docs: re-measure every figure at 0.7.4

.claude-plugin/marketplace.json

@@ -10,7 +10,7 @@
       "name": "agent-memory",
       "source": "./",
       "description": "Durable cross-repository memory for coding agents — built to survive an IT security review. Zero runtime dependencies, zero dev dependencies, no install script: nothing runs when you install it, and granting it your agents is a separate explicit command. Independently scanned, with a passing verdict. Adds /handoff, /remember and /recall over one local markdown store. Requires the CLI: npm install -g @vib795/agent-memory (Node >= 22.5).",
-      "version": "0.7.3",
+      "version": "0.7.4",
       "author": {
         "name": "Utkarsh Singh",
         "url": "https://github.com/vib795"

.claude-plugin/plugin.json

@@ -1,7 +1,7 @@
 {
   "name": "agent-memory",
   "displayName": "agent-memory",
-  "version": "0.7.3",
+  "version": "0.7.4",
   "description": "Durable cross-repository memory for coding agents — built to survive an IT security review. Zero runtime dependencies, zero dev dependencies, no install script: nothing runs when you install it, and granting it your agents is a separate explicit command. Independently scanned, with a passing verdict. Adds /handoff, /remember and /recall over one local markdown store. Requires the CLI: npm install -g @vib795/agent-memory (Node >= 22.5).",
   "author": {
     "name": "Utkarsh Singh",

ARCHITECTURE.md

@@ -3,8 +3,8 @@
 How agent-memory is built, and why it is built that way.
 
 Every figure here was read from the source rather than written from memory, at
-v0.6.4: 15 modules, 3,574 lines of JavaScript, zero runtime dependencies and zero
-dev dependencies, 97 tests.
+v0.7.4: 15 modules, 4,168 lines of JavaScript, zero runtime dependencies and zero
+dev dependencies, 111 tests.
 
 ---
 
@@ -242,7 +242,7 @@ quietly ending.
 
 ---
 
-## Two-tier routing
+## Three-tier routing
 
 The expensive resource is not disk or CPU. It is the model's context window, and the
 premium request budget behind it.
@@ -472,22 +472,22 @@ The properties that must stay true. Each is covered by the test suite.
 
 | Module | Lines | Responsibility |
 |---|--:|---|
-| `cli.js` | 980 | 13 commands; one process, one answer |
+| `cli.js` | 1114 | 14 commands; one process, one answer |
+| `digest.js` | 559 | tiers 1 and 2: both descriptions, tree, brief |
+| `compact.js` | 372 | dedupe, supersede, decay, regenerate |
 | `index-db.js` | 317 | SQLite cache: DDL, reindex, FTS5 search |
+| `setup.js` | 297 | install into detected agent surfaces |
 | `pii.js` | 274 | export-time disclosure control |
 | `store.js` | 256 | notes as source of truth; frontmatter; atomic upsert |
-| `compact.js` | 283 | dedupe, supersede, decay, regenerate |
-| `setup.js` | 282 | install into detected agent surfaces |
 | `staleness.js` | 213 | commit-distance staleness and capture gap |
-| `digest.js` | 209 | two-tier routing: description + tree |
-| `config.js` | 191 | engagement resolution; two-file config split |
+| `config.js` | 197 | engagement resolution; two-file config split |
 | `targets.js` | 140 | where each agent looks for instructions |
 | `graph.js` | 99 | recursive CTE traversal + retrieval budget |
 | `schema.js` | 98 | validate and normalize; reports every problem |
 | `redact.js` | 97 | capture-time fail-closed secret redaction |
 | `promptfile.js` | 78 | VS Code prompt files derived from SKILL.md |
 | `atomic.js` | 57 | atomic write; imports nothing from this package |
-| **total** | **3,574** | zero dependencies, 97 tests |
+| **total** | **4,168** | zero dependencies, 111 tests |
 
 `atomic.js` deliberately imports nothing from the package: `store.js` already imports
 `config.js`, so putting the atomic write in either would create a cycle.

README.md

@@ -440,7 +440,7 @@ equivalent and is not: npm links the global install to that folder rather than c
 it, which shows up as an arrow in `npm list -g`:
 
 ```
-`-- @vib795/agent-memory@0.6.5 -> .\..\..\..\agent-memory
+`-- @vib795/agent-memory@0.7.4 -> .\..\..\..\agent-memory
 ```
 
 Move or delete the clone afterwards and the global install points at nothing — the same

package.json

@@ -1,6 +1,6 @@
 {
   "name": "@vib795/agent-memory",
-  "version": "0.7.3",
+  "version": "0.7.4",
   "description": "Durable cross-repo knowledge graph for GitHub Copilot and Claude Code. Markdown source of truth, disposable SQLite index, zero runtime dependencies.",
   "keywords": [
     "github-copilot",

src/digest.js

@@ -3,22 +3,25 @@ import { createHash } from 'node:crypto';
 import { captureGap, currentRepo } from './staleness.js';
 
 /**
- * Two-tier routing.
+ * Tiers 1 and 2 of a three-tier ladder.
  *
- * Tier 1 is the `recall` skill's description, which is loaded into every chat
- * whether or not memory is ever used. It is standing cost, so it has to read like
- * a description rather than a document.
+ * Tier 1 is a skill description, loaded into every chat whether or not memory is ever
+ * used. It is standing cost, so it has to read like a description rather than a
+ * document. It has two occupants, not one: `recall`'s description advertises what the
+ * store knows; `remember`'s advertises what it is missing. Both are the same mechanism
+ * — a line of frontmatter that code regenerates and every conversation loads —
+ * pointed at opposite halves of the same problem.
  *
- * Tier 2 is the tree, printed only when `recall` actually fires. Per-invocation
- * cost, paid once, and only when someone is already looking something up.
+ * Tier 2 is printed only when a skill actually fires: the tree for `recall`, the
+ * capture brief for `remember`. Per-invocation cost, paid once, and only when someone
+ * is already looking something up or about to write one down.
  *
- * Neither tier costs a premium request. A request is charged per prompt, not per
- * tool call, so both of these ride inside a turn that was already paid for.
+ * Tier 3 — note bodies, and the colliding note that `write` hands back — lives
+ * outside this module, because by then the question is which note rather than which
+ * of them.
  *
- * Tier 1 has two occupants, not one. `recall`'s description advertises what the store
- * knows; `remember`'s advertises what it is missing. Both are the same mechanism — a
- * line of frontmatter that code regenerates and every conversation loads — pointed at
- * opposite halves of the same problem.
+ * Neither tier here costs a premium request. A request is charged per prompt, not per
+ * tool call, so both of these ride inside a turn that was already paid for.
  */
 
 // Never dropped from either tier. A constraint is what stops an agent from burning