v0.2.0

What's Changed

• docs: close Phase 7 after v0.1.0 by @fazpu in #134
• docs: select Phase 8 benchmark portfolio by @fazpu in #135
• feat: add guarded LoCoMo benchmark setup by @fazpu in #136
• feat: compose full pipeline and LoCoMo system benchmark by @fazpu in #137
• RS-LoCoMo-Full-v2: stronger judge, and fix local .env test contamination by @fazpu in #138
• Make Selection's verdict and drop reason one value by @fazpu in #140
• Retry empty completions once and say why a completion was unusable by @fazpu in #141
• Remove unverified claims from binding documents and the retry built on one by @fazpu in #142
• Render the required array in canonical order, not declaration order by @fazpu in #143
• RS-LoCoMo-Full-v3: strict-representable agent step by @fazpu in #145
• Extract and persist claim valid-time (D41) from Claimify by @fazpu in #151
• Recipe ergonomics v4: hydrated hybrid RRF, tool guidance, agent loop guards by @fazpu in #152
• Measurement hardening: deterministic extraction, bounded transcripts, per-model reasoning effort (v5) by @fazpu in #162
• D79: deterministic structure skeleton + bottom-up, consumed summaries (design) by @fazpu in #164
• Claimify-loss ledger: no kept span dies silently (#161) by @fazpu in #166
• D79 addition: skeleton sanity check — stats-informed flash judge on the parser path by @fazpu in #167
• D79 Wave 1: the structure stage — deterministic skeleton, sanity check, anchor fallback (#165) by @fazpu in #169
• D79 Wave 2: bottom-up summaries + placement (#165) by @fazpu in #170
• D79 Wave 3: summaries consumed as orientation — closes #165 by @fazpu in #171
• fix: explicit max_completion_tokens on OpenRouter chat calls by @fazpu in #172
• fix: D32 layer-2 grounds against the union of source-derived texts by @fazpu in #173
• feat: opt-in observability — Sentry-protocol errors + Langfuse eval traces by @fazpu in #175
• benchmark: RS-LoCoMo-Full-v5-strong — the strong-agent protocol variant by @fazpu in #176
• fix(locomo): bounded retry on invalid reader completions + protocol-pinned reasoning effort by @fazpu in #178
• fix(e2): ground relative claim times against in-document session anchors (#158) by @fazpu in #179
• fix(e2): token-tolerant D32 union grounding — stop rejecting prompt-mandated attribution scaffolding by @fazpu in #180
• fix(e2): temporal resolution applies to quoted/attributed claim forms (#158 follow-up) by @fazpu in #181
• release: prepare RememberStack 0.2.0 by @fazpu in #182

Full Changelog: v0.1.0...v0.2.0