Projects, recovery, and checkpoints

An .equarith project is the normal way to preserve a working session. In v1.0.3 it is a versioned ZIP-based archive, but it should be opened and saved through Equarith rather than edited with an archive utility.

What a project stores

A project can contain:

• project identity and document metadata;
• an embedded dataset or a reference to an external source file;
• selected target and inputs;
• search settings, functions, and complexity costs;
• workspace layout and relevant presentation state;
• published Pareto solutions; and
• an optional compatible search checkpoint.

Project data, results, and checkpoints remain local files. Saving does not upload them.

Embedded versus referenced data

When an unchanged imported file is available, Equarith can ask whether to embed it or keep a reference.

Embedded data makes the project self-contained and portable, but increases the archive size. It is the safest choice when sending the project to another computer or when the source file may move.

Referenced data keeps the project smaller. Equarith stores the expected location, schema, and fingerprint. When reopening the project it verifies that the source still matches. If the file is missing or changed, Equarith warns instead of silently associating saved solutions with different observations. The affected dataset and solutions are not treated as valid until the problem is resolved.

Edited, pasted, and generated datasets are embedded because no unchanged external source represents their current contents.

Safe saving and dirty state

Equarith writes and closes a complete neighboring temporary file, then replaces the destination atomically when the filesystem supports it, with a regular replacement fallback. This pattern reduces the chance of leaving a partially written project.

Saving is asynchronous. If you edit the document again while a save is finishing, Equarith does not incorrectly mark the newer revision as saved. Watch the document state before closing the application.

Use Save As before making a risky experiment when you want an explicit branch. Normal filesystem backups and versioned copies remain valuable; project recovery is not a substitute for backup.

Autosave and crash recovery

Equarith keeps a local active-session marker and recovery archives under the application-data directory. After an abnormal exit it can offer to restore recoverable work. Autosave frequency is configurable from 0 to 120 minutes; zero disables scheduled autosave.

Recovery files are a safety net:

• they can lag behind the latest edit;
• forced process termination may interrupt the newest write;
• they are kept on the same machine unless you back them up; and
• they should not be your only copy of important research.

After recovery, inspect the data, configuration, and solution list, then save a named project.

Search checkpoints

A checkpoint captures enough engine state to continue a compatible unfinished search. Equarith creates checkpoints automatically during search and requests one during a normal stop or restart when possible.

When a project with an unfinished compatible checkpoint is opened, the next Start can resume it. A completed search is not resumed. Compatibility includes the dataset fingerprint, split and objective, constraints, function selection and costs, engine configuration, and format version. Changing any of these can deliberately invalidate the checkpoint, in which case a fresh search starts.

Stop is cooperative first. If workers cannot finish within the bounded shutdown period, Equarith can interrupt them; the very latest checkpoint may then be unavailable. Previously published solutions remain visible.

Do not move a standalone checkpoint between unrelated requests and do not alter its files. Equarith rejects truncated, oversized, inconsistent, unsafe, or future-format archive content.

Exchanging configurations and seeds

Search settings can be imported from or exported to versioned JSON. This is useful for reviewing configurations and repeating them with another dataset, but it does not embed the dataset itself.

Seed formulas use UTF-8 plain text with one expression per line. Blank lines and lines beginning with # are ignored. Each other line is parsed independently, so valid lines can still load when another line is reported as invalid. A seed is a starting suggestion; the search is not required to retain it.

If you need unattended operation and standalone checkpoints, see Automation, headless CLI, and SDK roadmap. For format-related failures, see Troubleshooting and known limitations.