iranti 0.2.2 → 0.2.3

package/README.md

@@ -9,10 +9,10 @@
 
 Iranti gives agents persistent, identity-based memory. Facts written by one agent are retrievable by any other agent through exact entity+key lookup. Iranti also supports hybrid search (lexical + vector) when exact keys are unknown. Memory persists across sessions and survives context window limits.
 
-**Latest release:** [`v0.2.0`](https://github.com/nfemmanuel/iranti/releases/tag/v0.2.0)  
+**Latest release:** [`v0.2.2`](https://github.com/nfemmanuel/iranti/releases/tag/v0.2.2)  
 Published packages:
-- `iranti@0.2.0`
-- `@iranti/sdk@0.2.0`
+- `iranti@0.2.2`
+- `@iranti/sdk@0.2.2`
 
 ---
 
@@ -31,6 +31,7 @@ Iranti is a knowledge base for multi-agent systems. The primary read path is ide
 - **Library**: Active truth store (`knowledge_base`) in PostgreSQL.
 - **Archive**: Historical/superseded truth store (`archive`) in PostgreSQL.
 - **Archivist**: Maintenance worker that archives stale/low-confidence facts and processes resolved escalations.
+- **Resolutionist**: Interactive CLI reviewer that guides humans through pending escalation files and writes valid authoritative resolutions.
 
 ---
 
@@ -82,10 +83,10 @@ Iranti now also has an adversarial conflict benchmark that measures contradictio
 | Suite | Score | Notes |
 |---|---|---|
 | **Direct contradiction** | `4/4` | Same entity+key conflicts are explicitly resolved or escalated |
-| **Temporal conflict** | `3/4` | One known-failing edge remains |
-| **Cascading conflict** | `0/4` | Cross-key contradiction detection not implemented yet |
-| **Multi-hop conflict** | `0/4` | Graph-aware conflict reasoning not implemented yet |
-| **Total** | `7/16 (44%)` | Honest baseline for the current Librarian |
+| **Temporal conflict** | `4/4` | Equal-score ties now use deterministic temporal tie-breaks |
+| **Cascading conflict** | `4/4` | Deterministic same-entity cross-key contradiction checks |
+| **Multi-hop conflict** | `4/4` | Narrow relationship-aware contradiction checks across related entities |
+| **Total** | `16/16 (100%)` | Current benchmark coverage for the Librarian |
 
 Conflict benchmark methodology: [`docs/internal/conflict_benchmark.md`](docs/internal/conflict_benchmark.md)
 
@@ -212,7 +213,7 @@ The current landscape splits into three buckets:
 Iranti is strongest today as infrastructure for developers building multi-agent systems who need shared, structured, queryable memory rather than pure semantic recall. The current evidence base is now more concrete than a positioning claim alone:
 
 - `16/16` fictional-fact transfer in retrieval validation
-- `7/16 (44%)` on an adversarial conflict benchmark
+- `16/16 (100%)` on the current adversarial conflict benchmark
 - `4/4` on empirical consistency validation for serialized writes and read visibility
 
 That is not a claim that multi-agent memory is solved. It is a claim that Iranti now has reproducible evidence for three things at once:
@@ -292,6 +293,7 @@ Security quickstart: [`docs/guides/security-quickstart.md`](docs/guides/security
 Claude Code guide: [`docs/guides/claude-code.md`](docs/guides/claude-code.md)
 Codex guide: [`docs/guides/codex.md`](docs/guides/codex.md)
 Release guide: [`docs/guides/releasing.md`](docs/guides/releasing.md)
+Vector backend guide: [`docs/guides/vector-backends.md`](docs/guides/vector-backends.md)
 
 ### Claude Code via MCP
 
@@ -318,6 +320,28 @@ When `iranti codex-setup` is run from a project directory, it automatically capt
 
 Guide: [`docs/guides/codex.md`](docs/guides/codex.md)
 
+### Resolve Pending Escalations
+
+Review unresolved human-escalation files from the CLI:
+
+```bash
+iranti resolve
+```
+
+Use `--dir` to point at a non-default escalation root. Guide: [`docs/guides/conflict-resolution.md`](docs/guides/conflict-resolution.md)
+
+### Native Chat
+
+Start a CLI chat session against the configured Iranti instance:
+
+```bash
+iranti chat
+```
+
+Use `--agent`, `--provider`, and `--model` to pin the session identity and model routing.
+The chat surface now includes slash commands for fact history, relationships, conflict-resolution handoff, and confidence updates in addition to memory search/write operations.
+Guide: [`docs/guides/chat.md`](docs/guides/chat.md)
+
 ---
 
 ## Install Strategy (Double Layer)
@@ -438,12 +462,14 @@ Use the CLI doctor command before first run or before a release check:
 iranti doctor
 iranti doctor --instance local
 iranti status
-iranti upgrade
+iranti upgrade --check
+iranti upgrade --dry-run
+iranti upgrade --yes
 ```
 
 This validates the active env file, database URL, API key presence, provider selection, and provider-specific credentials.
 `iranti status` shows the current runtime root, known instances, and local binding files.
-`iranti upgrade` prints the supported upgrade commands for npm and Python installs.
+`iranti upgrade` detects repo/global/Python install paths, compares current vs latest published versions, prints the exact plan, and executes the selected upgrade path when you pass `--yes`.
 `iranti configure ...` updates instance/project credentials without manual env editing.
 `iranti auth ...` manages registry-backed API keys and can sync them into instance or project bindings.
 
@@ -687,7 +713,7 @@ middleware.after_receive(
 
 ## Architecture
 
-Iranti has four internal components:
+Iranti has five internal components:
 
 | Component | Role |
 |---|---|
@@ -695,6 +721,7 @@ Iranti has four internal components:
 | **Librarian** | Manages all writes. Detects conflicts, reasons about resolution, escalates when uncertain. |
 | **Attendant** | Per-agent working memory manager. Implements `attend()`, `observe()`, and `handshake()` APIs. |
 | **Archivist** | Periodic cleanup. Archives expired and low-confidence entries. Processes human-resolved conflicts. |
+| **Resolutionist** | Interactive CLI helper that walks pending escalation files, writes `AUTHORITATIVE_JSON`, and marks them resolved for the Archivist. |
 
 ### REST API