clearotron 0.2.2 → 0.2.3

package/driver/publish/render.mjs

@@ -456,8 +456,8 @@ function meter(name, entry) {
 // (emitted by the report-card stage, parsed into card.meta.ord) and the driver already orders +
 // provenance-checks cards by it — so match on THAT exact, injective key. The old fuzzy owner/mark
 // containment collapsed distinct findings onto one card (`.find()` first-match): two findings sharing
-// a mark, or whose mark is a substring of another's ("aura" ⊂ "aurora"), all matched the first card,
-// rendering one card's prose under every finding (the PETCARY "02 On-field conflicts" duplication). The
+// a mark, or whose mark is a substring of another's ("nova" ⊂ "novadex"), all matched the first card,
+// rendering one card's prose under every finding (the "02 On-field conflicts" duplication). The
 // fuzzy match is kept ONLY as a fallback for legacy artifacts whose cards predate the `- ord:` line.
 function matchCard(finding, cards) {
   const list = cards || [];

package/driver/reference-strip-signatures.mjs

@@ -30,6 +30,26 @@ export const SIGNATURES = [
     // the strip's residue always leaves whitespace after the hyphen.
     re: /\bpre-\s+[a-z]/,
   },
+  {
+    name: "a parenthesis opening on a comma",
+    // THE THIRD RESIDUE, and the floor could not see it for two years because the first two are
+    // WORD-level — a bare possessive, a dangling `pre-` — and this one is PUNCTUATION. It looks like
+    // ordinary code to a reader scanning for broken English, which is exactly why a census built from
+    // the shapes somebody had noticed reads like a census of the whole problem.
+    //
+    // Where the removed reference was the entire content of a parenthesis, nothing is left but `(,)`.
+    // Where it opened one, the sentence now begins on a comma: `(, 2026-08-21)`. Two ran off the end of
+    // their line.
+    //
+    // THE LOOKAHEAD IS THE WHOLE PRECISION AND IT IS DERIVED, NOT GUESSED. `(,` also occurs in code —
+    // inside a character class (`[(,=:…]`, `[;{\s(,]`) and in an alternation (`(,|$)`) — where those two
+    // characters are syntax rather than damage. Every code instance is followed by a regex metacharacter
+    // and no residue instance is, so the exclusion is `|`, `]` and `=` rather than a list of files that
+    // would go stale. Measured against the tree as it stood before the repair: 123 matches, which is the
+    // residue count the finding reported, and 0 after it. A signature that reproduces the number it was
+    // derived from is one that is reading the same thing.
+    re: /\(\s*,(?![|\]=])/,
+  },
 ];
 
 // THE RULE'S OWN DEFINITION IS THE ONLY EXEMPTION, and it is named rather than pattern-matched.

package/driver/search-policy.mjs

@@ -877,7 +877,7 @@ export const RECIPE_KEYS = [
  * The saved-search SCOPE block — where the machinery points, as against `base`/`components`, which
  * choose which machinery runs.
  *
- * This is what makes a saved search a saved SEARCH rather than a saved depth. "Zephyr Beverages knockouts — US
+ * This is what makes a saved search a saved SEARCH rather than a saved depth. "Drinks knockouts — US
  * focus" is a label over exactly this: base `knockout`, scope `{jurisdictions:["US"]}`. Without it a
  * saved search could only ever restate a level that is already one click away, which is why the recipe
  * store existed for weeks with nothing worth putting in it.
@@ -1066,7 +1066,7 @@ let recipeCache = null;
  * THE STORE IS NAMED, NEVER GUESSED. `CLEAROTRON_RECIPES_DIR` unset means this deployment has no saved
  * searches — not that it should use the ones bundled with the source.
  *
- * `driver/recipes/` ships synthetic demos for two FICTIONAL customers (aurora, zephyr) so the dev
+ * `driver/recipes/` ships synthetic demos for two test accounts so the dev
  * cockpit and the tests have something to render. It used to be the fallback when the env var was
  * unset, which is a foot-gun that only fires in production: the real deployment does not set the
  * variable, so switching saved searches on there would have surfaced invented customers inside the

package/driver/skills/blind-frame/SKILL.md

@@ -27,7 +27,7 @@ Decompose the mark into its element(s) and name the **dominant element** (the sp
 
 - **Drop** characters: shorten the element (VELTRIN → VELTRI). A dropped letter is the commonest missed-cluster cause — the shorter root is its own crowded field.
 - **Add** characters / **composite**: the element living inside a larger mark (DELPHI Diagnostics, Osler Delphi, Delphic). A composite that shares your dominant element is on the board.
-- **Phonetic / homophone**: sound-alikes (ZEPHYR / ZEFFYR / ZEPHUR).
+- **Phonetic / homophone**: sound-alikes (PHAROLIS / FAROLIS / PHAROLLIS). Carry the `ph`/`f` pair in particular: it sounds identical, files under a different letter, and survives no letter-distance measure — which is the whole reason this class is separate from the two above.
 - **Neighbour**: a one-keystroke real-word or famous-mark neighbour (CHROME on a NOVAPULSE clearance). A famous neighbour is carried for diligence even when off-field.
 
 For each, give the value, the direction, and one line of rationale.

package/driver/skills/knockout-assess/SKILL.md

@@ -119,6 +119,90 @@ Measured tone only — no "extremely difficult", "most dangerous", "massive", "e
 colour communicates urgency. No quantitative claims from research ("major streaming hit", never
 "294M streams"; "significant market presence", never revenue figures).
 
+## Plain language — the two-register rule
+
+The report goes to a lawyer who layers advice on top, and that lawyer's client reads the same page. The
+band, the summary, the basis line and the one-liners are the whole product for the second reader. They
+were the hardest lines on the page.
+
+**Everything a reader sees before opening a fold is written for the client's business owner.** Ordinary
+nouns, one idea per sentence, no sentence over 25 words, the conclusion first. On this product that is:
+`chunkSummary`, `basis`, `factors`, `counterFactors`, `mitigation`, every finding's `net`, every
+`registerReads[].read`, the `purpleNotes`, and `standardCaveats`.
+
+**Inside a fold the lawyer's vocabulary is allowed** where a plain word would lose precision — a
+finding's `basis`, the provenance, the long `assessment`. Even there, a term is explained the first time
+it appears on the page.
+
+The rule is the test, not the list. Ask of every noun: *does the reader already own this word?* These
+are worked examples of the one failure, not its boundary:
+
+| Instead of | Write |
+|---|---|
+| proprietor | owner |
+| subsisting | live |
+| senior | earlier, or came first |
+| specification | goods list |
+| DELPH-formative | names built on DELPH- |
+| prevail | win |
+| citable prior rights | earlier marks the office can raise against you |
+| vulnerable to a non-use attack | could be cancelled for not being used |
+| on the record as it stands | on what we found |
+| the marks-and-goods comparison | same name, same goods |
+| the confusion comparison meets on every limb | same name, same goods, same shops |
+| belt-and-braces classes | extra classes |
+| dispatch, instructed | the request, what was asked |
+| lane | class, or channel — whichever you mean |
+| chunk | never; say the batch, or these names |
+
+The target is the level of these, each the standard for its line:
+
+> **Basis.** "ORBIT is already the name of two satellite-tracking apps on the same app stores, and of an
+> established satellite-communications company. Any of them would likely win a dispute over this name
+> for this software. The word is a weak mark for these goods, which is why this is High and not Very
+> High."
+
+> **A finding's `net`.** "Same name, same goods, and their filing came first in every territory we
+> searched. We see no argument against it."
+
+> **The batch opener.** "One name screened: ORBIT, rated High."
+
+**Shortening by dropping the reason is not the fix.** The "why" stays, in plain words. A visible
+line that is short because it no longer says why is worse than the long one it replaced, and it
+passes every length rule on this page while telling the reader less.
+
+**Say it once.** A caveat, a limit or a finding belongs in one place. Four `standardCaveats` that each
+restate "this is not a clearance" are one caveat and three repetitions — and the report's scope block
+already carries that sentence in the renderer's own words, so a caveat repeating it is dropped before
+the reader sees it. Write the one thing the scope block does not say, or write none.
+
+**A note is verb-first, one line, and about one of two things.** Write each `purpleNotes` entry as an
+instruction to the reviewing lawyer, opening on the verb. Every note is about the REQUEST — what was
+asked, and whether it holds together — or about the NAME. The report prints the first kind at the top,
+above the conflicts, and the second under that name's cards, so a note that mixes the two lands in one
+place and reads oddly in it. Write two notes.
+
+A note about the request NAMES the request in the note — "the request", or "what we were asked". That is
+not a formality: the report sorts the two kinds by what the note talks about, so a request note that
+never mentions the request is filed under the name and prints in the wrong place.
+
+> "Check the request. The client is described as a beverages and supplements business, but we were asked
+> to screen satellite-tracking software in Class 9. We screened the software. If beverages is the real
+> business, this screen looked at the wrong market."
+
+> "Ask the client whether it already uses ORBIT. The request does not say, and the client's own earlier
+> use would change the picture."
+
+A note about the name opens on what the lawyer should do about the name:
+
+> "Pull EG Tech's full goods list at clearance. It is the record most likely to change the picture in
+> either direction."
+
+> "Search Classes 42 and 38 in their own right. Both carry live ORBIT filings, not spillover from
+> Class 9."
+
+A note that restates a finding already on the page is not a note. Cut it.
+
 ## The per-mark opening read (`assessment`)
 
 Each mark's report is delivered on its own, and `assessment` is the first thing its reader meets —

package/driver/skills/matter-frame/SKILL.md

@@ -75,11 +75,11 @@ You do not write a file. Hand the frame back by calling `record_matter_frame`; t
 
 ### Scope jurisdictions — search wide, cite narrow (instructed scope honored; worldwide / brand-signalled scope leans wide + discloses)
 
-Two scopes live here, and the doctrine is **narrow at citation, never silently at search**: the **search scope** (where the register sweep looks) must be wide enough to catch the conflict; the **citation scope** (what the report flags and ranks) is where rights legally matter. The failure to avoid is the PETCARY miss — *silently* pinning a globally-signalling brand to one country, so the real EU/UK/US conflict is never even searched. Derive both from THIS matter; a customer profile selects the **risk framework** only — it never sets territory.
+Two scopes live here, and the doctrine is **narrow at citation, never silently at search**: the **search scope** (where the register sweep looks) must be wide enough to catch the conflict; the **citation scope** (what the report flags and ranks) is where rights legally matter. The failure to avoid is the pinned-scope miss — *silently* pinning a globally-signalling brand to one country, so the real EU/UK/US conflict is never even searched. Derive both from THIS matter; a customer profile selects the **risk framework** only — it never sets territory.
 
 - **Instructed scope is honored (narrow or wide).** When the client names territories — a list, or "X only" — that is the scope for BOTH search and citation; do not override a client who genuinely wants a narrow scope (a sweep of a country with no instructed interest and no reach is wasted). The profile's `defaultJurisdictions`, when present, is the *customer's own standing instruction* — treat it like an instructed set.
-- **Worldwide, or the brand's own signals point broader → the SEARCH widens (it never silently narrows).** When the instruction is worldwide, OR the brand signals broader use — a global gTLD (a `.io`/`.com` launch), an investor / expansion announcement, stated target markets, a genuinely global customer base — the **search set includes the major markets (US / EU / UK / CN / JP) plus the signalled markets**. Decide which to actually cite and rank at **selection** (where rights matter), and **disclose** any material market you could not fully cover (a `coverage-limited` row — never a silent clean). The one move forbidden: quietly pinning an unpinned, globally-signalling brand to a single country (the PETCARY miss). **The cost guard stays:** an abstract *"the sector tends to be global"* is **not, by itself, the widener** — the widener is a CONCRETE signal (a worldwide instruction, or the brand's own global footprint), so a deliberately local matter stays local.
-- **Genuinely unpinned, no global signal → reason the likely scope, LABEL it inferred, and DISCLOSE it.** Read the product / customer base / channels and scope to where the brand plausibly operates; never silently default to the requester's home country (that is the PETCARY mechanism). Where reach is plausibly broad, lean wide; where clearly local, scope local — either way set `Scope basis: inferred` and record a one-line `Scope assumption:` so a narrowing is visible and the coverage gate can clamp + disclose it.
+- **Worldwide, or the brand's own signals point broader → the SEARCH widens (it never silently narrows).** When the instruction is worldwide, OR the brand signals broader use — a global gTLD (a `.io`/`.com` launch), an investor / expansion announcement, stated target markets, a genuinely global customer base — the **search set includes the major markets (US / EU / UK / CN / JP) plus the signalled markets**. Decide which to actually cite and rank at **selection** (where rights matter), and **disclose** any material market you could not fully cover (a `coverage-limited` row — never a silent clean). The one move forbidden: quietly pinning an unpinned, globally-signalling brand to a single country (the pinned-scope miss). **The cost guard stays:** an abstract *"the sector tends to be global"* is **not, by itself, the widener** — the widener is a CONCRETE signal (a worldwide instruction, or the brand's own global footprint), so a deliberately local matter stays local.
+- **Genuinely unpinned, no global signal → reason the likely scope, LABEL it inferred, and DISCLOSE it.** Read the product / customer base / channels and scope to where the brand plausibly operates; never silently default to the requester's home country (that is the pinned-scope mechanism). Where reach is plausibly broad, lean wide; where clearly local, scope local — either way set `Scope basis: inferred` and record a one-line `Scope assumption:` so a narrowing is visible and the coverage gate can clamp + disclose it.
 - **In scope by reach.** A territory also counts when a right is **legally effective in an instructed territory by any route** — filed there; an international (Madrid) registration **designating** it; a regional/supranational right covering it (an EUTM covers every EU member; an instructed EU member is covered by an EUTM); a treaty / priority effect. Tag such a territory `in-scope-by-reach` and name the route. Reason the route from the right — never assume it.
 - **Excluded.** Territories you considered and set aside — neither instructed, nor brand-signalled, nor a major on a worldwide matter — name each, the reason, and a **reopen trigger** (a right or signal pointing there surfaces). A hit effective ONLY outside the scope set is out of *citation* scope (drop it from findings); but a market the brand actually signals belongs in the SEARCH set, not here.
 - **Primary vs only.** If the instruction names territories of "primary" / "first" interest WITHOUT "only", treat the named ones as the citation core but **let the search lean wide** to the majors / brand-signalled markets (the "primary" wording signals more may matter), and **record the assumption** — a one-line `Scope assumption: <text>`. (A single clarifying question is acceptable instead; the recorded default never blocks the run.)
@@ -88,7 +88,7 @@ Two scopes live here, and the doctrine is **narrow at citation, never silently a
 
 *Worked illustrations:*
 *— Instructed "Switzerland and the EU, only" → search CH + the EUTM/EU layer (+ any Madrid designation reaching them); majors beyond (CN/JP/US/UK) are out of scope — an honored narrow instruction (`Scope basis: instructed`).*
-*— A brand on a `.io` domain with an investor deck naming a US/EU launch and no explicit territory list → lean wide: search US/EU/UK/CN/JP + the named launch markets; cite where rights matter; disclose any gap (`Scope basis: inferred`). This is the PETCARY fix — do not pin it to CH.*
+*— A brand on a `.io` domain with an investor deck naming a US/EU launch and no explicit territory list → lean wide: search US/EU/UK/CN/JP + the named launch markets; cite where rights matter; disclose any gap (`Scope basis: inferred`). This is the pinned-scope fix — do not pin it to CH.*
 *— A local artisan, national-only product, no global signal → scope the home market, `Scope basis: inferred`, `Scope assumption:` recorded — narrow but disclosed, never silent.*
 
 ### Off-field sectors
@@ -111,7 +111,7 @@ Two scopes live here, and the doctrine is **narrow at citation, never silently a
 ### Sector-convergence flags
 - Sectors off-field today where commercial trajectories may bring them adjacent
 - Brief reason
-- Example: "Aurora Interactive's broader AI / cloud AI narrative could theoretically bring vector-DBs adjacent to game-dev tooling, but Aurora dev kit is specifically game-tooling not LLM-tooling; convergence path exists but weak"
+- Example: "Foxglade Interactive's broader AI / cloud AI narrative could theoretically bring vector-DBs adjacent to game-dev tooling, but Foxglade dev kit is specifically game-tooling not LLM-tooling; convergence path exists but weak"
 
 ### Machine lines the task message dictates
 
@@ -132,7 +132,7 @@ driver's fixed query shapes:
 
 ## Reasoning posture
 
-- When in doubt about an **off-field / adjacency** decision, INCLUDE rather than exclude — matter-frame should not silently filter a *field*; downstream review (`placement-inquiry`) can deprioritise later with reasoning. **Jurisdiction scope follows its own rule (see *Scope jurisdictions*): search wide, cite narrow.** An instructed-narrow scope is honored; a worldwide or brand-signalled scope widens the SEARCH to the majors + signalled markets; an inferred scope is labelled and disclosed. Do not silently pin a globally-signalling brand to one country (the PETCARY miss), and do not widen a deliberately-local matter on abstract "the sector is global" reasoning (a wasted sweep).
+- When in doubt about an **off-field / adjacency** decision, INCLUDE rather than exclude — matter-frame should not silently filter a *field*; downstream review (`placement-inquiry`) can deprioritise later with reasoning. **Jurisdiction scope follows its own rule (see *Scope jurisdictions*): search wide, cite narrow.** An instructed-narrow scope is honored; a worldwide or brand-signalled scope widens the SEARCH to the majors + signalled markets; an inferred scope is labelled and disclosed. Do not silently pin a globally-signalling brand to one country (the pinned-scope miss), and do not widen a deliberately-local matter on abstract "the sector is global" reasoning (a wasted sweep).
 - Be concrete about WHY for each item. A senior lawyer reading matter-context.md should be able to disagree with specific items and see exactly where your reasoning broke — not just see a conclusion.
 - If you find yourself rating something as in-lane that a senior lawyer would obviously drop, surface the tension rather than smoothing it.
 

package/driver/skills/matter-frame/watchlist-reference.md

@@ -47,6 +47,6 @@ watchlists come from in any case, and is why this file was never authority.
 
 ## Cross-sector
 
-- **Aurora Interactive** — `major_brand_owners`. Broad AURORA / NEBULA / VANGUARD families;
-  on an Aurora Interactive-adjacent matter watch the relevant family token and name
+- **Foxglade Interactive** — `major_brand_owners`. Broad FOXGLADE / NEBULA / VANGUARD families;
+  on a Foxglade Interactive-adjacent matter watch the relevant family token and name
   ecosystem partners with their partnership context rather than as bare conflicts.

package/driver/skills/prelim-common-law/SKILL.md

@@ -173,7 +173,7 @@ For game-title rows, the `developer_of_record` and `publisher_of_record` columns
 | Finding | Source / Platform | URL | Notes |
 |---|---|---|---|
 | Sony "Pulse Elevate" portfolio | Sony products | https://... | Sony uses "Elevate" in audio products; flagged for prelim-register cross-check |
-| Aurora "Borealis" console "Raise Your Play" tagline (prior usage) | Aurora Interactive marketing | https://... | Client's own prior use — note as supportive |
+| Foxglade "Borealis" console "Raise Your Play" tagline (prior usage) | Foxglade Interactive marketing | https://... | Client's own prior use — note as supportive |
 
 ### PR / reputational risk
 

package/driver/skills/prelim-search/SKILL.md

@@ -259,7 +259,7 @@ The driver runs the delivery stage on **every** completed run — the search is
 
 **Methodology lives in the Excel Methodology tab, NOT in the email body.** Do not include paragraph-length descriptions of search approach, sub-skills invoked, variants generated, etc. in the client-facing email. That bloat is what the render-layer split fixes.
 
-**The email is a COVER NOTE; the HTML report is the single master document (wp50).** The house default for every customer: the email carries only the headline risk tier, the report link, and any surviving scope/verdict/handling flags plus reviewer's open questions — no findings body, no client-voice prose, no hand-holding preamble. Client-voice findings live in the **HTML report** (the one surface the client reads). The last exception — Aurora's review-table overlay (`delivery.email:'table'`) — is DELETED as of 2026-07-28, along with the knob that asked for it: a second full findings surface in the mail was the seam that let per-conflict tiers diverge from the report, and on the knockout lane it shipped internal purple notes over the wire. A lawyer who needs a bespoke, forwardable client mail asks the assistant to draft one from the run's `report-data.json` — formatting is judgement, and it belongs where judgement lives, not in a profile enum. The operative delivery spec is [delivery-contract.md](delivery-contract.md) + the driver's `report-synthesis` stage. (The `client-summary` stage, which re-voiced the findings into a separate client-facing document, is RETIRED as of 2026-08-01 — the one report is the one surface.)
+**The email is a COVER NOTE; the HTML report is the single master document (wp50).** The house default for every customer: the email carries only the headline risk tier, the report link, and any surviving scope/verdict/handling flags plus reviewer's open questions — no findings body, no client-voice prose, no hand-holding preamble. Client-voice findings live in the **HTML report** (the one surface the client reads). The last exception — a customer's review-table overlay (`delivery.email:'table'`) — is DELETED as of 2026-07-28, along with the knob that asked for it: a second full findings surface in the mail was the seam that let per-conflict tiers diverge from the report, and on the knockout lane it shipped internal purple notes over the wire. A lawyer who needs a bespoke, forwardable client mail asks the assistant to draft one from the run's `report-data.json` — formatting is judgement, and it belongs where judgement lives, not in a profile enum. The operative delivery spec is [delivery-contract.md](delivery-contract.md) + the driver's `report-synthesis` stage. (The `client-summary` stage, which re-voiced the findings into a separate client-facing document, is RETIRED as of 2026-08-01 — the one report is the one surface.)
 
 Structure (content shared across templates):
 

package/driver/skills/prelim-search/report-prose.md

@@ -19,6 +19,88 @@ writes lives in that stage's own file — the finding sentence and the grouped r
 [synthesis-rules.md](synthesis-rules.md), the card and section shape in
 [delivery-contract.md](delivery-contract.md). What is below is general and is stated only here.
 
+## Two registers, and the reader of the first one is not a lawyer
+
+**The reader is a lawyer who layers advice on top — and that lawyer's client reads the same page.** The
+band, the summary, the basis line and the one-liners are the whole product for the second reader, and
+they were the hardest text on it: single sentences of seventy-odd words in the vocabulary of the
+profession.
+
+So the page has two registers, and which one a line is written in is decided by whether the reader has
+to open something to see it.
+
+**DEFAULT-VISIBLE TEXT CARRIES NO LEGAL OR ENGINE VOCABULARY.** Default-visible means anything a reader
+meets before opening a fold: the summary and the opening line, the basis for the rating, the points for
+and against and what would soften them, the one sentence under each name, the four answers, the
+reviewer's notes, the caveats, the coverage rows and the lines saying what needs a decision. Ordinary
+nouns. One idea per sentence. **No visible sentence over 25 words.** The conclusion first.
+
+Which stored fields those are is not prose's business and is not listed here — `DEFAULT_VISIBLE_FIELDS`
+in `driver/plain-register.mjs` names them for both products, in one place, so a field renamed once does
+not leave a doctrine file quietly describing a shape that no longer exists.
+
+**INSIDE A FOLD THE PROFESSION'S VOCABULARY IS ALLOWED**, where a plain word would lose precision: the
+card's detail paragraph, provenance, the drill bullets. Even there a term is glossed at first use on the
+page. Nothing is removed from the drill and the workbook is unchanged.
+
+### The swaps, as worked examples rather than a banned list
+
+A list of forbidden words is not the mechanism and must never become one. These are what the plain form
+sounds like; the instruction is always to rewrite the sentence, never to substitute the word and leave
+the rest of a lawyer's sentence standing around it.
+
+| written for a lawyer | what the reader needs |
+|---|---|
+| proprietor | owner |
+| subsisting | live |
+| senior | earlier, or came first |
+| specification | goods list |
+| DELPH-formative | names built on DELPH- |
+| prevail | win |
+| citable prior rights | earlier marks the office can raise against you |
+| vulnerable to a non-use attack | could be cancelled for not being used |
+| on the record as it stands | on what we found |
+| the marks-and-goods comparison | same name, same goods |
+| the confusion comparison meets on every limb | same name, same goods, same shops |
+| belt-and-braces classes | extra classes |
+| dispatch, instructed | the request, what was asked |
+| lane | class, or channel — whichever is meant |
+| chunk | never; it is an engine word |
+
+### The standard, in full sentences
+
+A basis line, rewritten: *"ORBIT is already the name of two satellite-tracking apps on the same app
+stores, and of an established satellite-communications company. Any of them would likely win a dispute
+over this name for this software. The word is a weak mark for these goods, which is why this is High and
+not Very High."*
+
+An answer, rewritten: *"Blocked. An identical earlier mark for the same goods stops registration in
+Switzerland, the EU and the US. Below that, earlier DELPH- marks in EU class 5 and US class 42 can be
+raised against the application."*
+
+A card's detail — allowed the fold's vocabulary, still written plainly: *"Same name, same goods, and
+their Swiss filing came first in every territory we searched. We see no argument against it."*
+
+A note, rewritten: *"Ask the client whether it already uses ORBIT. Its own earlier use would change the
+picture and is not reflected here."*
+
+**Shortening by dropping the reason is not the fix.** The "why" stays, in plain words. A visible line
+that is short because it no longer says why is worse than the long one it replaced.
+
+### What reads this, and what does not
+
+**No delivery gate fails on any of it, and no report is ever marked for it.** The reviewing stage that
+already rejects engine vocabulary applies this test to the default-visible fields and hands back a
+rewrite, in the same pass and the same voice it uses for everything else. A hit is a sentence rewritten.
+It is never a disclosure to the client, never a run failure, and it moves no band, no evidence and
+nothing that is searched — this is presentation.
+
+The reviewer is helped by `driver/plain-register.mjs`, which offers the plain form beside the term it
+found and never decides anything. **It cannot flag the mark the run is clearing**: half of these words
+are ordinary English and several — PREVAIL, SENIOR, SPECIFICATION — are perfectly good trademarks, so
+every term the run is about is removed from the text before it is read. A check that could not tell a
+mark from the profession's vocabulary would put its noise on the one report where it matters most.
+
 ## Fact · assessment · prescription
 
 Three different things, and only two of them belong in a report.

package/driver/skills/prelim-search/risk-framework-demo.manifest.json

@@ -2,7 +2,7 @@
   "schema_version": 1,
   "framework_key": "demo",
   "title": "Demo Brand Owner risk framework",
-  "source_deck": "Demo Brand Owner risk framework, doc 50 shape",
+  "source_deck": "Demo Brand Owner risk framework, authored in-house for the demonstration account",
   "entity_label": "Demo Brand Owner",
   "bands": [
     { "label": "Very High", "tone": "severe" },

package/driver/skills/prelim-search/risk-framework-triage.manifest.json

@@ -2,7 +2,7 @@
   "schema_version": 1,
   "framework_key": "house-triage",
   "title": "House knockout triage ladder (Stage 0)",
-  "source_deck": "Ported from the interactive knockout-searches skill's 5-tier rating system (SKILL.md rating table + calibration rules), doc 50 shape",
+  "source_deck": "Cordillera house triage scale for knockout screening, authored in-house; five rating tiers",
   "entity_label": "the client",
   "bands": [
     { "label": "Very High", "tone": "severe" },

package/driver/skills/prelim-search/risk-framework.manifest.json

@@ -2,7 +2,7 @@
   "schema_version": 1,
   "framework_key": "house-default",
   "title": "Generic default risk framework",
-  "source_deck": "Generic house default — IP Risk Assessment Framework.pptx (Privileged & Confidential), completed by the reviewing lawyer 2026-08-31; supersedes the 2026-07-05 transcription (doc 50), which was partial",
+  "source_deck": "Cordillera house risk framework, authored in-house and reviewed by counsel; 2026-08-31 revision",
   "entity_label": "the company",
   "bands": [
     { "label": "Very High", "tone": "severe" },

package/driver/skills/prelim-search/synthesis-rules.md

@@ -14,7 +14,7 @@ Across the synthesis, two postures are load-bearing:
 
 3. **Optics is annotation, not a rating.** Partner-relationship, channel-conflict, PR, and reputational concerns — however prominently `matter-context` or the manifest raise them — are surfaced as client-facing annotations (the separate PR / reputational section below; a "relationship-level heads-up"). PR is a separate category, **never rated on the framework's bands**, and an existing business relationship is an *either-way* practical factor, not a band multiplier. The rating answers what the framework in force asks — anchored in *if this party sued us, would they win?*, set by the consumer-confusion read (mark-as-whole × G&S). Never let optics/PR/partner/size move the band; annotate beside it.
 
-4. **The framework in force rates the matter (doc 50).** Every run reads exactly one risk framework — **the customer's own framework if the profile has one on file, else the house default. Nothing in between.** That framework is the customer's own legal judgment written down: reason each conflict *with* it — its band definitions (Legal position × Practical position × Potential outcomes), or its matrix where it states one — and state the rating as **its band word, verbatim**. It is a reasoning authority, never a lookup table: no numeric thresholds, no score-to-band mapping, no vocabulary borrowed from another framework. **The band follows the words** — state the likelihood in plain words first (*"the prior owner is likely to win, though credible defences exist"*), then give the band those words require under the framework's own definitions; the prose read and the band are one judgment and may not disagree. Where the framework states **ceilings or matrix mappings, honour them exactly as written** — they are that framework's own anti-escalation mechanism, and no practical or optics factor lifts a rating past what its stated method yields. A conflict the client **clearly wins with no material risk is not a rated conflict at all** (most frameworks' lowest band still describes *real* residual risk — nuisance claims, weak strategic complaints, registration obstacles — never clear wins): surface it unrated as commercial awareness if worth knowing, else omit it. Voice the client side as the framework's **entity** names it (*"the company"*, *"Zephyr/Nimbus/Kestrel"*, *"Aurora Interactive"*). The three-question discipline stays: **legal risk** (*would they win?* — the confusion read net of merits defences), **practical risk** (*would they act?* — enforcement reality), and **impact** (*if they did, what follows?*). The framework's band definitions integrate the first two — reason both, in that order, through its own terms. **Impact stays client-surfaced, never rating-moving**: name the consequences for the client to weigh (we do not conclude acceptability), beside the band, exactly as before.
+4. **The framework in force rates the matter (doc 50).** Every run reads exactly one risk framework — **the customer's own framework if the profile has one on file, else the house default. Nothing in between.** That framework is the customer's own legal judgment written down: reason each conflict *with* it — its band definitions (Legal position × Practical position × Potential outcomes), or its matrix where it states one — and state the rating as **its band word, verbatim**. It is a reasoning authority, never a lookup table: no numeric thresholds, no score-to-band mapping, no vocabulary borrowed from another framework. **The band follows the words** — state the likelihood in plain words first (*"the prior owner is likely to win, though credible defences exist"*), then give the band those words require under the framework's own definitions; the prose read and the band are one judgment and may not disagree. Where the framework states **ceilings or matrix mappings, honour them exactly as written** — they are that framework's own anti-escalation mechanism, and no practical or optics factor lifts a rating past what its stated method yields. A conflict the client **clearly wins with no material risk is not a rated conflict at all** (most frameworks' lowest band still describes *real* residual risk — nuisance claims, weak strategic complaints, registration obstacles — never clear wins): surface it unrated as commercial awareness if worth knowing, else omit it. Voice the client side as the framework's **entity** names it (*"the company"*, *"Coastline/Nimbus/Kestrel"*, *"Foxglade Interactive"*). The three-question discipline stays: **legal risk** (*would they win?* — the confusion read net of merits defences), **practical risk** (*would they act?* — enforcement reality), and **impact** (*if they did, what follows?*). The framework's band definitions integrate the first two — reason both, in that order, through its own terms. **Impact stays client-surfaced, never rating-moving**: name the consequences for the client to weigh (we do not conclude acceptability), beside the band, exactly as before.
 
 ## Contents
 

package/driver/skills/prelim-variants/SKILL.md

@@ -125,8 +125,8 @@ LUMENGARDE is the distinctive anchor — risk concentrates on coined-word axes a
 
 **A variant VALUE is a mark term, never a note about one.** A parenthetical, sentence punctuation
 (`—`, `;`) or a space-flanked slash makes it a label, and a label dispatched verbatim returns a
-confident zero over marks that exist — a nil search that reads as a clean. Write `ZEPHYR`, not
-`ZEPHYR (root)`; the rationale column is where the note belongs. The compiler refuses annotated values
+confident zero over marks that exist — a nil search that reads as a clean. Write `COASTLINE`, not
+`COASTLINE (root)`; the rationale column is where the note belongs. The compiler refuses annotated values
 whatever their length, so a two-word one is caught the same way a long one is.
 
 **If a value genuinely IS a mark carrying that punctuation** — a device mark recorded with its Vienna
@@ -178,7 +178,7 @@ silently narrows what was searched, and the deferred row is the honest outcome.
 
 ### Watchlists
 
-- **Aggressive enforcers:** Nordwave, Aurora Interactive
+- **Aggressive enforcers:** Nordwave, Foxglade Interactive
 - **Major brand owners:** Sony, Aureon, Nintendo, HP, Activision, Take-Two, EA
 - **Competitors:** Epic Games, Valve, Unity, Riot Games
 

package/driver/stages.mjs

@@ -1301,7 +1301,7 @@ export const STAGES = {
       job.jurisdictions ? `Instructed territories (AUTHORITATIVE scope — do NOT widen to "major markets"): ${Array.isArray(job.jurisdictions) ? job.jurisdictions.join(", ") : job.jurisdictions}` : "",
       job.customer ? `Customer/applicant (from the intake brief — drives the self-exclusion set): ${job.customer}` : "",
       // WS-B profile defaults — falsy-omitted, so an empty profile renders this message byte-identical
-      // to the pre-profile shape (the aurora/generic regression anchor).
+      // to the pre-profile shape (the regression anchor).
       // Industry is CONTEXT not a rule (Design Law #1): it sharpens which sectors/adjacencies matter for
       // the vertical (e.g. food/ingestible adjacency for a beverage brand), it never dictates a conclusion.
       profile?.industry ? `Customer industry (context for sector framing — let it sharpen the relevant sectors and adjacencies for this vertical; it is context, never a rule that decides a finding): ${profile.industry}.` : "",
@@ -1813,7 +1813,7 @@ export const STAGES = {
       // DETERMINISTIC GRID (robust fix, 2026-06-14): when the driver wrote a grid-spec, the model is OUT
       // of the grid data path entirely — it passes grid_spec_path, the plugin runs the dictated cells and
       // WRITES common-law-grid.json from the API response (no truncation, no dropped/mis-keyed cells), and
-      // the model only judges the returned candidates. This is the structural cure for both the Zephyr
+      // the model only judges the returned candidates. This is the structural cure for both the dense-marketplace
       // truncation and the NOVA PULSE dropped-cell failures (no tier bump would fix the output ceiling).
       if (gridSpecPath) {
         return lines(
@@ -2353,7 +2353,7 @@ export const STAGES = {
       },
       "placements.json — the structured mirror, keys EXACTLY {mark, owner, jurisdiction, records, tier, reason} + optional borderline": {
         class: "mechanical:code-rendered", tokens: ["placementmodel_missing", "placements_unparseable", "placements_key_unknown", "placement_invalid", "placement_key_unknown"],
-        why: "The driver renders it: renderPlacementsJson() in placement-form.mjs over the union, landed by gateway.mjs:706 (the #562 union-then-render block; re-verified 2026-08-29 — the old :507 predated this branch and pointed at the engine-resolution doc comment). The skill file was not updated with #562, so the stage's two sources contradict each other — the contract that escapes if E1 is authored against stages.mjs alone. [citation unverified]",
+        why: "The driver renders it: renderPlacementsJson() in placement-form.mjs over the union, landed by `syncPlacementForm` in gateway.mjs (the union-then-render block). Cited by SYMBOL because the number has now moved twice: an earlier :507 pointed at an engine-resolution comment, and :706 went blank when an unrelated block was inserted above it. The skill file was not updated with #562, so the stage's two sources contradict each other — the contract that escapes if E1 is authored against stages.mjs alone.",
       },
       "mark / owner / records / territories / classes on a SELECTED row": {
         class: "mechanical:code-extracted", tokens: [],
@@ -3222,7 +3222,7 @@ export const STAGES = {
       // lane's own prose contract. Two of the four longest sentences in the delivered report were
       // coverage/gap prose, and NOTHING governed it: the code-stamped `coverage_line:` front-matter
       // (scope-facts.mjs,) is EXCLUDED from predelivery-lint's prose scan by design
-      // (stripFrontMatterBlock, predelivery-lint.mjs:374), so a coverage number re-typed into prose
+      // (stripFrontMatterBlock, predelivery-lint.mjs:375), so a coverage number re-typed into prose
       // beside it is caught only if the prose form itself trips SCOPE_NUMBER_RE — which the narrative's
       // own phrasings routinely dodge. Prompt-only per §7; the fix is to stop authoring the duplicate.
       `COVERAGE PROSE (the lane that runs longest — hold it to the house budgets): the register coverage line a reader sees is COMPUTED from this run's own record and STAMPED BY CODE as front-matter (the proportion, the class states, the searched registers). Do NOT re-type its numbers anywhere in prose — not in a coverage[] note, not in the coverage_judgment reason, not in the narrative. Nothing catches the duplicate for you (the code-stamped line sits outside the prose checks precisely because ITS numbers are the authoritative ones), so a re-typed count does not disagree with the record — it silently drifts from it a redelivery later. Carry the SUBSTANCE and drop the number: "the remaining forms are non-Latin script" says the useful half; the code says how many. State each coverage fact ONCE, in ONE place — an area's state belongs in its coverage[] row, the sufficiency read belongs in coverage_judgment.reason, and neither is re-narrated in the other or in the findings. And say WHICH KIND of negative you hold every time: a source this run actually queried and got nothing from reads "searched — none found"; a source it did not reach reads "not searched this run" or "could not be searched — <the reason>". The same source must never wear both readings in one report.`,