Skip to content

v1.5.2

Choose a tag to compare

@github-actions github-actions released this 15 Nov 13:16
· 8 commits to main since this release

PageSift 1.5.2 Release Notes

Release Version: v1.5.2
Release Scope: Transaction, temporal, scheduling, and adaptive-observation correctness fixes on the 1.5 line (v1.5.1 -> v1.5.2)

PageSift 1.5.2 resolves seven core correctness issues covering crash-atomic storage transactions, source identity isolation, censored adaptive statistics, point-in-time historical fidelity, and scheduling boundary safety across the storage, graph, planning, sync engine, HTTP API, CLI, MCP, C FFI, Node.js, and Python layers.

Highlights

  • Crash-Atomic AtomicTx Commits (pagesift-store): Replaced contiguous plain-record batch buffering with explicit TxPut, TxDel, and TxCommit records guarded by a streaming BLAKE3 transaction digest. Mid-batch crashes, partial writes, ENOSPC/sync failures, or corrupt trailing bytes roll back the entire transaction on recovery while keeping existing 1.5.0/1.5.1 segment data readable and compactable, with u32 segment offset bounds enforced cleanly.
  • Delete-and-Recreate ABA Protection (pagesift-store, pagesift-engine, pagesift-api, pagesift, bindings): Added persistent monotonic sgen:<source_id> generation tracking and unique instanceId validation on CAS spec updates and commit_tick_atomic, preventing stale in-flight ticks from a deleted source from committing into a newly created source sharing the same id. Checked source existence consistently across delta queries (/v2/sources/{id}/deltas, CLI, MCP, Node.js, Python).
  • Censored Adaptive Learning (pagesift-plan, pagesift-engine): Classified sync fetch failures via structured ErrorCode instead of error-message matching. Transport errors (DNS failure, timeouts, connection resets), browser launch/render errors, 304, and DirectBrowser captures now leave http_sufficient / browser_gain censored (None) rather than recording false negatives, and fully censored observations no longer create empty FamilyStats entries. Hard budget exhaustion emits explicit BudgetExhausted deferral without fabricating TransientFailure or Unverified states.
  • Point-in-Time Claim History Fidelity (pagesift-graph, pagesift-store, pagesift-api, pagesift, pagesift-mcp): Stopped lossy ClaimTimelineEntry interval coalescing from merging distinct evidence/reason/metadata transitions or forging last_confirmed_at_ms at asOf query time. Added checked claim and conflict historical queries that propagate historyUnavailable across GraphStore, /v2/graph/*, pagesift graph inspect, pagesift_graph_query, C FFI, Node.js, and Python.
  • Numeric & Freshness Scheduling Bounds (pagesift-engine, pagesift-api): Eliminated compute_next_due_at_ms clamp panics when user-configured minIntervalSecs exceeds the default tombstone interval, guarded TickBudgetLedger against u64::MAX / u128 -> u64 overflow, and neutralized non-finite importance_weight inputs.
  • Pre-Creation & Legacy Inventory Point-in-Time Isolation (pagesift-store, pagesift-api, pagesift, bindings): Separated NotCreatedYet (spec: null), Complete, and HistoryUnavailable in load_spec_as_of() / build_state_view(), preventing future or stale-instance specs from leaking to earlier timestamps and fixing JavaScript safe-integer f64 timestamp coercion for asOf queries in loop_ops. Removed the time-filtered live get_inventory() fallback when checkpoints are missing or evicted, surfacing historyUnavailable and propagating corrupt/I/O errors instead of silent fallbacks.

Verification

  • Workspace checks run after the final edit: check_release.py, check_wall_clock.py, check_workflows.py, check_links.sh, check_file_length.sh, cargo fmt --all --check, cargo deny --config .github/deny.toml check, clippy -D warnings, release tests across touched Rust crates, Node.js @pagesift/node, and Python pagesift.