Abstract
Intentrail is a local-first mission validation workspace for synthetic robotics scenarios. It keeps a declared mission, a finite command sequence, the resulting trace, and the validation outcome in one portable record.
The project addresses a narrow handoff problem: a route specification and its reported outcome can become separated. Intentrail makes the computation repeatable. A second browser can import a receipt, rerun the embedded mission under the same validator rules, and compare canonical SHA-256 digests.
This protocol evaluates discrete data. It does not operate a robot, model physical dynamics, authenticate sensors, prove location, or certify safety.
The handoff problem
A mission is more than a final coordinate. Ordered checkpoints, obstacles, stop limits, and failed attempts all contribute to the meaning of a run. A screenshot of a passing state cannot reveal whether the route was replayable or whether intermediate conditions were preserved.
Intentrail treats the mission specification and trace as one validation object. The user can inspect every command and stop reason before exporting evidence. No hosted account is required to create or verify that object.
Mission model
A mission manifest contains a rectangular grid, one start, one goal, zero or more ordered checkpoints, obstacle cells, a maximum-step constraint, and a finite sequence of cardinal commands or waits.
Grid, markers, constraints, and commands are normalized before a manifest digest is created.
Each command maps exactly one prior position to one attempted and resulting position.
Collision, boundary exit, exhausted commands, and step limits remain visible outcomes.
The local route helper can prepare a path through visible cells. It is a deterministic convenience, not an autonomous system and not a claim about physical feasibility.
Deterministic validation
The validator starts from the declared start cell and applies commands in order. A collision or out-of-bounds attempt stops the run immediately. Checkpoints count only in their declared order. A run completes only after all checkpoints and the goal are reached within the step limit.
Trajectory animation is downstream of the trace. It can be paused or replayed, but visual timing never influences computation. Path efficiency compares executed steps with an obstacle-agnostic Manhattan baseline only for completed runs. The displayed constraint score is a local collision/boundary heuristic, not a safety metric.
Manifest and run receipts
The mission manifest digest covers normalized intent. The run receipt digest covers the complete receipt body excluding the digest field itself. Canonical JSON sorts object keys, rejects ambiguous values, and gives an independent verifier a stable byte sequence for hashing.
A valid receipt establishes that its embedded mission recomputes to the recorded trace and result under this validator version, and that the canonical payload has not changed since the digest was created.
Physical execution, sensor truth, operator identity, trusted time, hardware ownership, route safety, regulatory compliance, external endorsement, or reward eligibility.
Web3 evidence layer
The Web3 workspace keeps connection, reads, local points, optional payment review, and anchor preparation as separate actions. Connecting reads a user-selected public address and active network identifier. It awards nothing and requests no signature or transaction.
A verified run receipt can be reduced to digest references inside a chain-neutral anchor envelope. The current envelope is unsigned and not submitted. Signature and anchor fields remain empty. A read-only inspector can examine an actual transaction and test whether exact digest bytes occur in transaction input, but byte presence alone does not prove the transaction's purpose.
After a passed receipt is imported, a user may separately activate 100 browser-local points scoped to the current address and network. An optional 0.21 BOT upgrade is review-only in this release: recipient and target network are unavailable, confirmation is disabled, and no broadcast path exists.
Ecosystem contribution
Intentrail contributes small, reusable building blocks for mission-evidence workflows. Builders can produce compatible records, verify them without this interface, or adapt the read-only inspection boundary.
The project is independent. Linking to broader ecosystem channels documents participation context and does not imply affiliation, integration, approval, or endorsement.
Limits and future work
This release uses a discrete 2D model and browser memory. It does not include physics, sensor feeds, fleet orchestration, cloud storage, hardware credentials, autonomous model inference, or a deployed anchoring contract.
Future work may improve schema interoperability, verifier ergonomics, and independently reviewed anchor integrations. Any real write path would require separately verified configuration, network checks, a complete recipient and fee review, explicit user confirmation, and new testing. Those steps are not implied by this release.