memhtml 0.2.4 → 0.3.0

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "memhtml",
-  "version": "0.2.4",
+  "version": "0.3.0",
   "description": "An agent's long-term memory: one fact per semantic HTML file in git, four-arm retrieval, and a nightly sleep cycle.",
   "keywords": [
     "memory",

package/state-migrations/S0002_entity_corroboration.sql

@@ -0,0 +1,55 @@
+-- The corroboration counter on a machine-proposed ENTITY MERGE, the second one-way door that earns a
+-- place in the durable plane.
+--
+-- Sleep's entity resolution asks a model to partition one entity type's names into identity clusters.
+-- A proposal the model makes with no external evidence is a guess, and acting on it renames every
+-- `memhtml-entity` meta across the corpus and fuses two subjects' memories permanently — there is no
+-- inverse commit that separates them again. So a merge the model alone proposes is COUNTED here on the
+-- night it is seen and applied only once `detections >= 2`, meaning two different nights reached the
+-- same conclusion over independently re-read corpora.
+--
+-- The same discipline `state.edge_corroboration` holds for a machine-detected `contradicts`, and for
+-- the same reason: the counter is derived, high-churn, and gates a write into the files, so it belongs
+-- in the plane that is gitignored and re-exported rather than in the memory files themselves. Once the
+-- merge is applied the rewrite is file-borne and this row is decoration.
+--
+-- A merge backed by a DECLARED alias never reaches this table. A person file stating
+-- `<meta name="memhtml-alias" content="laith">` is a human's (or an authoritative directory's)
+-- assertion of identity, not a machine's suspicion, so it applies on the first night.
+--
+-- Known and accepted: the ORIENTATION is part of the key, so a merge whose direction flips restarts the
+-- counter. The phase's weight-then-lexicographic rule reads the corpus's own file counts, and those churn
+-- as memories are written and evicted — so a night that counted `laith -> laith al-saadoon` and a later
+-- night that reaches `laith al-saadoon -> laith` are two rows at one detection each, and neither
+-- promotes. That is the safe direction to be wrong in: a corpus whose weights are still moving has not
+-- settled which name is canonical, and a merge recorded under one orientation and applied under the other
+-- would rewrite every meta toward a name a later night disagrees with.
+
+CREATE TABLE state.entity_corroboration (
+  -- Keyed on the merge, not on a file. An entity name is a value in a `memhtml-entity` meta and
+  -- appears on many files, so the pair being corroborated is `(alias_name -> canonical_name)` within
+  -- one `entity_type`. `person:api` and `service:api` are two different subjects whose names collide,
+  -- so the type is part of the key rather than a column beside it.
+  entity_type    TEXT NOT NULL,
+  -- The name that would be REWRITTEN AWAY, normalized (lowercased, whitespace-collapsed) exactly as
+  -- the phase normalizes before comparing. Storing the raw authored form would make `Checkout API` and
+  -- `checkout api` two counters for one merge, and neither would ever reach two detections.
+  alias_name     TEXT NOT NULL,
+  -- The name that would SURVIVE, normalized. Which of the two this is follows from the phase's own
+  -- weight-then-lexicographic rule and never from the model's choice, so the row records the merge the
+  -- code would perform rather than the one the model described.
+  canonical_name TEXT NOT NULL,
+  detections     INTEGER NOT NULL DEFAULT 1 CHECK (detections >= 1),
+  confirmed      INTEGER NOT NULL DEFAULT 0 CHECK (confirmed IN (0,1)),
+  promoted       INTEGER NOT NULL DEFAULT 0 CHECK (promoted IN (0,1)),
+  updated_at     TEXT NOT NULL,
+  PRIMARY KEY (entity_type, alias_name, canonical_name)
+);
+-- The schema name goes on the INDEX, not on the table: `CREATE INDEX x ON state.entity_corroboration
+-- (...)` is a syntax error on this driver, while `CREATE INDEX state.x ON entity_corroboration (...)`
+-- is accepted and lands the index in the attached schema. Unqualified `entity_corroboration` resolves
+-- within it. (The same fact S0001 records for `state.access`.)
+--
+-- The lookup this serves is "every pending merge for one type", which the phase reads to report how
+-- many proposals are waiting on a second night.
+CREATE INDEX state.entity_corroboration_pending ON entity_corroboration (entity_type, promoted);