This commit implements comprehensive production hardening across multiple
layers to prepare StemeDB for enterprise pilot deployments:
## API Layer
- Add rate limiting middleware with configurable limits per endpoint
- Enhance error handling with detailed context and proper HTTP status codes
- Add security hardening tests for input validation and boundary conditions
- Create store_helpers module for defensive storage access patterns
## Storage & WAL
- Optimize group commit batching for higher throughput
- Add defensive error handling in hybrid backend with proper fallbacks
- Enhance WAL journal durability guarantees with fsync validation
- Improve index store query performance with better caching
## Operations & Deployment
- Add comprehensive operations documentation (deployment, monitoring, DR)
- Create systemd units for backup, WAL archival, and verification
- Add monitoring configs (Prometheus alerts, metrics exporters)
- Implement backup/restore scripts with verification and S3 archival
- Add DR drill automation and runbook procedures
- Create load balancer configs (nginx, envoy) with health checks
## Documentation
- Update CLAUDE.md with operations and troubleshooting guides
- Expand roadmap with production readiness milestones
- Add pilot success criteria and deployment reference architecture
- Document TLS setup, monitoring integration, and incident response
## Configuration
- Add .env.example with all required environment variables
- Document resource sizing for different deployment scales
- Add configuration examples for various deployment topologies
This positions StemeDB for successful enterprise pilots with proper
operational discipline, monitoring, backup/DR, and security hardening.
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
## Problem
CLI-created community corpus items (tier 3) were stored correctly but
invisible via API queries. Two issues blocked discoverability:
1. **Prefix mismatch**: API hardcoded 'community://pattern/' for
aggregated patterns, but CLI creates 'community://rust/http/...' URIs
2. **Query parameter parsing**: Axum's default parser doesn't support
bracket notation (?sources[]=value) used by the dashboard
Result: 0/22 CLI-created items were queryable.
## Solution
### Fix 1: Broaden Community Prefix
- Changed: 'community://pattern/' → 'community://' in corpus handler
- Impact: Now matches both aggregated patterns AND CLI-created items
- Backward compatible: Broader prefix includes narrower results
### Fix 2: Add QsQuery Extractor
- Added: serde_qs dependency + custom QsQuery extractor
- Supports: Bracket notation for array parameters (?sources[]=a&sources[]=b)
- Compatible: Works with JavaScript URLSearchParams standard
- Tested: 3 new unit tests for extractor behavior
## Verification
- ✅ All 22 CLI-created community items now queryable (was 0)
- ✅ Source filtering works: community (22), RFC (2), vendor (5)
- ✅ Multi-source queries work: ?sources[]=community&sources[]=rfc → 24
- ✅ All 89 API tests pass + 3 new extractor tests
- ✅ Clippy clean (0 warnings)
- ✅ No regressions in existing functionality
## Files Changed
- crates/stemedb-api/Cargo.toml: Add serde_qs dependency
- crates/stemedb-api/src/extractors.rs: New QsQuery extractor (117 lines)
- crates/stemedb-api/src/handlers/aphoria/corpus.rs: Use QsQuery, broaden prefix
- crates/stemedb-api/src/lib.rs: Export extractors module
Also includes: Scale-adaptive thresholds, wiki corpus extraction,
documentation updates, and dashboard UI improvements from prior work.
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
Add CRC32C checksums to WAL record format (v2), implement crash recovery
with automatic truncation of corrupt records, add feature-gated group commit
buffer for batched fsync under concurrent load, and implement log rotation
via segment files with global offset addressing.
Key changes:
- Record format v2: [len:u32][crc32c:u32][blake3:32][payload:N]
- recover_file() scans and truncates corrupt tail records
- GroupCommitBuffer batches fsync via MPSC channel (tokio feature gate)
- SegmentManager with binary search resolution and cursor-based cleanup
- Journal::read() auto-refreshes segments on miss for writer/reader split
- Split recovery.rs and key_codec.rs into directory modules for 500-line max
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add Layered() method to Go SDK for per-source-class consensus queries
- Add LayeredQueryParams, LayeredResult, TierResolution types to Go SDK
- Create conflict example demonstrating Skeptic and Layered endpoints
- Update quickstart.md with sections 6 (conflict detection) and 7 (authority tiers)
- Remove tracked Go binary and add data/ to .gitignore
The new quickstart sections demonstrate Episteme's differentiating features:
- Skeptic endpoint shows "Trust but Verify" conflict analysis
- Layered endpoint shows per-tier resolution (Clinical vs Anecdotal)
Note: Pre-existing large files flagged by pre-commit hook (technical debt from prior sessions)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
