Role: This page controls the Matrix distributed matrix multiplication lane (WASM ABI V2). Compute tasks are leased and executed inside browser workers over HTTP via the admin server scheduler.
Operator Block Controls: Located in the Matrix Operator Configuration & Controls card below:
Reload Config (re-reads active server configuration),
Stage Config (stages form parameters in memory),
Save as Default (persists current configuration as server startup default),
Reset Staged Changes (discards pending staged modifications),
Stage & Start New Run (stages parameters and initiates execution),
Start New Run (initializes a new matrix workload run),
Pause Run (halts issuing new task leases),
Resume Run (resumes leasing tasks to workers),
Stop Run (stops active run; in-flight tasks may report),
Restart Run (stops current run and initializes new run ID with active/staged configuration),
Refresh this panel (queries latest operator state),
Archive & Clear Previous Run Data (archives run state into archive/matrix/<timestamp>/ and resets to idle),
Restart Server (restarts the pool_admin service; note: this temporarily affects both Matrix and SSPrime).
Three Configuration Columns: SAVED DEFAULT / ACTIVE displays the baseline loaded at server boot and the settings actively running; STAGED NEXT-RUN shows in-memory pending modifications; RUN displays current phase, run ID, and completed/total task progress. Staged config applies at the next Initialize or Restart Run.
Configuration Lock & Safety: Configuration is locked while a run is active (RUNNING or PAUSED); you must stop or restart the run before staging new parameters. Workloads above 1000 tasks require explicit confirmation before starting.
Archived Checkpoint Persistence: The legacy h5 Checkpoint Persistence card is retired/archived in a collapsed details block below because it is superseded by the shared Persistence panel. Its internal mode switch remains fully functional.
Shared Persistence Panel: The unified persistence panel (persistence_panel.js) above the archive section displays coordinator storage metrics, checkpoint frequency, and allows live switching between compact-adaptive, compact-immediate, and full-legacy modes.
Glossary:
leaseWant)./matrix/init to build task set and write new run ID to audit chain. Keeps active configuration; previous progress is not carried over. Use when starting a new matrix run from IDLE or STOPPED./matrix/run/pause to stop scheduler issuing new task leases. Keeps run ID, completed task counters, and in-flight tasks (which may still report). Use to temporarily suspend task dispatching./matrix/run/resume to resume leasing tasks to workers. Keeps same run ID, task counters, and evidence (refused if run is STOPPED). Use to continue a paused matrix run./matrix/run/stop to stop the run and cease leasing. Keeps counters and evidence, and in-flight tasks may still report; stopped run cannot be resumed. Use to terminate execution./matrix/run/restart to stop current run and initialize a NEW run ID with active or staged config. Resets progress under new run ID. Use to restart execution.Note: these controls drive the HTTP matrix scheduler on the admin server. They are independent of the prime/computation WebSocket coordinator on the Main Director and of the segmented sieve on the SSP Director. Starting a matrix run does not disturb either.
Workers join this lane by opening worker.html?family=matrix, or from PowerShell with mirror.ps1 open -Lanes 4 -Family matrix. There is no separate matrix worker page - the generic worker page carries the lane.
Staged in memory and applied by Start Run or Restart Run. Nothing here is written to disk - persisting operator config edits is a separate decision.
Read-only. Set by CCGSC_PROFILE at container start. To change one, edit the deploy profile and restart the stack - deliberately not something a dashboard button can do.
| Gate | Value | Source | Env var | Changeable | What it controls |
|---|---|---|---|---|---|
| loading… | |||||
A task moves queued → leased → running → reported → validated → acknowledged. failed and retryQueued rising together is normal churn; failed rising alone is not.
| Run ID | - |
|---|---|
| Config Hash | - |
| WASM ABI Version | - |
| Generator Version | - |
| Seed Derivation | - |
| Audit Schema | - |
| Uptime | - |
| Current Chain Tip | - |
|---|
The audit sequence must only ever increase. A value that resets or moves backwards is a persistence failure and is the highest-severity symptom on this page - stop the run and preserve the volume.
| Saves since server start | - |
|---|---|
| Batch interval (compact-adaptive) | - |
| Last Director/admin switch | - |
| Failed saves (retried) | - |
A switch writes one full checkpoint first, then applies at once, is kept across restarts and is audited. compact-adaptive saves accepted reports in batches (0.5-10 s): after a crash those few tasks are computed again; their results are already in the audit and task logs. Order of authority: default < settings file < environment < this switch.
| Persist mode | |
|---|---|
| Archive closed leases after (ms) | |
| Expiry sweep every (ms) | |
| Coalesce wait (ms) | |
| Archive (prune on) | |
| States | - |
| Artifact Available | - |
|---|---|
| Artifact Hash | - |
| Artifact Error | - |
| Restart restore | - |
Initialisation is fail-closed: if the qualified WASM artifacts are missing or do not match the declared ABI version, /matrix/init returns 503 rather than starting a run on unverified binaries.
| Worker ID | Tab ID | Status | Lane | Task | TTL | CPU | Browser / OS | Memory (heap) | IP | Last Seen |
|---|
SAVED DEFAULT loads at server start · ACTIVE is what the current run uses · STAGED applies on the next Initialize or Restart Run. Matrix dimensions (M, K, N) are not the number of tasks — that is Tasks This Run.
/matrix/init to build task set and write new run ID to audit chain. Keeps active configuration; previous progress is not carried over. Use when starting a new matrix run from IDLE or STOPPED./matrix/run/pause to stop scheduler issuing new task leases. Keeps run ID, completed task counters, and in-flight tasks (which may still report). Use to temporarily suspend task dispatching./matrix/run/resume to resume leasing tasks to workers. Keeps same run ID, task counters, and evidence (refused if run is STOPPED). Use to continue a paused matrix run./matrix/run/stop to stop the run and cease leasing. Keeps counters and evidence, and in-flight tasks may still report; stopped run cannot be resumed. Use to terminate execution./matrix/run/restart to stop current run and initialize a NEW run ID with active or staged config. Keeps server state; resets progress under a new run ID. Use to restart under a new run ID.PUT /matrix/config and immediately calls /matrix/init to start a fresh run. Promotes staged parameters; resets progress under a new run ID. Use to apply form edits and launch computation in one step.