Separate written duties from interpretation drift before an agent moves money. The lens preserves source text, extracts operative terms, and identifies which trigger or evidence standard changed.
Clause and independent readings
Load the clearly synthetic dispute or enter three text blocks.
Local text comparison only. No evidence verification, legal opinion, GenLayer integration, validator, wallet, payment, rejection, or external action.
EXECUTION GATE
HUMAN ADJUDICATION REQUIRED
Reading B replaces the written hash trigger and independent-verifier standard with a private quality review.
A parse can expose disagreement. It cannot prove delivery, validate evidence, determine legal meaning, or authorize machine-speed payment.
PROVENANCE-PRESERVING REDLINE
5 parsed sentences · Compromise 14.15.0
$12,000PAYMENT
2 hoursWINDOW
2 objective checksWRITTEN EVIDENCE
2 substituted termsREADING B
WRITTEN DUTY
Buyer releases $12,000.
WRITTEN TRIGGER
Delivery hash matches the signed manifest.
WRITTEN REJECTION TEST
Independent verifier records a hash mismatch.
PAYMENT TRIGGER
A: manifest hash match at 14:00 starts the two-hour clock.
B: private quality review replaces the written hash trigger.
Reading B introduces its own quality review and output-quality expectations, neither of which appears in the controlling clause. The disagreement changes when payment is due and when rejection is allowed, so automatic execution would choose an interpretation rather than merely follow text.
Autonomous agents executing smart contracts rely on exact text triggers. When natural language contracts are given to language agents, semantic drift creates conflicting interpretations. The tool parses sentences locally using natural language token extraction. In the loaded synthetic dispute, it maps the payment duty of twelve thousand dollars and extracts two written objective checks. Next, the algorithm compares normalized phrases across readings. Reading B introduces two unwritten substitutions: own quality review and output quality expectations. Because unwritten criteria change when payment is due and when rejection is permitted, the execution gate halts machine payment and demands human adjudication.
Contract Disagreement Analysis and Gatekeeper Mechanics
How does the tool distinguish between compliant execution and semantic interpretation drift in automated agent contracts?
The tool uses Compromise NLP to tokenize input contract clauses and evaluate whether alternate readings alter agreed operational conditions. In the default scenario, a controlling clause establishes an objective verification trigger (a manifest hash match) and an objective rejection threshold (an independent verifier recording a mismatch). When a party's reading introduces extraneous subjective gates—such as unwritten internal quality reviews—the script detects the ungrounded terms, flags trigger and rejection conflicts, and flips the execution gate from automated processing to mandatory human adjudication.
This demonstration uses targeted regular expressions and string matching rules tailored to the synthetic sample. It does not perform formal legal interpretation, semantic entailment across open-ended natural language, cryptographically verify data hashes, or interface with live smart contract ledgers or consensus networks.
Try a worked example
Clicking 'Load worked dispute' populates the controlling clause with a $12,000 duty contingent on a manifest hash match within 2 hours. Clicking 'Compare operative meaning' executes the local text normalization pipeline. The analyzer detects two substituted phrases in Reading B ('own quality review' and 'output quality expectations') that do not appear in the controlling clause. Because these substitutions alter the payment trigger and rejection standard, the tool opens the analysis modal, displaying 2 substituted terms and setting the execution gate to 'HUMAN ADJUDICATION REQUIRED'.
Deterministic Parsing vs. Open Semantic Drift
The implementation extracts payment obligations ($[\d,]+) and delivery timing (within \d+ hours) via regular expressions. It then normalizes clause text to identify expected objective standards ('signed manifest' and 'independent verifier'). In autonomous agent workflows, downstream financial execution requires strict alignment between textual triggers and system state.
When alternate agent readings introduce ungrounded conditions, such as subjective internal assessments, the tool flags substituted terms. If any substitution affects the trigger or rejection condition, the algorithm halts automatic execution gating to avoid unauthorized asset transfer.