artdag: Phase 7 cost-directed opt-improvement/opt-cheaper? + 5 tests

artdag/opt-improvement compares the original output cone (dce to id) vs the
maude-reduced DAG under an injected cost-fn, returning before/after total-work and
critical-path. opt-cheaper? asserts optimisation never increases cost: the 5-node
chain drops to 2 (work 5->2, path 5->2) and stays cheaper under radius-weighted cost
(5->3); over dedup and untouched DAGs are never pessimised. Consumes cost.sx. Phase 7
base + (later) cost box done. maude-optimize 38/38, total 196/196.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

lib/artdag/optimize-rules.sx

@@ -186,3 +186,28 @@
     (let
       ((o (artdag/opt-reduce dag id)))
       (artdag/opt-last (artdag/dag-order o)))))
+
+; ---- cost-directed: the maude-optimised cone never costs more than the original ----
+; compares the original output cone (dce to id) against the maude-reduced DAG under an
+; injected cost-fn (op params). Monotone per-node costs => optimisation is never a
+; pessimisation: fewer nodes (DCE/dedup) and fused ops (one blur(M+N) for two blurs).
+
+(define
+  artdag/opt-improvement
+  (fn
+    (dag id cost-fn)
+    (let
+      ((orig (artdag/dce dag (list id))) (opt (artdag/opt-reduce dag id)))
+      {:before (artdag/total-work orig cost-fn)
+        :after (artdag/total-work opt cost-fn)
+        :before-path (artdag/critical-path orig cost-fn)
+        :after-path (artdag/critical-path opt cost-fn)
+        :optimized opt})))
+
+(define
+  artdag/opt-cheaper?
+  (fn
+    (dag id cost-fn)
+    (let
+      ((imp (artdag/opt-improvement dag id cost-fn)))
+      (<= (get imp :after) (get imp :before)))))

lib/artdag/scoreboard.json

@@ -10,9 +10,9 @@
     "serialize": {"pass": 13, "fail": 0},
     "stats": {"pass": 12, "fail": 0},
     "fault": {"pass": 14, "fail": 0},
-    "maude-optimize": {"pass": 33, "fail": 0}
+    "maude-optimize": {"pass": 38, "fail": 0}
   },
-  "total_pass": 191,
+  "total_pass": 196,
   "total_fail": 0,
-  "total": 191
+  "total": 196
 }

lib/artdag/scoreboard.md

@@ -14,5 +14,5 @@ _Generated by `lib/artdag/conformance.sh`_
 | serialize | 13 | 0 | 13 |
 | stats | 12 | 0 | 12 |
 | fault | 14 | 0 | 14 |
-| maude-optimize | 33 | 0 | 33 |
+| maude-optimize | 38 | 0 | 38 |
-| **Total** | **191** | **0** | **191** |
+| **Total** | **196** | **0** | **196** |

lib/artdag/tests/maude-optimize.sx

@@ -268,3 +268,36 @@
   (artdag/node-params
     (artdag/dag-get mo-plain-opt (artdag/opt-last (artdag/dag-order mo-plain-opt))))
   {:radius 3})
+
+; ---- cost-directed: optimisation never increases cost ----
+
+(define
+  mo-rcost
+  (fn (op params) (if (= op "blur") (max 1 (get params :radius)) 1)))
+
+(artdag-test
+  "opt-improvement: const-cost total work drops on fused chain"
+  (let ((imp (artdag/opt-improvement mo-chain5 mo-chain5-id artdag/const-cost)))
+    (list (get imp :before) (get imp :after)))
+  (list 5 2))
+
+(artdag-test
+  "opt-improvement: critical path shrinks under const cost"
+  (let ((imp (artdag/opt-improvement mo-chain5 mo-chain5-id artdag/const-cost)))
+    (< (get imp :after-path) (get imp :before-path)))
+  true)
+
+(artdag-test
+  "opt-cheaper?: fused chain is cheaper under radius-weighted cost"
+  (artdag/opt-cheaper? mo-chain5 mo-chain5-id mo-rcost)
+  true)
+
+(artdag-test
+  "opt-cheaper?: over dedup is cheaper"
+  (artdag/opt-cheaper? mo-dedup mo-dedup-id artdag/const-cost)
+  true)
+
+(artdag-test
+  "opt-cheaper?: untouched DAG keeps equal cost (never a pessimisation)"
+  (artdag/opt-cheaper? mo-plain mo-plain-id artdag/const-cost)
+  true)

plans/artdag-on-sx.md

@@ -31,7 +31,7 @@ edges.
 
 ## Status (rolling)
 
-`bash lib/artdag/conformance.sh` → **191/191** (11 suites: dag, analyze, plan, execute, optimize, fed, cost, serialize, stats, fault, maude-optimize)
+`bash lib/artdag/conformance.sh` → **196/196** (11 suites: dag, analyze, plan, execute, optimize, fed, cost, serialize, stats, fault, maude-optimize)
 
 Base roadmap (Phases 1–6) COMPLETE. Now extending.
 
@@ -172,7 +172,10 @@ declaratively and prove it equivalent to the hand-written `optimize.sx`.
   needs. API also: `mau/critical-pairs`, `mau/joinable?`.
 - [x] `lib/artdag/tests/maude-optimize.sx` — bridge round-trip, each law,
   result-preserving equivalence, `(mau/confluent? rules-module)` holds (33 tests).
-- [ ] (later) cost-directed choice among confluent-equivalent forms; (optional)
+- [x] cost-directed: `artdag/opt-improvement`/`opt-cheaper?` compare the optimised
+  cone vs the original cone under an injected `cost-fn` — optimisation is never a
+  pessimisation (fewer nodes + fused ops ⇒ total-work and critical-path never
+  increase, under const and radius-weighted costs). (optional, not pursued)
   miniKanren scheduling.
 
 maude is a READ-ONLY consumed substrate (like datalog/persist) — load it, don't
@@ -188,6 +191,18 @@ be an op token.
 
 ## Progress log
 
+- **2026-06-19 Phase 7 — cost-directed: optimisation is never a pessimisation**
+  (maude-optimize 38/38, total 196/196). `artdag/opt-improvement dag id cost-fn` compares
+  the original output cone (`artdag/dce` to `id`) against the maude-reduced DAG under an
+  injected `cost-fn (op params)` — returns `{:before :after :before-path :after-path
+  :optimized}` (total-work + critical-path each side). `artdag/opt-cheaper?` asserts
+  `after <= before`. Under monotone per-node costs the optimised DAG never costs more:
+  the 5-node chain drops to 2 (const work 5→2, critical path 5→2) and stays cheaper under
+  a radius-weighted cost (5→3 — one `blur(M+N)` costs the same as the two it replaces);
+  the `over` dedup and an untouched DAG are both `opt-cheaper?`. Consumes `cost.sx`'s
+  `total-work`/`critical-path`. Phase 7 base + the "(later)" cost box now done; only the
+  optional miniKanren scheduling remains.
+
 - **2026-06-19 Phase 7 — opt-reduce: bridge normal form back to a DAG** (maude-optimize
   33/33, total 191/191). `artdag/opt-reduce dag id`: encode the DAG cone at `id` into an
   opt-term (`artdag/dag->opt-term` — leaves→nullary const, `:radius` nodes→`op(inputs…,