TunnelVision

TunnelVision 0.5.0

This release adds terrain exposure: the plugin now knows whether an ore was visible in open air before it was broken. That's what makes it possible to tell a player exploring a cave apart from one digging straight at an invisible vein — the main source of false positives so far.

Terrain exposure (new)

For every ore broken, the plugin re-reads the actual terrain through the Bukkit API and determines whether the block was exposed to air just before it was mined. This feeds a new scoring signal, blind_target_rate: the share of veins hit without a single face having been visible beforehand.

• Enabled by default, under the exposure: section in config.yml.
• Runs on an async thread — the server tick is never blocked.
• Automatic history backfill. Two cursors work in parallel with nothing to configure: a forward cursor for recent breaks, and a backward cursor that gradually works through your existing CoreProtect history. Installing the plugin on a server that's been running for months therefore doesn't leave older sessions unanalyzed.
• The backward pass resumes where it left off after a restart.
• The /exposure/coverage endpoint reports how far the backfill has progressed over a given time window.
• Fully optional: disable the section and sessions are scored as before — nothing breaks.

Cave sessions: far fewer false positives

Sessions taking place in natural caves are now detected and tagged automatically, then evaluated with dedicated scoring instead of the general one. A player exploring a cave legitimately finds a lot of already-exposed ore — they're no longer penalized for it. Reports flag these sessions explicitly.

Performance

• Exposure backfill is significantly faster: the ore filter was pushed down into SQL, neighbor block history is now cached, and the scan window is bounded (scan-window-rowids, roughly 62 ms per million rowids).
• Sync to the analysis service is 2.5× faster thanks to a page cache on the mirror side.
• Configurable sync batch size via max-page-size. Measured over a 61 M block pass: ~12,500 blocks/s, with 96% of the time spent sending over the network and 4% reading locally. Raising the value cuts the number of HTTP round trips proportionally, at the cost of more memory per batch.
• Chunks are now preloaded in parallel during terrain reads.
• Lower memory use when loading break data.

3D reports

• Ore selection with color coding.
• Format adjusted to surface cave sessions.

Fixes

• The backward backfill stopped after a single batch instead of continuing.
• The backward backfill ignored terrain exposure.
• Application memory thresholds now follow the actual allocated limit.

Upgrade notes

No action required: the new settings ship with working defaults and the history backfill starts on its own. The plugin still never writes to the CoreProtect file.