scrimshawlife-ctrl · Sep 25, 2026 · Sep 25, 2026
diff --git a/docs/vernacular/VERDICT-RULES.md b/docs/vernacular/VERDICT-RULES.md
@@ -0,0 +1,140 @@
+# Vernacular verdict rules — `metrics-rule-v1.4`
+
+Status: in use since 2026-09-25 PT. This page describes, in plain language, how the Vernacular export decides a verdict for
+each tracked slang term. The machine-readable Jev question contracts are in [`contracts/`](contracts/).
+
+This page holds rules only. It contains no posts, no author handles, no term example posts, no holdout ids and no database
+contents. Any example below is either a term name used to illustrate a ruling, or clearly marked as synthetic.
+
+## 1. The four verdicts
+
+| Verdict | Meaning | Who decides |
+|---|---|---|
+| **PASS** | Real slang with one coherent meaning, and the evidence shows real, independent, sustained use. | Jev says slang, then the deterministic rule finds at least one evidence path and no guard flag. |
+| **OBSERVE** | Real slang, but the evidence is not strong enough yet (or it is a meaningless meme, see §5). Keep watching. | Jev says slang, then the rule finds no evidence path, or a guard flag, or the null-meaning cap applies. |
+| **QUARANTINE** | Not a slang term, or the record is unusable: a proper name, a hashtag or tag, only an ordinary non-slang sense, a sentence fragment, or incoherent/polluted meanings. | Jev. Final. |
+| **HOLD** | Meaning empty, unknown or unconfirmed; several unrelated senses with none dominant; or Jev is unsure. Abstain and fail closed. | Jev. Final. |
+
+## 2. Two steps: Jev decides, the rule measures
+
+**Step 1: Jev** ([`jev_vote_contract.json`](contracts/jev_vote_contract.json)) answers one question: is this real slang with
+one coherent slang meaning? The answer is one of `SLANG_COHERENT`, `SLANG_NULL_MEANING`, `QUARANTINE`, `HOLD`.
+
+- Jev sees the recorded meanings, meaning history, variants and a few sample posts. It does **not** see volume, adoption,
+  lifecycle or scores, so popularity cannot bias the slang judgment.
+- A top probability below 0.5 becomes HOLD.
+- Borderline voting: if the first answer's top probability is between 0.35 and 0.65 (inclusive), two more independent calls
+  are made and the majority (at least 2 of 3) wins. No majority means HOLD.
+- **Jev decides slang vs QUARANTINE/HOLD. Nothing downstream overrides it.** QUARANTINE and HOLD are final: the rule never
+  upgrades them, and a dictionary entry never overrides Jev (a term with a matching dictionary entry that Jev quarantines
+  stays QUARANTINE).
+
+**Step 2: the deterministic rule** runs only when Jev says `SLANG_COHERENT` (or `SLANG_NULL_MEANING`, see §5).
+A term is **PASS** if at least one evidence path is met **and** no guard flag is raised. Otherwise it is **OBSERVE**.
+
+## 3. Evidence is counted on slang-sense posts only
+
+Each post is labelled by Jev ([`jev_sense_contract.json`](contracts/jev_sense_contract.json)) as using the term in its slang
+sense (`SLANG_SENSE`, which includes mentioning or explaining the slang) or as an unrelated same-spelling use (`OTHER_SENSE`:
+a name, school, company, ticker, technical term, literal word or another language). Answers below 0.5 are `UNKNOWN` and are
+excluded. For a term with an operator definition, posts are labelled against that definition instead (a slang sense not in
+the definition counts as `OTHER_SENSE`).
+
+All the author, week, day, community and attestation counts below use **slang-sense posts only**.
+
+## 4. Evidence paths
+
+Every path requires real, independent use. The core minimum is **at least 3 distinct slang-sense authors over at least
+2 active weeks**.
+
+| Path | Condition (all on slang-sense posts) | Idea |
+|---|---|---|
+| **A_BROAD_ADOPTION** | ≥ 8 authors **and** ≥ 3 communities **and** ≥ 2 active weeks | Many people in several communities, more than one week. |
+| **B_PERSISTENT_USE** | ≥ 3 authors **and** ≥ 3 active weeks **and** a lifespan of ≥ 14 days | Few but independent users who keep using it for more than two weeks. |
+| **C — attested** | ≥ 3 authors **and** ≥ 2 active weeks **and** one attestation source (below) | People explain it (or the operator or a dictionary does), plus minimal independent use. |
+| **D_LINEAGE** | ≥ 1 tracked compound relation (child or parent term) **and** ≥ 3 authors **and** ≥ 2 active weeks | A generative term (for example aura → aura farming), plus minimal independent use. |
+
+**Path C attestation sources.** Only one C label is recorded, checked in this order:
+
+1. **`C_ATTESTED`**: ≥ 2 independent people in the corpus explaining the term (definition-style posts or explainer accounts).
+2. **`attested_operator`**: otherwise, the operator has defined the term.
+3. **`attested_dictionary`**: otherwise, an Urban Dictionary, Green's Dictionary of Slang or Wiktionary sense **matches** the
+   term's slang-sense meaning ([`jev_dictionary_contract.json`](contracts/jev_dictionary_contract.json): one MATCH/DIFFERENT
+   question per dictionary sense; a sense counts only at MATCH with p ≥ 0.6).
+
+An operator definition or a matching dictionary sense fills **only the "≥ 2 explainers" part** of path C. The ≥ 3 slang-sense
+authors, the ≥ 2 active weeks and every guard still apply, and neither changes Jev's answer. Only the matching sense counts:
+a different sense of the same spelling (for example "slay" meaning "to kill") does not. An unmatched dictionary hit count
+never counts toward path C.
+
+**What is never a path.** Raw volume (number of posts), velocity and search-trend scores are recorded for information but
+appear in no path. Repetition alone can never produce PASS.
+
+## 5. Guard flags (any one blocks PASS)
+
+| Flag | Raised when |
+|---|---|
+| `ONE_AUTHOR_DEPENDENCE` | One author wrote more than half of the slang-sense posts. |
+| `BOTLIKE_REPETITION` | Half or more of the posts are identical copies (after removing URLs and handles). |
+| `SINGLE_BURST` | Fewer than 2 active days, or 80 % or more of the posts fall on one day. |
+| `FIXTURE_EVIDENCE` | Half or more of the evidence is test/fixture data. |
+| `DORMANT_OR_UNDATED` | No use in the last 60 days (measured against the newest post in the database), or no dated post. |
+| `LOW_SENSE_PURITY` | Fewer than 30 % of the posts use the slang sense **and** fewer than 8 slang-sense authors. |
+| `SENSE_UNVERIFIED` | More than half of the posts have no usable sense label, or none do. |
+| `NULL_MEANING_CAP` | Jev answered `SLANG_NULL_MEANING` (see below). Always OBSERVE. |
+
+**Meaningless memes are capped at OBSERVE.** Operator ruling 2026-09-25: a real slang term whose point is that it means
+nothing (for example "67") is answered `SLANG_NULL_MEANING` by Jev and is always **OBSERVE**, even if an evidence path is met.
+The paths are still recorded. A term that merely has a vague, mild or emotional meaning (an intensifier, general approval)
+is *not* null-meaning.
+
+## 6. Compounds are mutations
+
+Operator ruling 2026-09-25: when a phrase combines two base terms, it is a **mutation** of those terms, not a separate
+adoption and not a member of an unrelated family. Example: "brain rot" and "aura" are terms; "brain rot aura" is a mutation
+of the two. Compounds count as mutation/lineage evidence for their parents (path D), not as independent adoption. (An
+earlier export folded "brainrot aura" into the `ai-native` family; that was a bug, fixed in Hyperlex PR #123.)
+
+## 7. Meaning: context first, operator as fallback
+
+- The meaning of a term is worked out **from context first**: Jev picks among the recorded glosses (or short meanings quoted
+  verbatim from the term's own slang-sense posts) the one the posts support
+  ([`jev_meaning_contract.json`](contracts/jev_meaning_contract.json)). Jev never writes a meaning, so nothing is invented.
+  A resolved context meaning is `INFERRED`.
+- The operator is asked **only when context cannot settle it** (`UNRESOLVED`, and the verdict is not QUARANTINE). A term the
+  operator says they don't know is marked `operator_unknown` and is not asked again.
+- Precedence of meanings: operator definition (`OBSERVED`) > context (`INFERRED`) > none (`UNRESOLVED`).
+- A context meaning is not attestation and never changes a verdict.
+- **An operator definition counts as attestation** (`attested_operator`, §4), **but real use is still required**: the posts
+  must still show ≥ 3 slang-sense authors over ≥ 2 weeks with no guard flag.
+
+## 8. Fail-closed behaviour
+
+- A Jev error, a missing answer or an out-of-set answer leaves the term's note untouched. No verdict is guessed.
+- A slang answer with no metrics row, a missing sense label or a missing dictionary judgment also leaves the note untouched.
+- Any vote error in borderline voting fails the term closed; partial vote sets are never used.
+- A verdict is a label only and never triggers publishing.
+- Any threshold change requires bumping the rule version.
+
+## 9. Version history
+
+| Version | Date (PT) | Change |
+|---|---|---|
+| metrics-rule-v1 | 2026-09-25 10:39 | Deterministic PASS/OBSERVE rule with paths A–D and guards. |
+| v1.1 | 2026-09-25 10:43 | Null-meaning cap (meaningless memes → OBSERVE). |
+| v1.2 | 2026-09-25 10:50 | Evidence on slang-sense posts only; `LOW_SENSE_PURITY` and `SENSE_UNVERIFIED` guards; borderline 3-vote majority. |
+| v1.3 | 2026-09-25 10:57 | Operator definition satisfies path C's explainer part (`attested_operator`). |
+| **v1.4** | 2026-09-25 11:20 | A sense-matched Urban Dictionary / Green's / Wiktionary entry does too (`attested_dictionary`); the unmatched dictionary hit count no longer feeds path C. |
+
+## Contracts in this folder
+
+| File | Contract | Question |
+|---|---|---|
+| [`contracts/jev_vote_contract.json`](contracts/jev_vote_contract.json) | `rebuilt-v3.2-metrics` | Verdict vote: real slang with one coherent meaning? (with borderline voting) |
+| [`contracts/jev_sense_contract.json`](contracts/jev_sense_contract.json) | `rebuilt-sense-v1` | Per post: slang sense or unrelated same-spelling use? |
+| [`contracts/jev_meaning_contract.json`](contracts/jev_meaning_contract.json) | `rebuilt-meaning-v2` | Which recorded meaning do the posts support? |
+| [`contracts/jev_dictionary_contract.json`](contracts/jev_dictionary_contract.json) | `rebuilt-dictmatch-v1` | Per dictionary sense: same slang sense as the corpus usage? |
+
+Each JSON file is a sanitized published copy: its `_publication` block names the source file and its sha256. The contracts
+embed no real posts; the `synthetic_example` blocks are invented placeholders (the made-up term "zorblet" and posts marked
+`SYNTHETIC`).
diff --git a/docs/vernacular/contracts/jev_dictionary_contract.json b/docs/vernacular/contracts/jev_dictionary_contract.json
@@ -0,0 +1,77 @@
+{
+  "_publication": {
+    "source_file": "jev_dictionary_contract.json",
+    "source_sha256": "0616564b2b45fa503db95a4a482b7f5db77aeb2ae9f71340e485c15637783efa",
+    "note": "Published copy for documentation. Sanitized on 2026-09-25 PT: local file paths, cache paths, account names and people's names were removed or generalized; the question text, closed sets, thresholds and fail-closed rules are unchanged. This contract embeds NO real posts, X post text, author handles, term example posts, holdout ids or database contents: the {term}/{pid}/STATE fields are template placeholders filled at run time. The illustrative state under 'synthetic_example' (where present) is invented and clearly synthetic."
+  },
+  "contract_id": "vernacular-dictionary-sense-match",
+  "version": "rebuilt-dictmatch-v1",
+  "provenance": "NEW 2026-09-25 11:24 PT, rebuilt-series label (not an original contract). External dictionary attestation (Urban Dictionary, Green's, Wiktionary): an entry whose sense matches the term's slang-sense meaning satisfies path C's '>= 2 explainers' part only (path label attested_dictionary); only the matching sense counts. Jev only judges whether each dictionary sense matches the corpus usage. General instructions; no term-specific hints.",
+  "approved_by": "operator, 2026-09-25 PT (standing Jev permission, $1/run cap)",
+  "model": "jev-latest",
+  "sources": {
+    "urbandictionary": {
+      "url": "https://api.urbandictionary.com/v0/define?term={q}",
+      "max_senses": 6,
+      "min_interval_s": 1.0,
+      "note": "exact-word entries in API order; the API currently returns 0 votes for every entry, so no vote floor is possible (user-generated: weakest source)"
+    },
+    "wiktionary": {
+      "url": "https://en.wiktionary.org/w/index.php?action=raw&title={title}",
+      "max_senses": 8,
+      "min_interval_s": 1.0,
+      "note": "English-section definition lines with their {{lb}} labels; quotation lines (##*) skipped"
+    },
+    "greens": {
+      "search_url": "https://greensdictofslang.com/search/basic?q={q}",
+      "max_entries": 3,
+      "max_senses": 4,
+      "min_interval_s": 2.0,
+      "note": "exact-headword search results, then the entry page's definition text"
+    }
+  },
+  "fetch": "Crawl4AI (self-hosted, free); no Firecrawl, no X. Fetched pages are cached permanently, so a term is fetched only the first time it is looked up.",
+  "max_posts": 8,
+  "max_post_chars": 240,
+  "min_match_probability": 0.6,
+  "question": {
+    "type": "choice",
+    "instructions": "STATE.dictionary_senses['{d}'] is one dictionary sense (source: {src}) of '{term}'. STATE.posts show how '{term}' is used as slang in our corpus; STATE.meaning, if not empty, is its recorded slang meaning. Does this dictionary sense describe the SAME slang sense as that corpus usage (a close paraphrase, or a broader or narrower wording of the same sense, counts)? A different sense of the same spelling (literal, older, unrelated, or a different slang meaning) is DIFFERENT.",
+    "criteria": {
+      "MATCH": "the dictionary sense is the same slang sense as the corpus usage",
+      "DIFFERENT": "a different sense of the same spelling, or unrelated"
+    }
+  },
+  "statuses": {
+    "MATCH": ">= 1 dictionary sense matched (p >= 0.6) -> attestation",
+    "NO_MATCH": "senses found, none matched",
+    "NO_ENTRY": "no entry in any source",
+    "NO_SLANG_POSTS": "no slang-sense posts to compare against (not attested)"
+  },
+  "fail_closed": "fetch error with no match found, or Jev error: no row is written for the term (previous row kept); the export fails that term closed.",
+  "synthetic_example": {
+    "_synthetic": true,
+    "STATE": {
+      "term": "zorblet",
+      "meaning": "SYNTHETIC: impressive",
+      "posts": {
+        "p1": "SYNTHETIC: that fit is zorblet"
+      },
+      "dictionary_senses": {
+        "d1": {
+          "source": "urbandictionary",
+          "text": "SYNTHETIC: something extremely impressive"
+        },
+        "d2": {
+          "source": "wiktionary",
+          "text": "SYNTHETIC: a small pastry"
+        }
+      }
+    },
+    "expected": {
+      "d1": "MATCH",
+      "d2": "DIFFERENT"
+    },
+    "term_status": "MATCH"
+  }
+}
diff --git a/docs/vernacular/contracts/jev_meaning_contract.json b/docs/vernacular/contracts/jev_meaning_contract.json
@@ -0,0 +1,107 @@
+{
+  "_publication": {
+    "source_file": "jev_meaning_contract.json",
+    "source_sha256": "deb8f4f657662fd8836fd512b0517dba0e3518ed636aeb6886fac9fff4281aba",
+    "note": "Published copy for documentation. Sanitized on 2026-09-25 PT: local file paths, cache paths, account names and people's names were removed or generalized; the question text, closed sets, thresholds and fail-closed rules are unchanged. This contract embeds NO real posts, X post text, author handles, term example posts, holdout ids or database contents: the {term}/{pid}/STATE fields are template placeholders filled at run time. The illustrative state under 'synthetic_example' (where present) is invented and clearly synthetic."
+  },
+  "contract_id": "vernacular-context-meaning",
+  "version": "rebuilt-meaning-v2",
+  "supersedes": "rebuilt-meaning-v1 (sha256 6512419a994c946ea2b8275fa586e5784ae2db4caa61083cdae648761a76d64d; its rows are kept as history)",
+  "provenance": "NEW 2026-09-25 11:07 PT, rebuilt-series label (not an original contract). Synonym-split fix (meaning field only, not verdicts). Changes vs v1: (1) near-duplicate candidates are merged before scoring (Jev pairwise same-sense check, complete linkage) so synonyms no longer split probability; (2) corpus/evidence notes are no longer candidates; (3) when no stored gloss resolves, a second stage offers short meanings taken verbatim from the term's own slang-sense posts (deterministic 'X means / stands for / = ...' extraction). Jev only returns choice/boolean/score answers, so it cannot write free text and nothing is invented. Same confidence + post-agreement bar as v1.",
+  "approved_by": "operator, 2026-09-25 PT (standing Jev permission; feature 10:57-10:58 PT; v2 fix approved 11:05 PT)",
+  "model": "jev-latest",
+  "max_posts": 20,
+  "max_post_chars": 240,
+  "max_glosses": 6,
+  "max_post_candidates": 6,
+  "max_candidate_words": 8,
+  "merge": {
+    "min_same_probability": 0.6,
+    "linkage": "complete",
+    "representative": "first member in candidate order (stored glosses before post-derived)"
+  },
+  "voting": {
+    "band_lo": 0.35,
+    "band_hi": 0.65,
+    "votes": 3
+  },
+  "resolve_rule": {
+    "min_single_top_probability": 0.66,
+    "min_voted_mean_probability": 0.5,
+    "min_post_agreement": 0.6,
+    "min_supporting_posts": 2,
+    "min_post_answer_probability": 0.5
+  },
+  "gloss_exclude_patterns": [
+    "sacred NOT_COMPUTABLE",
+    "not slang",
+    "not a slang",
+    "not a lexeme",
+    "NOT_COMPUTABLE from body",
+    "corpus:",
+    "corpus meta",
+    "unconfirmed",
+    "unproven",
+    "claimed",
+    "claim ",
+    "gloss unclear"
+  ],
+  "unresolved_categories": {
+    "conflict": [
+      "overall=MIXED",
+      "overall=NO_MAJORITY",
+      "posts disagree"
+    ],
+    "obscure": [
+      "no slang-sense posts",
+      "no candidate meanings",
+      "overall=NONE_FIT",
+      "low confidence",
+      "too few supporting posts"
+    ]
+  },
+  "note": "A context meaning (INFERRED) fills the note's meaning field only; it is NOT attestation and never changes a verdict. An operator definition (OBSERVED) overrides it. Context first; the operator is asked only when context cannot settle the meaning (UNRESOLVED).",
+  "questions": {
+    "merge": {
+      "type": "boolean",
+      "instructions": "STATE.candidates are short candidate meanings for the slang term '{term}'. Do candidates {a} and {b} describe the same sense of the term (paraphrases or near-synonyms of one meaning), rather than different meanings?"
+    },
+    "overall": {
+      "type": "choice",
+      "instructions": "STATE.posts are posts that use the term '{term}' as slang. Each option lists one meaning (alternative phrasings of the same meaning are separated by ' | '). Which option best describes how the term is used in these posts? Pick an option only if most posts fit it. Pick NONE_FIT if no option fits the posts, MIXED if the posts use clearly different senses with none dominant.",
+      "criteria_extra": {
+        "NONE_FIT": "no listed meaning fits how the posts use the term",
+        "MIXED": "posts use clearly different senses; none dominant"
+      }
+    },
+    "support": {
+      "type": "choice",
+      "instructions": "Post {pid} in STATE.posts: is the term '{term}' used there in the sense given in STATE.meaning (same or a close variant of that sense)? Mentioning or explaining the term with that sense counts as SUPPORTS.",
+      "criteria": {
+        "SUPPORTS": "uses or explains the term in STATE.meaning's sense (or a close variant)",
+        "DIFFERENT": "uses the term in a different sense, or the post does not show the sense"
+      }
+    }
+  },
+  "synthetic_example": {
+    "_synthetic": true,
+    "STATE": {
+      "term": "zorblet",
+      "candidates": {
+        "M1": "SYNTHETIC: impressive | really cool",
+        "M2": "SYNTHETIC: a small pastry"
+      },
+      "posts": {
+        "p1": "SYNTHETIC: that fit is zorblet",
+        "p2": "SYNTHETIC: zorblet energy today"
+      }
+    },
+    "expected": {
+      "overall": "M1",
+      "support": {
+        "p1": "SUPPORTS",
+        "p2": "SUPPORTS"
+      }
+    }
+  }
+}