v3.5.0

What's Changed

• docs(type-aware-compression): refresh stale retrieval.py line refs by @robotrocketscience in #902
• docs(philosophy): drop stale v2.x conditional on filter-bubble exploration by @robotrocketscience in #904
• test(promotion): refresh stale #616-merge skip-reason text by @robotrocketscience in #903
• docs(readme): bump 'Latest stable' from v3.0.1 to v3.3.0 by @robotrocketscience in #905
• exp(bench): MAB LRU back-fill — n=171 3-run SEM=24.2% F1=37.2% (#899) by @robotrocketscience in #900
• docs(bench): land orphan LoCoMo recalibration paragraph from PR #896 by @robotrocketscience in #906
• exp(bench): MAB Conflict_Resolution back-fill (800q) — SEM 37.1% F1 38.2% (#899) by @robotrocketscience in #907
• exp(bench): MAB Test_Time_Learning back-fill — n=700, not measurable by @robotrocketscience in #908
• exp(bench): MAB Accurate_Retrieval back-fill — n=2000 split-rep SEM=79.6% F1=84.9% (#899) by @robotrocketscience in #913
• exp(bench): StructMemEval per-sub-task back-fill — 90.5%/15.4%/0%/0% spread (#899) by @robotrocketscience in #912
• fix(hook): conversation-aware UserPromptSubmit retrieval by @robotrocketscience in #911
• fix(bench): task-aware StructMemEval prediction scorers by @robotrocketscience in #914
• fix(bench): settlement-tuple scorer for StructMemEval accounting by @robotrocketscience in #917
• feat(bench): reader-prediction scoring seam for StructMemEval by @robotrocketscience in #918
• docs(bench): correction banner on StructMemEval back-fill numbers by @robotrocketscience in #916
• exp(bench): StructMemEval answer-correctness capture — canonical judge by @yoshi280 in #919
• chore: gitignore hook_audit.jsonl to prevent accidental commit by @yoshi280 in #921
• fix(hook): skip audit write for :memory: DB to stop CWD pollution by @yoshi280 in #922
• feat(cadence): p3_substantive dispatch + shadow capture + 4-policy scorer + offline replay (#876) by @yoshi280 in #920
• fix(setup): refuse worktree-pathed venv in hook resolvers (#928) by @yoshi280 in #939
• docs(latency): reconcile stack-r1-r3 p99 claim across README + ROADMAP by @yoshi280 in #926
• feat(search): AELF_SHOW_CONFLICTS — [!] slot-conflict column on aelf search (#938) by @robotrocketscience in #946
• feat(cli): aelf:stale subcommand — surface beliefs by age + retrieval recency (#933) by @yoshi280 in #942
• chore(deps): bump aiohttp from 3.13.5 to 3.14.0 in the uv group across 1 directory by @dependabot[bot] in #927
• feat(observability): .aelfrice-feed.jsonl write-event log + aelf:feed reader (#931) by @yoshi280 in #943
• feat(statusline): L0/L1/? count badges (#932) by @yoshi280 in #940
• feat(hooks): aelf-pre-issue-create guard — duplicate-detection before gh issue create (#941) by @robotrocketscience in #945
• chore(deps): bump starlette from 1.0.0 to 1.0.1 in the uv group across 1 directory by @dependabot[bot] in #949
• docs: full audit + factual correction sweep + marketing polish (130 files) by @yoshi280 in #923
• feat(review): weekly .aelfrice/review.md checkbox file + aelf:review --apply (#936) by @robotrocketscience in #948
• feat(cli): aelf audit-claude-memory — cross-store dedup audit (#935) by @robotrocketscience in #947
• feat(cli): aelf:speculative — list non-user-locked beliefs (#937) by @robotrocketscience in #944
• feat(hook): SessionStart belief-write recap injection (#934) by @robotrocketscience in #950
• release: v3.5.0 by @robotrocketscience in #951

Full Changelog: v3.3.0...v3.5.0