Keep one invalidation table, and derive the inverse a UI wants

There were two answers in the tree to "which stored bytes stop being valid when
this knob moves", and only one of them was checked.

`flow/address/block-knobs` is per block and asserted by biconditional —
`address-test` re-freezes the take once per knob and requires that the bytes
changed if and only if the key did. `domain/params`'s `:affects` was per area,
had no caller but a test asserting it returned what it was written as, and was
already wrong in both directions on the one entry where the two granularities
disagree: `:aperture-cut` claimed `#{:mouth}`, where it reaches no block, and
omitted the teeth, whose contour bytes it genuinely moves by gating
`condition/interior`'s smoothing. `:blink-cut` claimed `#{:eye}` and reaches no
block either, because a blink is `[:vis]` keys in tier 1.

So `:affects` and `affected-areas` are gone, and `address/knob-roles` is the
derived inverse of the table that is asserted — which is what a parameter panel
actually wants to ask. A knob absent from it invalidates no block, and that is
an answer rather than a gap.

Two new assertions keep the derivation from rotting at either edge: every role
in the table is reachable from some knob, and every knob a block declares is one
the registry defines. The second closes a real hole — `block-descriptor` checks
only that a knob was PASSED, and the freeze's `merge take/knobs` makes that true
of anything spelled like a keyword, so a typo in `block-knobs` would have named a
setting no slider can move.

228 CLJS tests, green.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

docs/architecture.md

@@ -204,6 +204,20 @@ bug waiting for a bigger project:
 This table is the argument for the split, and it should be derivable from the
 sub graph rather than maintained by hand.
 
+As built, half of it is, by a different route. `flow/address/block-knobs` is this
+table for tier 2 — per block, the settings its bytes depend on — and
+`address-test` asserts it by biconditional rather than deriving it, which is
+stronger than a derivation would have been: a derived table is only as right as
+the graph it reads. What is not written down anywhere is the rest of the column.
+Stages 3 to 5 are one call chain in `flow/take/measure` rather than a chain of
+subs, so there is no graph above `::base-scene` to read a dependency off, and the
+tier-1 half of a re-freeze is therefore whole-clip.
+
+An earlier arrangement had a second table in `domain/params`, per feature area,
+saying which areas a knob affected. It disagreed with the tested one on the first
+entry where the two granularities part company and it had no caller; see that
+namespace for why one checked table beats two.
+
 | Knob | Invalidates from |
 | --- | --- |
 | model / footage | 2 detect |

frontend/src/arthur/domain/params.cljs

@@ -1,12 +1,30 @@
 (ns arthur.domain.params
-  "Definitions for generated settings. Defaults, scope and value constraints live
-  here once; UI metadata and regeneration rules can be added to each entry.")
+  "Definitions for generated settings: defaults, scope and value constraints, once.
+
+  WHAT IS NOT HERE. A knob's INVALIDATION — which stored bytes stop being valid
+  when it moves — is `arthur.flow.address/block-knobs`, and it is there rather
+  than here for two reasons that are not about layering.
+
+  It is per BLOCK and this registry is per AREA, and the difference is not
+  granularity, it is disagreement. `:aperture-cut` is a mouth setting that reaches
+  the TEETH block's bytes and does not reach the mouth's, because it gates the
+  contour smoothing and the mouth's own geometry is smoothed either way.
+  `:blink-cut` is an eye setting that reaches no block at all, because a blink is
+  `[:vis]` keys in tier 1. An `:affects #{:eye}` on that entry read as documentation
+  and was wrong in both directions.
+
+  And `block-knobs` is TESTED — `address-test` re-freezes the take once per knob
+  and asserts the biconditional — while a field here could only be believed. An
+  earlier version of this namespace carried `:affects` and an `affected-areas`
+  reading it, and the only thing that ever called it was a test asserting it
+  returned what it was written as. Two tables where one is checked and one is not
+  is worse than one table, because the unchecked one is the one a parameter UI
+  would reach for first. `address/invalidates` is the derived inverse, and it
+  cannot drift from the thing that is asserted.")
 
 (def definitions
-  {:anchor-avg    {:area :subject :default 2 :type :integer :min 0
-                   :affects #{:head :mouth :eye :brow :teeth}}
-   :contour-avg   {:area :subject :default 1 :type :integer :min 0
-                   :affects #{:mouth :eye :brow}}
+  {:anchor-avg    {:area :subject :default 2 :type :integer :min 0}
+   :contour-avg   {:area :subject :default 1 :type :integer :min 0}
    :verts         {:area :mouth :default 8 :type :integer :min 4 :even? true}
    :aperture-cut  {:area :mouth :default 0.12 :type :number :min 0 :max 1}
    :eye-verts     {:area :eye :default 8 :type :integer :min 4 :even? true}
@@ -34,10 +52,6 @@
 
 (def areas #{:subject :mouth :eye :brow :teeth})
 
-(defn affected-areas [id]
-  (when-let [spec (get definitions id)]
-    (or (:affects spec) #{(:area spec)})))
-
 (defn for-area [wanted-area]
   (into {} (keep (fn [[id {:keys [area default]}]]
                    (when (= wanted-area area) [id default])))

frontend/src/arthur/flow/address.cljs

@@ -152,6 +152,28 @@
    "teeth"      [:anchor-avg :aperture-cut :blob-grow :cavity-erode :min-area
                  :teeth-on :teeth-smooth :teeth-verts :tongue-reject :top-bias]})
 
+(def knob-roles
+  "knob -> the roles whose bytes it moves. The inverse of `block-knobs`, DERIVED.
+
+  This is what a parameter UI wants — \"what stops being valid if I drag this\" —
+  and deriving it is the whole point: `block-knobs` is the table `address-test`
+  asserts by biconditional, so an answer computed from it cannot drift from an
+  answer that is checked, and an answer written down beside it could.
+
+  A knob absent from this map invalidates NO BLOCK, and that is a real answer
+  rather than a gap. `:blink-cut`, `:iris-size` and `:pupil-size` are all absent,
+  because a blink is `[:vis]` keys and the two sizes are framed channels: tier 1,
+  editable, and rewritten by a re-freeze without a byte of tier 2 moving."
+  (reduce (fn [m [role knobs]]
+            (reduce (fn [m knob] (update m knob (fnil conj #{}) role)) m knobs))
+          {}
+          block-knobs))
+
+(defn invalidates
+  "The roles one knob's bytes depend on, or an empty set."
+  [knob]
+  (get knob-roles knob #{}))
+
 (defn block-descriptor
   "The canonical text naming one dense block.
 

frontend/test/arthur/domain/feature_test.cljs

@@ -26,9 +26,6 @@
     (is (= :person-4 (get-in scene [:features :eye-8 :subject])))))
 
 (deftest parameter-definitions-drive-the-current-take-defaults
-  (is (= #{:eye} (params/affected-areas :blink-cut)))
-  (is (= #{:head :mouth :eye :brow :teeth}
-         (params/affected-areas :anchor-avg)))
   (is (params/valid-settings? :eye {:blink-cut 0.2}))
   (is (not (params/valid-settings? :eye {:verts 8}))))
 

frontend/test/arthur/flow/address_test.cljs

@@ -113,6 +113,31 @@
                              (str "add " id " to block-knobs for " (pr-str role))
                              (str "remove " id " from block-knobs for " (pr-str role)))))))))))
 
+(deftest the-knob-inverse-is-derived-and-total
+  ;; `invalidates` is the parameter UI's question — "what stops being valid if I
+  ;; drag this" — and the reason it is derived rather than written down is that
+  ;; the thing it is derived FROM is the thing the biconditional above asserts.
+  ;; These two assertions are what keep the derivation honest in the only two ways
+  ;; it could rot.
+  (testing "every role in the table is reachable from some knob"
+    (is (= (set (keys address/block-knobs))
+           (into #{} (mapcat val) address/knob-roles))))
+  (testing "every knob a block declares is a knob the registry defines"
+    ;; The other direction of the same edge: a typo in `block-knobs` would name a
+    ;; setting no slider can move, and `block-descriptor` only checks that the
+    ;; knob was PASSED, which the freeze's `merge take/knobs` makes true of
+    ;; anything spelled like a keyword.
+    (is (empty? (remove #(contains? params/definitions %)
+                        (into #{} (mapcat val) address/block-knobs)))))
+  (testing "a knob that moves no bytes says so, rather than having no answer"
+    ;; Tier 1 knobs. Each one rewrites keys or a framed value on a re-freeze and
+    ;; no block's address changes, which is why the empty set is the right answer
+    ;; and not a missing entry.
+    (doseq [knob [:blink-cut :iris-size :pupil-size]]
+      (is (= #{} (address/invalidates knob)) (str knob)))
+    (is (= #{"teeth"} (address/invalidates :aperture-cut))
+        "the mouth's threshold reaches the teeth block and no other")))
+
 (deftest the-teeth-table-is-asserted-at-the-descriptor
   ;; Every knob the teeth block declares must reach its key. The other half — that
   ;; each one also moves its bytes — needs real pixels, because the crop, the otsu