Add MiMo speech provider target contract

docs/architecture.md

@@ -11,7 +11,8 @@ CultCache/CultMesh/Eve authority map.
 Build a procedural alien-language engine whose outputs are grounded in
 worldview, semantics, grammar, phonology, morphology, and history. As a daemon,
 Weksa also turns writing plus agent state into typed conversational intent, then
-lowers that intent into pronunciation and utterance handoff documents.
+lowers that intent into written realizations, pronunciation plans, speech-provider
+requests, and utterance handoff documents.
 
 ## Current Mechanism
 
@@ -33,6 +34,7 @@ through its ontology.
 - Diachrony owns plausible irregularity.
 - Renderer owns output, not interpretation.
 - Weksa owns conversational intent and utterance lowering.
+- External speech providers own rendered audio, not meaning or delivery truth.
 - AquaSynth owns learned utterance embeddings, synth controls, Faust
   compilation, and audio output.
 
@@ -41,8 +43,10 @@ through its ontology.
 Future implementation should turn the pipeline in `state/map.yaml` into a small,
 inspectable daemon. The first executable slice should validate a tiny
 interlingua packet, read a tiny language project definition, render glossed
-output, lower an accepted intent into a pronunciation/utterance handoff, and
-produce a trace explaining which stage made each decision.
+output, lower an accepted intent into target-language realizations, optionally
+project the accepted realization into a speech-provider request or
+pronunciation/utterance handoff, and produce a trace explaining which stage made
+each decision.
 
 Runtime should be fast and deterministic. Model-assisted English decomposition
 belongs in authoring tools or offline import flows, not in the required hot path.

docs/fixtures/weksa-provider-advertisement.json

@@ -2,7 +2,7 @@
   "schema": "gamecult.eve.provider_advertisement.v1",
   "providerId": "weksa.intent.service",
   "title": "Weksa Intent Service",
-  "description": "Weksa converts writing, interlingua, and agent state into typed conversational intent, pronunciation plans, and utterance handoff documents for AquaSynth and other consumers.",
+  "description": "Weksa converts writing, interlingua, and agent state into typed conversational intent, target-language realizations, pronunciation plans, speech-provider requests, and utterance handoff documents for AquaSynth and other consumers.",
   "mode": "fixture-read-only",
   "serviceVerse": {
     "id": "verse.weksa.service.v1",
@@ -12,6 +12,7 @@
       "typed conversational intent",
       "pronunciation plans",
       "utterance handoff documents",
+      "speech-provider request projections",
       "lowering traces",
       "language-project and flavor-profile state"
     ],
@@ -19,6 +20,7 @@
       "AquaSynth learned embeddings",
       "Faust compilation",
       "rendered audio",
+      "MiMo provider audio",
       "Persona canonical state"
     ]
   },
@@ -51,6 +53,15 @@
         "weksa.utterance_embedding_handoff.v0.1"
       ]
     },
+    {
+      "id": "weksa.speech_provider",
+      "kind": "speech_provider",
+      "cultMeshKey": "weksa.speech-provider/requests",
+      "authority": "Weksa speech-provider request projection",
+      "schemas": [
+        "weksa.mimo_tts_request.v0"
+      ]
+    },
     {
       "id": "weksa.operator",
       "kind": "operator",
@@ -98,6 +109,14 @@
       ],
       "durableShape": ".weksa/handoff/{utteranceId}.cc"
     },
+    {
+      "family": "speech_provider",
+      "namespace": "weksa.speech_provider.*",
+      "witnesses": [
+        "weksa.mimo_tts_request.v0"
+      ],
+      "durableShape": ".weksa/speech-provider/{requestId}.cc"
+    },
     {
       "family": "trace",
       "namespace": "weksa.trace.*",
@@ -198,11 +217,31 @@
         "audio rendering",
         "AquaSynth synth driver truth"
       ]
+    },
+    {
+      "id": "weksa.command.speech_provider.project",
+      "owner": "Weksa",
+      "acceptedByFixture": false,
+      "allowedInputs": [
+        "accepted target-language realization",
+        "provider-neutral delivery intent",
+        "speech provider profile"
+      ],
+      "emits": [
+        "weksa.mimo_tts_request.v0"
+      ],
+      "forbiddenTargets": [
+        "source interlingua truth",
+        "Persona canonical state",
+        "rendered audio ownership"
+      ]
     }
   ],
   "loweringRules": [
     "Weksa owns conversational intent and utterance lowering.",
+    "Speech-provider requests are target projections from accepted realizations, not source meaning.",
     "AquaSynth owns learned embeddings, synth controls, Faust compilation, and audio output.",
+    "MiMo owns provider rendering and audio output, not Weksa delivery truth.",
     "Eve surfaces submit commands and inspect witnesses; they do not own intent truth.",
     "Fixture exports are discovery metadata until the daemon loop commits typed witnesses."
   ]

docs/mimo-tts-lowering.md

@@ -0,0 +1,123 @@
+# MiMo TTS Lowering
+
+Version: `0.1-draft`
+
+MiMo is an expressive spoken-output target for Weksa interlingua lowering. It is
+not the source of meaning, language choice, character state, or delivery truth.
+
+```text
+interlingua packet
+  -> projected speaker-local context
+  -> target-language utterance realization
+  -> Weksa delivery intent
+  -> MiMo TTS request projection
+  -> rendered audio artifact
+```
+
+MiMo sits beside written output and AquaSynth handoffs. Written realizations
+prove that meaning, register, and cultural ontology survived lowering.
+AquaSynth handoffs train or drive GameCult-owned synthesis. MiMo requests let an
+external provider speak a Weksa-authored utterance with expressive style control.
+
+## Authority
+
+Weksa owns:
+
+- target-language spoken text selected from interlingua
+- target-language register, script, cultural ontology, and trace
+- speaker-local delivery pressure
+- provider-neutral prosody and emphasis intent
+- the MiMo request projection as an export artifact
+
+MiMo owns:
+
+- provider voice selection within the requested language and voice constraints
+- interpretation of MiMo natural-language style guidance
+- interpretation of MiMo inline audio/style tags
+- rendered audio bytes or streaming audio chunks
+
+MiMo does not own:
+
+- source meaning
+- target-language cultural ontology
+- canonical speaker or Persona state
+- Weksa pronunciation plans
+- AquaSynth learned embeddings, synth controls, or audio truth
+
+## Projection Rule
+
+The MiMo request must be derived from an accepted target-language realization.
+It must not lower directly from English unless English is the selected target
+language for that realization.
+
+Natural-language style guidance belongs in the MiMo `user` message. The spoken
+target-language text belongs in the MiMo `assistant` message. Inline MiMo tags
+may appear in the assistant content only when they are derived from Weksa
+delivery intent or explicit target-language realization notes.
+
+## Request Shape
+
+```yaml
+schema_version: weksa.mimo_tts_request.v0
+source_realization_ref:
+speaker_agent_id:
+provider:
+  id: xiaomi-mimo
+  model: mimo-v2.5-tts
+target_language:
+  code:
+  locale:
+  script:
+  register:
+voice:
+  requested_voice:
+  fallback_voice:
+  selection_reason:
+messages:
+  user_style_instruction:
+  assistant_spoken_content:
+audio:
+  format:
+  streaming:
+trace:
+  source_fields:
+  delivery_controls_used:
+  provider_tags_used:
+  uncertainties:
+```
+
+## Delivery Mapping
+
+Weksa should map delivery pressure into provider-neutral labels before emitting
+MiMo-specific instructions:
+
+- pace, pause pressure, and urgency
+- pitch contour and pitch range
+- loudness and attack
+- articulation precision
+- warmth, threat, tenderness, contempt, play, fatigue, secrecy, formality
+- span-level emphasis and pauses
+
+The MiMo projection may turn those labels into director-style guidance or inline
+tags, but the trace must preserve the Weksa-owned source controls.
+
+## Language Rule
+
+Weksa should be able to write and speak in any target language whose utterance
+realization it can justify. MiMo output for Brazilian Portuguese should consume
+the accepted `pt-BR` spoken text. Japanese output should consume accepted
+Japanese text. Alien-language output should consume accepted alien rendered
+speech or phonetic text only after the language project owns that surface.
+
+Backtranslations are reviewer aids. They must not become MiMo spoken content.
+
+## First Proof
+
+The first proof should use the Nibu warning packet already used by multilingual
+lowering:
+
+- emit the accepted English, `pt-BR`, or Japanese realization
+- derive a provider-neutral delivery intent from Nibu's projected context
+- project that into a MiMo `mimo-v2.5-tts` request
+- verify that the request speaks the target text, not the backtranslation
+- verify that style guidance is traceable to Weksa delivery controls

docs/multilingual-utterance-lowering.md

@@ -3,13 +3,15 @@
 Version: `0.1-draft`
 
 Weksa can lower one structured utterance intent into multiple human-language
-realizations without making English the hidden source of truth.
+realizations and spoken-provider requests without making English the hidden
+source of truth.
 
 ```text
 interlingua packet
   -> projected speaker-local context
   -> utterance lowering request
   -> target-language realization packets
+  -> optional speech-provider request projections
 ```
 
 The first targets are English, Brazilian Portuguese (`pt-BR`), and Japanese
@@ -53,7 +55,7 @@ The utterance lowerer does not own:
 - new facts
 - raw agent-state internals
 - target-language alien forms
-- speech synthesis
+- rendered audio
 - phoneme parity or synth-control truth
 
 The target-language cultural ontology profile owns:
@@ -171,6 +173,21 @@ Trace should name the cultural ontology axes that shaped the line, especially
 where the target language requires a social or register choice the source packet
 does not spell out.
 
+## Speech Provider Targets
+
+Speech providers are lowering targets for accepted realizations, not alternate
+meaning sources. A provider request must consume the target-language
+`spoken_text` that Weksa already accepted and must carry a trace back to the
+interlingua packet, projected speaker context, and target-language profile.
+
+MiMo V2.5 TTS is the first external provider target. Its natural-language style
+instruction is a projection of Weksa delivery intent, and its assistant content
+is the selected target-language spoken text. See [MiMo TTS Lowering](./mimo-tts-lowering.md).
+
+AquaSynth remains the GameCult-owned synthesis path for learned utterance
+embeddings and synth-control work. MiMo and AquaSynth are sibling target
+projections from Weksa-authored utterances, not owners of Weksa meaning.
+
 ## First Proof
 
 The first proof uses the existing Nibu warning intent:

docs/utterance-embedding.md

@@ -3,7 +3,8 @@
 Version: `0.1`
 
 The utterance embedding is the compact learned handoff from Weksa-authored line
-intent to AquaSynth voice control.
+intent to AquaSynth voice control. It is one speech target projection among
+others, not the only way Weksa-authored utterances can become sound.
 
 ```text
 interlingua packet
@@ -44,6 +45,11 @@ AquaSynth owns:
 The point is one learned speech handoff, not two little neural courthouses
 arguing about who gets to wear the robe.
 
+External providers such as MiMo may consume a different Weksa target projection:
+accepted target-language `spoken_text` plus provider-neutral delivery intent
+lowered into provider-specific style controls. That path does not replace this
+AquaSynth contract and must not mutate AquaSynth-owned embedding fields.
+
 ## Packet Contract
 
 This is the first fixed training contract. Weksa may evolve later versions, but

examples/speech-provider/mimo/nibu-proof-001/en-US-request.yaml

@@ -0,0 +1,51 @@
+schema_version: weksa.mimo_tts_request.v0
+source_realization_ref: examples/multilingual-lowering/nibu-proof-001/realizations.yaml#nibu-proof-001-en
+speaker_agent_id: nibu
+provider:
+  id: xiaomi-mimo
+  model: mimo-v2.5-tts
+target_language:
+  code: en
+  locale: en-US
+  script: Latn
+  register: sharp technical casual
+  cultural_ontology_refs:
+    - data/target-language-ontology/en-US.yaml
+voice:
+  requested_voice: Chloe
+  fallback_voice: Mia
+  selection_reason: >
+    Use an English built-in voice for the first provider smoke. The requested
+    delivery should preserve Nibu's precise, abrasive, protective cockpit
+    warning without implying that MiMo owns her canonical voice.
+messages:
+  user_style_instruction: >
+    Character: a precise, abrasive, protective ship intelligence speaking
+    through cockpit audio. Scene: the player is about to touch a damaged live
+    control. Guidance: clipped technical casual delivery, contained urgency,
+    low warmth, sharp emphasis on "control" and "nervous system", slight dry
+    contempt, no melodrama, no soothing assistant tone.
+  assistant_spoken_content: >
+    (Serious Sharp Tired)Take your hand off that control unless you want your
+    nervous system to become my least useful circuit diagram.
+audio:
+  format: wav
+  streaming: false
+trace:
+  source_fields:
+    - realizations[realization_id=nibu-proof-001-en].outputs.spoken_text
+    - realizations[realization_id=nibu-proof-001-en].outputs.trace.projected_context_used
+  delivery_controls_used:
+    - clipped technical casual delivery
+    - contained urgency
+    - low warmth
+    - sharp emphasis on unsafe hardware and bodily risk
+    - dry contempt without random cruelty
+  provider_tags_used:
+    - Serious
+    - Sharp
+    - Tired
+  uncertainties:
+    - >
+      Provider-native style tags are a lossy projection. The Weksa trace, not
+      MiMo tags, remains the explanation of the intended delivery.

schemas/mimo-tts-request.md

@@ -0,0 +1,49 @@
+# MiMo TTS Request Contract
+
+This is the human-facing schema receipt for a Weksa-to-MiMo speech synthesis
+request projection.
+
+The packet sits after target-language utterance lowering. It is an export
+artifact for Xiaomi MiMo speech synthesis, not canonical Weksa meaning and not
+an AquaSynth handoff.
+
+## Required Fields
+
+- `schema_version`: currently `weksa.mimo_tts_request.v0`.
+- `source_realization_ref`: accepted Weksa target-language realization being
+  spoken.
+- `speaker_agent_id`: speaker identity.
+- `provider`: provider ID and model ID, initially `xiaomi-mimo` and
+  `mimo-v2.5-tts`.
+- `target_language`: code, locale, script, register, and profile refs copied or
+  referenced from the accepted realization.
+- `voice`: requested provider voice, fallback voice, and selection reason.
+- `messages.user_style_instruction`: MiMo natural-language style control derived
+  from Weksa delivery intent.
+- `messages.assistant_spoken_content`: the exact target-language content to
+  synthesize, optionally with MiMo inline tags derived from delivery controls.
+- `audio`: format and streaming mode.
+- `trace`: source fields, delivery controls used, provider tags used, and
+  uncertainties.
+
+## Boundary
+
+MiMo-specific fields must be projection fields. They must not be copied back
+into interlingua, target-language ontology, Persona state, or AquaSynth-owned
+embedding contracts.
+
+The request may include provider-native style tags when useful, but those tags
+are not the source of truth. Weksa's provider-neutral delivery intent and trace
+remain the inspectable explanation.
+
+## Language Rule
+
+The assistant spoken content must use the accepted target-language realization.
+For `pt-BR`, speak the Portuguese line. For `ja-JP`, speak the Japanese line. For
+English, speak the English line. Do not synthesize a natural backtranslation
+unless that backtranslation is itself the accepted target-language realization.
+
+## Streaming Rule
+
+Streaming requests should use MiMo's stream-compatible audio format. Non-streaming
+fixtures may use an archival format such as `wav`.

schemas/utterance-lowering-packet.md

@@ -1,7 +1,8 @@
 # Utterance Lowering Packet Contract
 
 This is the human-facing schema receipt for lowering one Weksa interlingua
-packet into one or more target-language utterance realizations.
+packet into one or more target-language utterance realizations and optional
+downstream target projections.
 
 Executable schema code does not exist yet. When it does, it must reconcile with
 this document and
@@ -30,6 +31,9 @@ It combines:
 - `projected_context_ref`: source reference for the consumed projection.
 - `lowering_controls`: constraints shared by all target lowerers.
 - `realizations`: non-empty list of target-language outputs.
+- `target_projections`: optional downstream projections derived from accepted
+  realizations, such as AquaSynth utterance embedding handoffs or MiMo TTS
+  requests.
 
 ## Target Language
 
@@ -88,3 +92,8 @@ Trace should identify:
 This packet may feed pronunciation planning or AquaSynth handoff work later, but
 it does not contain phoneme-parity, vocal-tract, reference-synth, or realtime
 voice artifacts.
+
+MiMo TTS requests are allowed as derived target projections. They must reference
+an accepted realization, speak that realization's target-language text, and keep
+provider-native style tags out of source meaning. MiMo owns rendered audio;
+Weksa owns the delivery intent and trace used to create the request.

state/map.yaml

@@ -12,9 +12,10 @@ objective:
     Create a durable workspace where Weksa can develop from a persistent map
     before runtime architecture decisions are made.
   current: >
-    Emit concrete v0.1 utterance handoff artifacts for AquaSynth's first
-    speech-training curriculum while keeping Weksa's ownership limited to
-    inspected inputs.
+    Treat written output, AquaSynth utterance embeddings, and MiMo expressive
+    speech requests as sibling target projections from accepted interlingua
+    lowerings while keeping Weksa's ownership limited to inspected intent,
+    delivery controls, and trace.
 
 constraints:
   - Runtime implementation language is undecided.
@@ -49,6 +50,9 @@ constraints:
     prosody/emphasis hints, and Ghostlight/Epiphany-shaped character-state
     vectors; AquaSynth owns the learned utterance embedding encoder, gradient
     descent, and synth-control decoding.
+  - MiMo V2.5 TTS is an external speech-provider target. Weksa may project an
+    accepted target-language realization plus delivery intent into MiMo style
+    controls, but MiMo owns only provider rendering and audio output.
   - Large offline text-embedding jobs should use `bge-m3` on the neighboring
     Ollama inference box documented in `E:\Projects\gamecult-ops` rather than
     running on the local Weksa workspace by default.
@@ -81,6 +85,9 @@ data_flow:
       owns the target pronunciation layer
     - optionally encode the line as a learned expression vector for downstream
       synth control
+    - optionally project an accepted realization into a provider-specific
+      speech request such as MiMo TTS without making the provider an owner of
+      source meaning
     - apply alien ontology and cultural salience
     - select grammatical commitments
     - derive morphology and lexical families
@@ -272,6 +279,10 @@ landed_baseline:
     English text semantics may use `bge-m3`; IPA strings and alien phonetic
     material belong in PanPhon sequence evidence that AquaSynth compresses into
     the phonetic realization embedding.
+  - MiMo V2.5 TTS is accepted as a speech-provider target projection from
+    Weksa utterance realizations. MiMo requests speak the accepted target text
+    and receive Weksa-derived style guidance; backtranslations and provider
+    tags do not become source meaning.
   - First Nibu interlingua-to-English lowering fixture.
   - First Nibu multilingual utterance lowering fixture at
     `examples/multilingual-lowering/nibu-proof-001/`, covering English,