LLM: fix strict schema 400 — correct providerOptions + chat model

Previous attempt set `strictJsonSchema:false` on createOpenAI()
(factory ignores unknown keys), so wire was still strict:true and
sparse edits {brightness:0.3} were rejected: Missing 'arrangement'.

- src/llm/llmTweak.js: use provider.chat(modelId) (/chat/completions,
  correct for Meta via gateway, not /responses) and pass
  `providerOptions:{ openai:{ strictJsonSchema:false } }` on
  generateObject where the SDK actually reads it.
- src/llm/spec.js: edits fields nullable instead of optional — strict
  requires every key in `required` when `additionalProperties:false`;
  nullable makes required = all 7 (model emits null for unchanged),
  clampEdits already strips nulls so NO_EDITS still works. Prompt now
  instructs "use null for unchanged, example { bpm:null,... }".
- test/llm.test.js: minimal valid result now all-null with one edit.

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

flow-state/src/llm/llmTweak.js

@@ -97,17 +97,17 @@ export async function requestSongTweak({ spec, track, look, hint = '', config =
         baseURL: isDev ? '/__llm/v1' : cfg.gatewayUrl,
         apiKey: cfg.apiKey,
         fetch: isDev ? proxyFetch : undefined,
-        strictJsonSchema: false,
     });
     const modelId = cfg.model;
 
     try {
         const result = await generateObject({
-            model: provider(modelId),
+            model: provider.chat(modelId),
             schema: TweakResultSchema,
             messages,
             temperature: 0.4,
             maxRetries: 0,
+            providerOptions: { openai: { strictJsonSchema: false } },
         });
 
         const raw = result.object || {};

flow-state/src/llm/spec.js

@@ -12,13 +12,13 @@ const ARRANGEMENT_VALUES = Object.keys(ARRANGEMENTS);
 export { ARRANGEMENT_VALUES };
 
 export const TweakEditsSchema = z.object({
-    bpm: z.number().int().min(60).max(180).optional(),
-    brightness: z.number().min(0).max(1).optional(),
-    noise: z.number().min(0).max(1).optional(),
-    noiseColour: z.number().min(0).max(1).nullable().optional(),
-    dynamics: z.number().min(0).max(1).optional(),
-    transition: z.number().min(0).max(1).optional(),
-    arrangement: z.enum(ARRANGEMENT_VALUES).optional(),
+    bpm: z.number().int().min(60).max(180).nullable(),
+    brightness: z.number().min(0).max(1).nullable(),
+    noise: z.number().min(0).max(1).nullable(),
+    noiseColour: z.number().min(0).max(1).nullable(),
+    dynamics: z.number().min(0).max(1).nullable(),
+    transition: z.number().min(0).max(1).nullable(),
+    arrangement: z.enum(ARRANGEMENT_VALUES).nullable(),
 });
 
 export const TweakResultSchema = z.object({
@@ -106,8 +106,9 @@ export function buildTweakPrompt({ spec, track, look, lookDigest = null, hint =
         '- Prefer lowering agitation for breakdowns/intros; keep section count ≥3 if possible.',
         '- Valid bpm 60-180 int, brightness/noise/dynamics/transition/noiseColour 0-1, arrangement one of: ' + ARRANGEMENT_VALUES.join(', ') + '.',
         '- noiseColour may be null (follows brightness) or 0-1; leave it null unless you have a reason.',
+        '- All 7 edit fields MUST be present in edits — use null for unchanged (not omitted). Example: { bpm: null, brightness: 0.6, noise: null, noiseColour: null, dynamics: null, transition: null, arrangement: null }',
         '- Explain your reasoning in 1-2 sentences (rationale).',
-        '- If the look is already good, return empty edits {}.',
+        '- If the look is already good, return all-null edits.',
         hintBlock,
     ].join('\n');
 

flow-state/test/llm.test.js

@@ -52,8 +52,13 @@ test('buildTweakPrompt truncates long hint in system', () => {
 });
 
 test('TweakResultSchema validates a minimal valid result', () => {
-    const parsed = TweakResultSchema.parse({ rationale: 'Lower noise to calm the drop.', edits: { noise: 0.3 }, confidence: 0.8 });
+    const parsed = TweakResultSchema.parse({
+        rationale: 'Lower noise to calm the drop.',
+        edits: { bpm: null, brightness: null, noise: 0.3, noiseColour: null, dynamics: null, transition: null, arrangement: null },
+        confidence: 0.8,
+    });
     assert.equal(parsed.edits.noise, 0.3);
+    assert.equal(parsed.edits.bpm, null);
     assert.ok(parsed.rationale.length > 0);
 });