Rediscover and compare useful mechanisms

Open Praxis Mine to survey what you have mined, recover why a mechanism mattered, or choose the next experiment. Decision field opens first in both environments: it groups findings by recorded disposition, puts each next falsifiable test beside its mechanism and source, and states which findings lack a current material-bound evaluation. These are categories, not stages or automatic recommendations. Source gallery in Blackglass — Crimson and Evidence circuit in SNAPFRAME let you follow the source trail. Both open full evidence in a focused reader and keep a comparison bench available across the collection.

Open your collection

Extract the complete customer ZIP. With Python 3.11 or newer available, double-click Open.cmd on Windows. On macOS, run python3 codex/praxis-mine/skills/praxis-mine/workspace/open.py from the extracted folder; Open.command runs the same action after chmod +x Open.command. Standalone skill packages include their own launchers.

The launcher confirms the matching local service, reconnects an existing instance, finds a free loopback port after a collision, and starts again after reboot. Runtime changes receive a distinct local service identity; a matching path alone cannot reuse an older implementation. The room uses PRAXIS_MINE_DATA_HOME when set, otherwise .praxis-mine/data in your user folder. Local ledger reveals the active location. Your agent uses the same SQLite store through scripts/praxis_mine.py.

If earlier work is missing, confirm its data location before recording it again. Launch with --data-root "your folder" to open an existing native data home; the CLI calls this option --data-home. Opening the original data home retains its full recorded history. Import native manifest brings source and candidate content into the selected collection and creates new run records; it does not import evaluations, revision histories, or existing runs. The package contains no private research history and does not silently import another collection.

Choose how to explore

Use Environment to choose between two presentations of the same records. The browser remembers your choice. Preferences from the retired environments reset to Blackglass — Crimson. In Decision field, the four measures describe only the findings currently in view: findings, linked sources, recorded next tests, and findings without a native evaluation. Each category shows its first three findings in the selected order; use Show more to reveal the rest. Search and disposition filters narrow every view.

Blackglass — Crimson uses near-black glass, neutral illuminated rims, soft shadows, and clean red interaction signals. Its Source gallery groups findings beneath their actual source. Each card exposes the useful mechanism and the next recorded test before you open it. The filter deck stays beside the gallery on wide screens; the gallery uses the remaining width instead of reserving space for an always-open dossier. Registered sources without findings remain available as source records.

SNAPFRAME uses precise decision nodes and branching circuit lines on a dark field. In Evidence circuit, choose a source from the horizontal source rail. The circuit connects that recorded source to its findings and each finding’s latest current evaluation and next test. Select a source, mechanism, or judgment node to read the corresponding native record. A registered source with no findings has an explicit empty state; no relationship is invented to complete a diagram.

A finding without a current evaluation may have historical evidence. Material changes to its source, locator, content, license or summary make old evaluations historical. Evaluations from older releases without a material snapshot also need reassessment. The circuit shows triage until a current judgment exists. Its latest-judgment view is an overview; the reader preserves the complete evaluation history.

Narrow the collection and follow its history

Use Find a mechanism or source, then select a recorded disposition to narrow the results. Dispositions are categories, not stages every finding must pass through. On small screens, Filter collection expands the disposition and order controls; search and view selection stay visible. Order sorts recent activity, earliest activity, or titles. The count shows how many findings match. Show the whole collection clears an empty search result and its filters.

Choose Activity to browse dated sources, findings, runs, and evaluations. Its Show control can focus on one record kind. Open an event to inspect it, then return to Decision field, Source gallery, or Evidence circuit. Refresh retrieves changes made by your agent or another window.

Read a finding without losing the collection

Select a card or circuit node to open the reader. It brings the mechanism, source, recorded evidence, risks, judgments, and next falsifiable test into focus. Source links and Recorded connections follow native keys. They are recorded relationships, not endorsements.

Use Previous and Next to move between findings in the current filtered collection. Back to collection, Close finding, or Escape closes the reader. Selection is retained so you can resume at the same finding. On wide screens, the reading content can spread into columns; on smaller screens it stacks.

Open Revision history for preserved native versions. Inspect complete native record exposes the full record and provenance. An absent evaluation is stated explicitly. Priority scores guide attention; they do not establish fitness or authorize adoption.

Compare two or three findings

Use + To bench on a gallery card or + Compare in a circuit or reader. The fixed comparison bench keeps the selected findings visible while you explore. Add at least two, then select Compare. The comparison aligns mechanism, capability delta, next falsifiable test, why you cared, possible use, evidence state, evidence, costs and risks, and source. Missing material reads Not recorded.

Remove a finding with its bench chip or Remove in the comparison. The bench holds at most three findings. Close comparison or Escape returns to the collection. Selected findings and bench choices are remembered in this browser for the active ledger; they do not alter recorded judgments.

Record, revise, and hand off

Ledger tools contains Record source, Record candidate, Import native manifest, Export ledger, and Export report. Begin with a source so each finding has provenance. Record its locator and access posture, then the mechanism, why you cared, applicability, and project connection. These are native candidate tags available to your agent.

Edit record updates a source or candidate; recorded runs allow explanatory notes only. Save record checks the expected revision inside a SQLite transaction. If another editor changed it, preserve your draft, close the editor, reopen the current record, and reconcile before saving again. Unknown native fields remain intact. A candidate’s source identity stays fixed to preserve its recorded relationship.

Record evaluation uses the native scoring and disposition gate. Enter the capability delta, evidence, next test/action, explicit benefit and risk scores, and decision basis. Name the baseline, comparison and limitations, rights status and basis, and reason for the choice. Owner and authorized action scope can be recorded when known. Unknown risk is not zero; risk inputs begin blank. Reject may omit a future test. Editing cannot bypass this gate or silently select adopt. An evaluation is superseded by a new evaluation rather than overwritten.

Project handoff downloads the finding, current assessment, evidence references, baseline, rights basis, limits and next action without the private data-home path. Check the recipient may see the source and project details before sharing. It neither installs nor executes the candidate. Export ledger downloads versioned JSONL with records, relations, events, and audit rows; Export report downloads the Markdown report. Back up the separate data home. Updating application files does not reset it.

Import and recovery

Manifest import validates the complete request and commits its source, candidate, relationship and run records together. Scans and evaluation changes also commit as units. A normal failed transaction leaves no partial record set; correct the error and retry. Temporary request transport is removed; normalized payload versions and the digest of supplied request bytes remain in provenance. Preserve your original input file separately. This is not a guarantee against every power-loss or filesystem failure.

If the room cannot load, use Refresh. Reopen through the launcher if its session expired. If launch fails, run the same command with --serve to see the error. The room binds only to 127.0.0.1, requires its launch token, and rejects foreign Host and Origin access. Keep the tokenised Open URL private.

The workspace has no cloud dependency and does not browse or execute candidates. Arm’s Reach remains dormant. Desktop shortcuts require your request. macOS launchers are supplied; native macOS execution has not been verified.

Diagnose a held or stale assessment

An explicit Reject or Monitor is never upgraded because the score is high. An actionable recommendation lacking comparison or rights evidence becomes Monitor, with its missing basis shown. A recorded LICENSE filename is insufficient; edit the candidate license conclusion and cite the actual basis. The gate records declared evidence, not independently verified facts. A changed material snapshot shows historical evaluations in the reader and triage in the decision field, circuit and report. Reassess the current material; never copy an old success label forward.

For an optional product-icon Windows shortcut or portable Mac app entry, use Add a Praxis Mine entry. Creation is explicit and refuses an existing entry; it does not capture a prior –data-root choice or modify your SQLite home.