About this application

Promise Ledger is a working account-memory demo over the fictional Acme archive supplied for the RELEX challenge. It combines reviewed, source-linked findings with local full-text search. It does not use a generative language model.

Sources and findings

All 45 documents are stored with stable line positions. Findings are human-authored annotations of the supplied archive, with explicit dates, scope and status. Receipts take you directly to the underlying passage. Unseen questions use lexical retrieval; matching a passage does not establish an answer.

Review cases

At import, the application creates six review cases from annotated evidence: partial completion, a mistaken agreement, old data copies, scope, acceptance and verification gaps. These appear without asking a question. Marking one reviewed only acknowledges it; it cannot certify completion. There is no general-purpose autonomous commitment extractor.

Deletion changes stored memory

Every browser gets an isolated SQLite workspace. Removing an identity purges attributed messages and turns, lines matching reviewed aliases, dependent findings and cases, saved answers and search entries. The search index is rebuilt from surviving text. SQLite secure deletion and compaction remove old database page contents. No embeddings, external inference caches or model training exist.

The supplied archive and an immutable baseline demo fixture are retained for new workspaces. They are outside the active workspace deletion boundary. The question engine never reads them to answer an existing workspace. Other browser workspaces and copied review packets are outside that boundary. Anonymous references and unknown aliases can escape matching; deleting a person is not a legal-compliance guarantee.

Residency

All application storage and computation run on this host. No corpus is sent to an external inference service. The host’s physical EU location must be verified with the infrastructure owner before making an EU-residency certification. Browser font assets are served locally.

Deployment limits

This public deployment uses fictional data and browser-isolated workspaces. There is no production user authentication or organization-level authorization. A production deployment needs those controls, a retention policy and an expanded identity review workflow. The exact dependencies, startup command, fixed project port and test instructions are in the repository documentation.