerance forty. Only source B is accepted; other deviations are five hundred and two hundred. Count drops from two to one below quorum two. Rain still passes but cannot compensate: every supplied signal must pass and both required names must exist. This is local simulation, not real action enforcement or feed authenticity proof.