Nishi Family › Compare › Sovereign Speech: text-to-speech, voice cloning, voice licensing

Nishi Compare · capabilities, evidence and delivery

Sovereign Speech: text-to-speech, voice cloning, voice licensing

Compare capabilities and delivery evidence. Source discovery, qualification, deployment and accepted outcomes are separate; an open contract names work still to verify.

Nishi VOICE-CLONE-001 lane vs Kokoro-82M and Chatterbox and Kyutai TTS and StepAudio 3 Gen

Overview

The purpose, declared position and evidence coverage of this domain. Source presence and completed acceptance are different measures.

Where we are. MEASURED 2026-09-21 by catalogue, capability search and a tree glob, not recalled. The estate already holds the VOICE-CLONE-001 lane in source: nx_pitch (integer YIN F0, rung 0), nx_voiceprint (PARCOR speaker identity plus a distance, rung 1), nx_voice_clone_synth (LPC source-filter speech in a captured voice at any pitch, rung 2), nx_phoneme_synth (formant articulation, rung 5a) and nx_voice_read (a cloned voice reading words by a vocal-tract-length warp, rung 5-FUSE), each with a gate source beside it; nx_g2p (rule grapheme-to-phoneme over lowercase ASCII, 27 phonemes), nx_voice_codec_v2 (LPC-residual codec), nx_postag, and SIX RIFF/WAVE writers in one tree (nx_wav, nx_wav_header, nx_wav_writer, wav_header, wav_writer, nx_audio_wav). A SECOND, older text-to-speech lane exists beside it, found by the ranker's incumbent index after this board first claimed there was none (RETRACTED on the log row): nx_tts_speak (R2.3, an exception dictionary plus letter rules feeding a formant cascade, any English text to a WAV, float arithmetic, a laptop output path hard-coded), nx_tts_seq, nx_tts_word and nx_tts_formant, all SOURCE-ONLY and never built; nx_voice_eval (a reference-free naturalness ruler over F0 statistics, SOURCE-ONLY) is the seed of the sovereign MOS twin; nx_rights_ledger (grant, revoke, state and the nx_rights_is_licensed predicate) and nx_license_check (the licence-tier wall) are the incumbents the voice ledger composes; nx_lipsync (text to visemes) is the avatar reader's seed. Nothing in either lane is built, registered or callable; nx_vocoder, nx_speech and nx_audio are absent; there is no neural stage, no consent or provenance organ, and no board carried either lane until today, so both were dark by construction. The floor is formant and LPC speech: intelligible words, not a narrator. The September 2026 listing, read the same day, says the state of the art is an autoregressive language model over a 12.5 Hz residual-vector-quantised audio tokenizer, its open weights are what our integer engine must learn to run, and its two open problems are long-form reliability and clone provenance, which are exactly this board's product cases [arxiv_tts_2026_09] [arxiv_clone_2026_09].

Where we need to go. OPERATOR 2026-09-21: gospelofwest.com reads the scriptures aloud in generated voices, in popular open-source voices, in family members' uploaded voices and in alpha and beta licensed voices, sovereign from the first byte up, with a /compare board carrying the ladder to state of the art for reading, listening and watching. This board owns the ENGINE: normalisation, lexicon G2P, the audio codec, the decoder, open weights imported as DATA with the licence beside them, consented enrolment, provenance marks, the licence ledger and the lab. /compare/scripturestudy owns the product surface and consumes every rung here. The oracles (UTMOSv2, seed-tts-eval, the preference arena, an outside ASR for WER) run OUTSIDE the build and serve path; only their numbers enter.

The unit. 1 u = one measured session-leg (organ plus gate plus live flip), the standing calibration; the partloft twin of 2026-09-19 (lib, gate 72 of 72, promote, register, roster) was one leg.

Where we are: 11 of 29 rows. The formant and LPC substrate is measured present; the narrator, the neural stage, the voices and the lab are eighteen named contracts.

Cost to the floor: 5 u. Through M1: one WAV writer, a lexicon G2P with scripture names as data, a text normaliser, a whole chapter read by the formant reader as the floor every later voice must beat, and the lab spine that journals WER, similarity and MOS per run.

Cost to a neural narrator: 18 u. Through M2: the RVQ codec twin, the codec-token decoder on the integer engine held to a pre-declared accept rule on seed-tts-eval, and open-weights voices imported as data under their licences.

Cost to family and licensed voices: 22 u. Through M3: consented enrolment with revocation, a provenance mark on every clip, and the licence ledger that refuses synthesis without a live grant.

Cost to the frontier: 30.5 u. Through M4: long-form detect-rollback-regenerate, streaming with word timestamps, style by natural language, cross-lingual transcript-free cloning.

Latest recorded release

2026-09-21 · SP0

speech.matrix, speech.plan, speech.refs admitted to regen.list; nine mirrors pinned by sha256 from nx_research_fetch provenance lines

Release history and work log

Release entries describe recorded changes; they do not establish that every capability passed evaluation.

14 of 35 capability source symbols found|0 of them authored stronger-claims|21 open|coverage 400/1000|adoption 7 full / 7 partial

Evidence profile — what the gaps on this board actually are

Unstamped. No evidence profile is published for this domain yet: run nx_swcompare_evidence speech and this section fills itself on the next beat. No numbers are shown, deliberately — a zero-filled profile would read as a board with no gaps, which is the one wrong answer nobody would question.

Production map

Follow the dependencies, declared acceptance criteria and recorded priorities. Inspect source binding before treating a rank as executable work.

Ranking source binding: PLAN_MATRIX_BOUND_ONLY. Recorded priorities require current acceptance evidence and resource checks before execution.

Ranking matches the captured plan and matrix only. Latest execution outcome, research freshness, accepted delivery and investment return are unverified.

Recorded priority estimates

Order from nx_compare_rank (nx_dr_ocm: (deficit + cost-of-delay + option + enables) x sponsor x self-sufficiency x momentum / cost). FINISH rows are rungs whose symbol is present but whose organ is short of full adoption: listed before new work by this heuristic. Priority is not measured delivery cost or execution readiness. Stamp: # asof=1790082204 domain=speech target_version=0.1 rungs=18 done=0 open=18 finish=0 ranker=nx_dr_ocm

#	Stage	Rung	Priority	Derivation
#1	0.1	Board, bar and census (SP0) sp_board dossier	10000	v=50 m=1 c=5
#2	later	One WAV writer, five retired reversibly (SP1) nx_wav_write_pcm16 dossier	9000	v=45 m=1 c=5
#3	later	Lexicon G2P with scripture names as data (SP2) g2p_lex_lookup dossier	8000	v=80 m=1 c=10
#4	later	Text normalisation for reading aloud (SP3) tn_normalize dossier	7000	v=70 m=1 c=10
#5	later	Register the dark VOICE-CLONE-001 lane on the MCP surface (SP17) sp_lane_registered dossier	7000	v=70 m=1 c=10
#6	later	The floor: a whole chapter through the formant reader (SP4) vr_read_chapter dossier	6000	v=60 m=1 c=10
#7	later	Speech lab spine (SP11) sl_run dossier	5000	v=50 m=1 c=10
#8	later	Neural audio codec twin (SP5) nc_encode dossier	4000	v=160 m=1 c=40
#9	later	Codec-token TTS decoder on the integer engine (SP6) tts_generate dossier	3000	v=180 m=1 c=60
#10	later	Open-weights voices as data under their licences (SP7) vw_import dossier	2000	v=60 m=1 c=30
#11	later	Family voice enrolment with consent and revocation (SP8) ve_enrol dossier	1000	v=15 m=1 c=15
#12	later	Long-form reliability (SP12) tts_longform dossier	1000	v=20 m=1 c=20
#13	later	Every rung an MCP verb, one workflow plane, an agent that runs it (SP16) sp_mcp_surface dossier	400	v=4 m=1 c=10
#14	later	Style by natural language (SP14) tts_style_nl dossier	150	v=3 m=1 c=20
#15	later	Provenance mark on every clip (SP9) vp_mark dossier	133	v=2 m=1 c=15
#16	later	Streaming synthesis with word timestamps (SP13) tts_stream_words dossier	133	v=2 m=1 c=15
#17	later	Licensed voice ledger (SP10) vl_grant dossier	100	v=1 m=1 c=10
#18	later	Cross-lingual transcript-free cloning (SP15) tts_xling dossier	66	v=2 m=1 c=30

Declared roadmap — contract, acceptance, executor, effort

Ladder to SOTA

Two dated targets, and every rung's role toward them. Best in class is the proven, deployed leader today; the frontier is the research edge as of its month. Each target is dated by a sotabar row and judged against the current month by the bar-age ruler (FRESH, ATTESTED with the fallback month named, or refused). Each rung is substrate (what the target's own methods consume), an arm (a sovereign arm that triangulates without competing for the number), a contender (its done-rule is a number against the target) or superseded (kept as capability, retired as a claim), and carries the rungs it stands on.

best-in-class (T-BIC-TTS), dated by SB-TTS-BIC = 2026-09 UNATTESTED. A codec-token TTS in NishiLang on the integer engine whose WER and speaker similarity on seed-tts-eval sit inside the band the named open-weights models publish, with open-weights voices run as data under their licences ref arxiv_tts_2026_09

frontier (T-FRONT-LONGFORM), dated by SB-TTS-LONGFORM = 2026-09 UNATTESTED. A whole scripture chapter read in one pass with worst-of-N WER under 5 percent over 10 seeds and zero catastrophic generations, by detect-rollback-regenerate ref arxiv_tts_2026_09

frontier (T-FRONT-CLONE), dated by SB-CLONE = 2026-09 UNATTESTED. A consented family voice reading any chapter from one short enrolment clip, every clip carrying an extractable provenance mark naming its grant, revocable by one row ref arxiv_clone_2026_09

Rung	Closes with	Definition of done (pre-declared)	Executor	Est.
Board, bar and census (SP0)	sp_board	LANDED 2026-09-21: speech.matrix, speech.plan, speech.refs with nine pinned mirrors, the lane census, the September bar and three dated targets; regen and page_verify GREEN	Seat	0.5 u
One WAV writer, five retired reversibly (SP1) after SP0	nx_wav_write_pcm16	nx_wav.nx carries the one PCM16 writer; the other five writers are retired by nx_retire_path or repointed as importers; nx_absent over buildroot/runtime proves no second RIFF header emitter; the gate round-trips a written file through a reader and refuses a planted bad header	Organ	0.5 u
Lexicon G2P with scripture names as data (SP2) after SP1	g2p_lex_lookup	A pronunciation lexicon as DATA (knowledge/speech/en.lex and names_scripture.lex with licence rows beside them) consulted before the rules; phoneme error rate on a held-out slice printed with counts; the gate proves a planted override wins over the rule and an unknown word falls to the rule	Organ	1 u
Text normalisation for reading aloud (SP3) after SP2	tn_normalize	Verse references, chapter numbers, ordinals, abbreviations and scripture names expand to spoken words from data rows; idempotent on normalised text; the gate proves each class on planted lines and a neg-control that a bare number is never read as a verse	Organ	1 u
The floor: a whole chapter through the formant reader (SP4) after SP3,SP17	vr_read_chapter	1 Nephi 1 from the public-domain text rendered to one WAV by nx_voice_read through SP2 and SP3, banked by hash and journaled with WER from the outside ASR oracle; ships as a selectable voice named formant; every neural rung must beat it on the spine	Organ	1 u
Speech lab spine (SP11) after SP4	sl_run	One row per synthesis run (organ, voice, text id, seconds, WER against the transcript by the outside ASR oracle, similarity to the reference by voiceprint distance now and a neural embedding later, MOS from UTMOSv2 run outside) appended to knowledge/status/speech/spine.jrnl; the emitter renders the last N rows; the gate proves append-only and refuses a row with no oracle field	Organ	1 u
Neural audio codec twin (SP5) after SP11	nc_encode	A residual-vector-quantised codec at 12.5 Hz frames in integer NishiLang (encoder, quantiser, decoder), trained on LibriTTS-R and the LibriVox scripture reading, round-trip judged on the spine; the gate proves determinism and refuses a planted codebook corruption	Organ	4 u
Codec-token TTS decoder on the integer engine (SP6) after SP5	tts_generate	An autoregressive model over SP5 tokens conditioned on SP2 phonemes and a speaker prompt, served by the nofloat engine class; PRE-DECLARED accept rule: WER and SIM on seed-tts-eval printed beside the rung, and it is not wired into the product until it beats the SP4 floor on both	Organ	6 u
Open-weights voices as data under their licences (SP7) after SP6	vw_import	Weights whose licence permits it (Kokoro-82M Apache-2.0, Kyutai TTS CC-BY 4.0, Chatterbox MIT, as read from the mirrored pages) convert to the estate tensor format with a LICENSE file beside them and run on OUR runtime; no third-party binary or interpreter in the path; the gate proves a licence-less weight file is refused	Organ	3 u
Family voice enrolment with consent and revocation (SP8) after SP7	ve_enrol	A consented upload (an ACCEPT row in knowledge/store/accept- naming speaker, uploader and scope) yields a voiceprint stored in the family's own space, processed locally, revocable by a superseding row that stops every synthesis; the gate proves no enrolment without the row and that revocation refuses	Organ	1.5 u
Provenance mark on every clip (SP9) after SP8	vp_mark	Every synthesised clip carries an imperceptible mark naming the grant id and the engine build, extractable by a sovereign reader after PCM16 and MP3 round trips; the gate proves extraction survives a planted re-encode and an unmarked clip reads unmarked	Organ	1.5 u
Licensed voice ledger (SP10) after SP8	vl_grant	Alpha and beta licensed voices are rows in the licence ledger (nx_acquire_license lineage) with term, scope and revocation; synthesis in a voice with no live grant is refused with the rule named; the gate proves the refusal and a positive control	Organ	1 u
Long-form reliability (SP12) after SP6	tts_longform	Detect an error onset during generation, roll back and regenerate under a temporary constraint; worst-of-N WER over 10 seeds on a 1500-word chapter printed beside the rung; accept below 5 percent and zero catastrophic generations	Organ	2 u
Streaming synthesis with word timestamps (SP13) after SP6	tts_stream_words	Audio frames stream as generated with a word-start table the reader page consumes for highlight; first-frame latency printed on the spine	Organ	1.5 u
Style by natural language (SP14) after SP6	tts_style_nl	Emotion and duration control from a text instruction as a post-training stage, judged by the MOS oracle beside the neutral read	Organ	2 u
Cross-lingual transcript-free cloning (SP15) after SP12	tts_xling	A voice enrolled in English reads Spanish scripture with no transcript of the prompt; speaker similarity printed against the English read	Organ	3 u
Register the dark VOICE-CLONE-001 lane on the MCP surface (SP17) after SP0	sp_lane_registered	OPERATOR 2026-09-21: every capability an MCP verb. nx_voice_read, nx_voice_clone_synth, nx_voiceprint, nx_pitch and nx_phoneme_synth get program wrappers, promote, /api/tools/register, nx_toolreg_reconcile and a minted cap so the floor is callable from any seat today, not only importable; their five gate sources run on the roster; nx_catalog reads REGISTERED and AUTHORISED for each	Organ	1 u
Every rung an MCP verb, one workflow plane, an agent that runs it (SP16) after SP4	sp_mcp_surface	Every organ this board ships is registered, reconciled and capped; a seeded plan_run plane speech-read runs normalise, g2p, synthesise, mark and bench in ONE call and journals each step, so an AI seat, a human or the beat can drive it as an agent; the gate proves every rung organ resolves by nx_catalog to REGISTERED and AUTHORISED and that the plane runs end to end on a planted sentence	Organ	1 u

Milestones

Milestone	Rungs	Cumulative
M0 · Board and bar	SP0	0.5 u
M1 · The floor: lane registered, one writer, lexicon, normaliser, a chapter read, the lab, the workflow plane	SP17,SP1,SP2,SP3,SP4,SP11,SP16	7 u
M2 · Neural: codec, decoder, open weights as data	SP5,SP6,SP7	20 u
M3 · Voices: enrolment, provenance, licence ledger	SP8,SP9,SP10	24 u
M4 · Frontier: long-form, streaming, style, cross-lingual	SP12,SP13,SP14,SP15	32.5 u

Rung	Role	Toward	Stands on	Why
SP17	substrate	T-BIC-TTS	SP0	a floor nobody can call is dark; registration is the first byte of adoption
SP16	substrate	T-BIC-TTS	SP4	the operator's order: every rung as an MCP verb, an API, a workflow plane and an agent
SP0	substrate	T-BIC-TTS	-	the board is what the ranker and every later seat read
SP1	substrate	T-BIC-TTS	SP0	one writer under every clip; six today is the duplicate-ruler defect wearing a file format
SP2	substrate	T-BIC-TTS	SP1	the phoneme stream every decoder conditions on; names are data
SP3	substrate	T-FRONT-LONGFORM	SP2	a chapter is references and numbers before it is words
SP4	arm	T-BIC-TTS	SP3,SP17	the formant floor triangulates the bench and never competes for the number
SP11	substrate	T-BIC-TTS	SP4	no rung is judged without a spine row
SP5	substrate	T-BIC-TTS	SP11	the token space the decoder speaks
SP6	contender	T-BIC-TTS	SP5	its done-rule is a number on seed-tts-eval
SP7	arm	T-BIC-TTS	SP6	a published model run on our runtime is a second voice and a calibration of the decoder
SP8	substrate	T-FRONT-CLONE	SP7	no family voice without a consent row
SP9	substrate	T-FRONT-CLONE	SP8	provenance is the field's own counterweight to cloning
SP10	substrate	T-FRONT-CLONE	SP8	a licensed voice is a grant row, never a file
SP12	contender	T-FRONT-LONGFORM	SP6	its done-rule is worst-of-N WER on a chapter
SP13	substrate	T-BIC-TTS	SP6	the reader page's highlight and the player's first byte
SP14	arm	T-BIC-TTS	SP6	style control is judged beside the neutral read, not instead of it
SP15	contender	T-FRONT-CLONE	SP12	its done-rule is similarity across a language

Ladder verdict. LADDERED. Targets, rung roles and bars are counted in the tables below, each partition re-derived beside its total. Bars read STALE against the current month 2026-10.

Ladder verdict, counted (17 rows, more than the 12-row disclosure bar)

Ladder verdict, counted
Class	Count	Note
targets	3	sotatarget rows on the plan
best in class	1	targets whose class is best-in-class, by code
frontier	2	targets whose class is frontier, by code
undated	0	targets naming no bar that exists
bad class	0	targets whose class is neither word
duplicate	0	target ids an earlier row already carries
rungs	18	rung rows on the plan
placed	18	rungs carrying a rungrole row
unplaced	0	rungs with no role toward any target
bad role	0	rungrole rows naming a role the ladder does not know
bad target	0	rungrole rows naming a target no sotatarget row declares
rungs, summed	18	RECONCILED: the parts sum to the total 18
substrate	12	what the target's own methods consume
arm	3	a sovereign arm that triangulates without competing for the number
contender	3	its done-rule is a number against the target
superseded	0	kept as capability, retired as a claim
placed, summed	18	RECONCILED: the parts sum to the total 18

Bar freshness, counted
Class	Count	Note
bars	4	sotabar rows on the plan
fresh	0	the bar month is the current month
attested	0	an older bar with a this-month barscan reporting no newer entry
stale	0	a this-month scan saw a newer month the bar ignores
unattested	4	nobody looked this month
malformed	0	a sotabar row missing a field
duplicate	0	bar ids an earlier row already carries
bars, summed	4	RECONCILED: the parts sum to the total 4
current month	2026-10	the month every bar is judged against

Treatment verdict. NOT DECLARED. No rung on this board declares its three-step treatment (a treat row naming the MCP row, the API route, the primitive it composes, the workflow that composes it and the agent beat that runs it); the ranker labels every rung TREATMENT-UNDECLARED until it does.

Inspect a rung and its prerequisites

Declared nodes 18. Rank input binding: PLAN_MATRIX_BOUND_ONLY. Dependency order is authored. Implementation, acceptance evidence, authority and resource readiness are unverified. No action is recommended or dispatched here.

Plan SHA-256 b6e641131002a2f0288ba2756707a6d1c1e78b1b2a46274d5ceb873873a21719. Target rows 3; role rows 18. Existing risks and release worklog retain their own scope; no node completion is inferred.

Runtime observation coverage: 0 of 18 rungs have current, identity-bound execution observations. Observation may report RED. Accepted completion remains unknown; source presence, qualification and deployment do not establish it.

Use Enter or Space on a rung to inspect its contract. Prerequisite links locate another rung in this list; open its summary to inspect it. Estimates are authored effort, not forecasts.

SP0 — Board, bar and census
Prerequisites: None declared; this does not establish execution eligibility.
Contract: sp_board
Acceptance: LANDED 2026-09-21: speech.matrix, speech.plan, speech.refs with nine pinned mirrors, the lane census, the September bar and three dated targets; regen and page_verify GREEN
Authored effort: 0.5. Executor kind: Seat. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.
SP1 — One WAV writer, five retired reversibly
Prerequisites: SP0 (acceptance unverified)
Contract: nx_wav_write_pcm16
Acceptance: nx_wav.nx carries the one PCM16 writer; the other five writers are retired by nx_retire_path or repointed as importers; nx_absent over buildroot/runtime proves no second RIFF header emitter; the gate round-trips a written file through a reader and refuses a planted bad header
Authored effort: 0.5. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.
SP2 — Lexicon G2P with scripture names as data
Prerequisites: SP1 (acceptance unverified)
Contract: g2p_lex_lookup
Acceptance: A pronunciation lexicon as DATA (knowledge/speech/en.lex and names_scripture.lex with licence rows beside them) consulted before the rules; phoneme error rate on a held-out slice printed with counts; the gate proves a planted override wins over the rule and an unknown word falls to the rule
Authored effort: 1. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.
SP3 — Text normalisation for reading aloud
Prerequisites: SP2 (acceptance unverified)
Contract: tn_normalize
Acceptance: Verse references, chapter numbers, ordinals, abbreviations and scripture names expand to spoken words from data rows; idempotent on normalised text; the gate proves each class on planted lines and a neg-control that a bare number is never read as a verse
Authored effort: 1. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.
SP4 — The floor: a whole chapter through the formant reader
Prerequisites: SP3 (acceptance unverified), SP17 (acceptance unverified)
Contract: vr_read_chapter
Acceptance: 1 Nephi 1 from the public-domain text rendered to one WAV by nx_voice_read through SP2 and SP3, banked by hash and journaled with WER from the outside ASR oracle; ships as a selectable voice named formant; every neural rung must beat it on the spine
Authored effort: 1. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.
SP11 — Speech lab spine
Prerequisites: SP4 (acceptance unverified)
Contract: sl_run
Acceptance: One row per synthesis run (organ, voice, text id, seconds, WER against the transcript by the outside ASR oracle, similarity to the reference by voiceprint distance now and a neural embedding later, MOS from UTMOSv2 run outside) appended to knowledge/status/speech/spine.jrnl; the emitter renders the last N rows; the gate proves append-only and refuses a row with no oracle field
Authored effort: 1. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.
SP5 — Neural audio codec twin
Prerequisites: SP11 (acceptance unverified)
Contract: nc_encode
Acceptance: A residual-vector-quantised codec at 12.5 Hz frames in integer NishiLang (encoder, quantiser, decoder), trained on LibriTTS-R and the LibriVox scripture reading, round-trip judged on the spine; the gate proves determinism and refuses a planted codebook corruption
Authored effort: 4. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.
SP6 — Codec-token TTS decoder on the integer engine
Prerequisites: SP5 (acceptance unverified)
Contract: tts_generate
Acceptance: An autoregressive model over SP5 tokens conditioned on SP2 phonemes and a speaker prompt, served by the nofloat engine class; PRE-DECLARED accept rule: WER and SIM on seed-tts-eval printed beside the rung, and it is not wired into the product until it beats the SP4 floor on both
Authored effort: 6. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.
SP7 — Open-weights voices as data under their licences
Prerequisites: SP6 (acceptance unverified)
Contract: vw_import
Acceptance: Weights whose licence permits it (Kokoro-82M Apache-2.0, Kyutai TTS CC-BY 4.0, Chatterbox MIT, as read from the mirrored pages) convert to the estate tensor format with a LICENSE file beside them and run on OUR runtime; no third-party binary or interpreter in the path; the gate proves a licence-less weight file is refused
Authored effort: 3. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.
SP8 — Family voice enrolment with consent and revocation
Prerequisites: SP7 (acceptance unverified)
Contract: ve_enrol
Acceptance: A consented upload (an ACCEPT row in knowledge/store/accept- naming speaker, uploader and scope) yields a voiceprint stored in the family's own space, processed locally, revocable by a superseding row that stops every synthesis; the gate proves no enrolment without the row and that revocation refuses
Authored effort: 1.5. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.
SP9 — Provenance mark on every clip
Prerequisites: SP8 (acceptance unverified)
Contract: vp_mark
Acceptance: Every synthesised clip carries an imperceptible mark naming the grant id and the engine build, extractable by a sovereign reader after PCM16 and MP3 round trips; the gate proves extraction survives a planted re-encode and an unmarked clip reads unmarked
Authored effort: 1.5. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.
SP10 — Licensed voice ledger
Prerequisites: SP8 (acceptance unverified)
Contract: vl_grant
Acceptance: Alpha and beta licensed voices are rows in the licence ledger (nx_acquire_license lineage) with term, scope and revocation; synthesis in a voice with no live grant is refused with the rule named; the gate proves the refusal and a positive control
Authored effort: 1. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.
SP12 — Long-form reliability
Prerequisites: SP6 (acceptance unverified)
Contract: tts_longform
Acceptance: Detect an error onset during generation, roll back and regenerate under a temporary constraint; worst-of-N WER over 10 seeds on a 1500-word chapter printed beside the rung; accept below 5 percent and zero catastrophic generations
Authored effort: 2. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.
SP13 — Streaming synthesis with word timestamps
Prerequisites: SP6 (acceptance unverified)
Contract: tts_stream_words
Acceptance: Audio frames stream as generated with a word-start table the reader page consumes for highlight; first-frame latency printed on the spine
Authored effort: 1.5. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.
SP14 — Style by natural language
Prerequisites: SP6 (acceptance unverified)
Contract: tts_style_nl
Acceptance: Emotion and duration control from a text instruction as a post-training stage, judged by the MOS oracle beside the neutral read
Authored effort: 2. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.
SP15 — Cross-lingual transcript-free cloning
Prerequisites: SP12 (acceptance unverified)
Contract: tts_xling
Acceptance: A voice enrolled in English reads Spanish scripture with no transcript of the prompt; speaker similarity printed against the English read
Authored effort: 3. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.
SP17 — Register the dark VOICE-CLONE-001 lane on the MCP surface
Prerequisites: SP0 (acceptance unverified)
Contract: sp_lane_registered
Acceptance: OPERATOR 2026-09-21: every capability an MCP verb. nx_voice_read, nx_voice_clone_synth, nx_voiceprint, nx_pitch and nx_phoneme_synth get program wrappers, promote, /api/tools/register, nx_toolreg_reconcile and a minted cap so the floor is callable from any seat today, not only importable; their five gate sources run on the roster; nx_catalog reads REGISTERED and AUTHORISED for each
Authored effort: 1. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.
SP16 — Every rung an MCP verb, one workflow plane, an agent that runs it
Prerequisites: SP4 (acceptance unverified)
Contract: sp_mcp_surface
Acceptance: Every organ this board ships is registered, reconciled and capped; a seeded plan_run plane speech-read runs normalise, g2p, synthesise, mark and bench in ONE call and journals each step, so an AI seat, a human or the beat can drive it as an agent; the gate proves every rung organ resolves by nx_catalog to REGISTERED and AUTHORISED and that the plane runs end to end on a planted sentence
Authored effort: 1. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.

Learning and practice paths

No structured learning path is declared for this plan. Existing research, roadmap and worklog remain available above.

Rung dossiers — every record that bears on a rung, in one place

Joined by symbol, never typed. Each dossier is assembled on every publish from the rows this board already carries: the plan's rung row, the matrix row that carries its contract symbol (its status re-measured by the one symbol ruler), the gate the organ declares and the verdict the evidence referee recorded on its last run, the reference marks in the rung text, the design rows that name the rung and the work log filed against it. The capability matrix links each cell to its dossier and each dossier links back to its cell and forward to the rungs it stands on; a dependency naming no rung on this plan is written as UNRESOLVED rather than linked. Absences are named: NO-MATRIX-ROW, NO-GATE-DECLARED, UNRECORDED.

Gate verdicts: UNSTAMPED — no per-gate verdict artifact is published for this domain yet (nx_swcompare_evidence speech writes it beside its stamp); every gate below reads UNRECORDED until it runs.

Gate declarations: knowledge/organ_gate.conf read from this working directory (1 rows) and from the estate root (196 rows); rows in both counted once: 0; declarations read: 197. Every declaration in either copy counts, so a stub beside the publisher can never hide the root.

SP0 Board, bar and census · cell NO-MATRIX-ROW · gate NO-MATRIX-ROW

Contract	`sp_board` · executor Seat · 0.5 u · stands on nothing
Definition of done	LANDED 2026-09-21: speech.matrix, speech.plan, speech.refs with nine pinned mirrors, the lane census, the September bar and three dated targets; regen and page_verify GREEN
Matrix cell	NO-MATRIX-ROW — no matrix row carries `sp_board`; the ranker reads this rung as unmapped and no cell can flip for it. Add the row and the join completes on the next publish.
Gate	the gate join needs a matrix row: no organ is named for this rung
Work log	2026-09-21 measure nx_catalog 2026-09-21: nx_tts nx_vocoder nx_speech nx_audio ABSENT; nx_g2p LIB 8150 B and nx_wav LIB 5889 B unbuilt; nx_synthroom PROMOTED 35473 B; nx_shelltool glob wav.nx over buildroot/runtime = 9 files of 25822 (corpus_complete=1), six of them writers 2026-09-21 measure nx_capsearch text to speech synthesis voice narration: nx_g2p first, then the VOICE-CLONE-001 libs (nx_voice_clone_synth, nx_voice_read, nx_voiceprint, nx_voice_codec_v2); no registered TTS tool exists 2026-09-21 land speech.matrix, speech.plan, speech.refs admitted to regen.list; nine mirrors pinned by sha256 from nx_research_fetch provenance lines 2026-09-21 lesson A lane with five gated rungs sat in the tree for weeks with no board and no registration, invisible to every seat that asked the catalogue by the obvious names; a capability with no board is dark by construction, which is the operator's 2026-09-06 order arriving from the audio side 2026-09-21 retract This board's first pos row said nx_tts was ABSENT by catalogue. Within the hour the ranker's incumbent index named nx_tts_speak, nx_tts_seq, nx_tts_word, nx_tts_formant, nx_voice_eval, nx_rights_ledger, nx_license_check and nx_lipsync, and nx_catalog confirmed all eight in source (SOURCE-ONLY or LIB, none built). A name-keyed catalogue query answers about the NAME asked, never the capability; the incumbent index over 21,294 organs is the instrument that answers the capability question. The claim is withdrawn; the pos row and four matrix rows now carry the eight, and SP4, SP10, SP11 and SP17 compose them. 2026-09-21 lesson Two refs rows raised the fleet's open-redirect ratchet because both cited URLs redirect (the arena page to /models, the YouVersion page to its canonical); the canonical pages were re-mirrored and repinned the same hour. Cite the canonical URL the fetch lands on, never the one typed. 2026-09-21 queue treat rows (mcp= api= prim= wf= agent=) land per rung as each verb registers, starting with SP17; declaring them before the verbs exist would refuse the ranker on a board with no callable organ yet, the permanently-red detector; the opt-in is deliberate and dated here. 2026-09-21 measure RANKER CHECK 2026-09-21 (operator: see if the ranker agrees). Seat order: SP17 then SP1, SP2, SP3, SP4, SP11. Ranker order before the fix: SP4, SP1, SP11, SP6, SP3, SP2, SP17. Evidence per pair: SP4 before SP3 and SP2 violates SP4's own dep chain; SP6 before SP5 violates SP6 deps SP5 and prices a 6 u rung at 12 tenths against a 4 u rung at 40 because five contracts share nx_tts.nx; SP17 at eighth is the ranker's honest reading of a rung whose only dependent is SP4 through a missing edge, so SP4's deps row now names SP17 as well. The ranker was improved (pmdash PR13) rather than overridden; after the fix it re-decides 2026-09-21 measure Ranker re-run on the shipped binary with SP4's missing edge declared (deps SP3,SP17): SP17 rose from twelfth to fifth, tied with SP3 at priority 7000, directly above SP4 that needs it; order now SP0 SP1 SP2 SP3 SP17 SP4 SP11 SP5 SP6 SP7 SP8 SP12; zero open rungs above a direct open prerequisite. The seat and the ranker agree, and the agreement came from an edge on the board, not a hand override 2026-09-21 measure nx_depbench speech (sovereign dependency-order ruler, shipped ef3a5048): rungs=18 ranked=18 edges_checked=18 violations=0 verdict=GREEN on the ranker shipped at 71b07255

SP1 One WAV writer, five retired reversibly · cell WATCHING · gate NO-GATE-DECLARED

Contract	`nx_wav_write_pcm16` · executor Organ · 0.5 u · stands on SP0
Definition of done	nx_wav.nx carries the one PCM16 writer; the other five writers are retired by nx_retire_path or repointed as importers; nx_absent over buildroot/runtime proves no second RIFF header emitter; the gate round-trips a written file through a reader and refuses a planted bad header
Matrix cell	One WAV writer, five retired reversibly · `runtime/nx_wav.nx` · WATCHING decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_wav`; declare `nx_wav TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	no log row names this rung yet

SP2 Lexicon G2P with scripture names as data · cell WATCHING · gate NO-GATE-DECLARED

Contract	`g2p_lex_lookup` · executor Organ · 1 u · stands on SP1
Definition of done	A pronunciation lexicon as DATA (knowledge/speech/en.lex and names_scripture.lex with licence rows beside them) consulted before the rules; phoneme error rate on a held-out slice printed with counts; the gate proves a planted override wins over the rule and an unknown word falls to the rule
Matrix cell	Lexicon G2P with scripture names as data · `runtime/nx_g2p.nx` · WATCHING decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_g2p`; declare `nx_g2p TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	no log row names this rung yet

SP3 Text normalisation for reading aloud · cell WATCHING · gate NO-GATE-DECLARED

Contract	`tn_normalize` · executor Organ · 1 u · stands on SP2
Definition of done	Verse references, chapter numbers, ordinals, abbreviations and scripture names expand to spoken words from data rows; idempotent on normalised text; the gate proves each class on planted lines and a neg-control that a bare number is never read as a verse
Matrix cell	Text normalisation for reading aloud · `runtime/nx_textnorm.nx` · WATCHING decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_textnorm`; declare `nx_textnorm TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	no log row names this rung yet

SP4 The floor: a whole chapter through the formant reader · cell WATCHING · gate NO-GATE-DECLARED

Contract	`vr_read_chapter` · executor Organ · 1 u · stands on SP3, SP17
Definition of done	1 Nephi 1 from the public-domain text rendered to one WAV by nx_voice_read through SP2 and SP3, banked by hash and journaled with WER from the outside ASR oracle; ships as a selectable voice named formant; every neural rung must beat it on the spine
Matrix cell	The floor: a whole chapter through the formant reader · `runtime/nx_voice_read.nx` · WATCHING decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_voice_read`; declare `nx_voice_read TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	no log row names this rung yet

SP11 Speech lab spine · cell WATCHING · gate NO-GATE-DECLARED

Contract	`sl_run` · executor Organ · 1 u · stands on SP4
Definition of done	One row per synthesis run (organ, voice, text id, seconds, WER against the transcript by the outside ASR oracle, similarity to the reference by voiceprint distance now and a neural embedding later, MOS from UTMOSv2 run outside) appended to knowledge/status/speech/spine.jrnl; the emitter renders the last N rows; the gate proves append-only and refuses a row with no oracle field
Matrix cell	Speech lab spine (WER, similarity, MOS per run) · `runtime/nx_speechlab.nx` · WATCHING decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_speechlab`; declare `nx_speechlab TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	no log row names this rung yet

SP5 Neural audio codec twin · cell WATCHING · gate NO-GATE-DECLARED

Contract	`nc_encode` · executor Organ · 4 u · stands on SP11
Definition of done	A residual-vector-quantised codec at 12.5 Hz frames in integer NishiLang (encoder, quantiser, decoder), trained on LibriTTS-R and the LibriVox scripture reading, round-trip judged on the spine; the gate proves determinism and refuses a planted codebook corruption
Matrix cell	Neural audio codec twin · `runtime/nx_neural_codec.nx` · WATCHING decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_neural_codec`; declare `nx_neural_codec TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	no log row names this rung yet

SP6 Codec-token TTS decoder on the integer engine · cell WATCHING · gate NO-GATE-DECLARED

Contract	`tts_generate` · executor Organ · 6 u · stands on SP5
Definition of done	An autoregressive model over SP5 tokens conditioned on SP2 phonemes and a speaker prompt, served by the nofloat engine class; PRE-DECLARED accept rule: WER and SIM on seed-tts-eval printed beside the rung, and it is not wired into the product until it beats the SP4 floor on both
Matrix cell	Codec-token TTS decoder on the integer engine · `runtime/nx_tts.nx` · WATCHING decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_tts`; declare `nx_tts TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	no log row names this rung yet

SP7 Open-weights voices as data under their licences · cell WATCHING · gate NO-GATE-DECLARED

Contract	`vw_import` · executor Organ · 3 u · stands on SP6
Definition of done	Weights whose licence permits it (Kokoro-82M Apache-2.0, Kyutai TTS CC-BY 4.0, Chatterbox MIT, as read from the mirrored pages) convert to the estate tensor format with a LICENSE file beside them and run on OUR runtime; no third-party binary or interpreter in the path; the gate proves a licence-less weight file is refused
Matrix cell	Open-weights voices as data read by our engine · `runtime/nx_voice_weights.nx` · WATCHING decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_voice_weights`; declare `nx_voice_weights TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	no log row names this rung yet

SP8 Family voice enrolment with consent and revocation · cell WATCHING · gate NO-GATE-DECLARED

Contract	`ve_enrol` · executor Organ · 1.5 u · stands on SP7
Definition of done	A consented upload (an ACCEPT row in knowledge/store/accept- naming speaker, uploader and scope) yields a voiceprint stored in the family's own space, processed locally, revocable by a superseding row that stops every synthesis; the gate proves no enrolment without the row and that revocation refuses
Matrix cell	Family voice enrolment with consent and revocation · `runtime/nx_voice_enrol.nx` · WATCHING decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_voice_enrol`; declare `nx_voice_enrol TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	no log row names this rung yet

SP9 Provenance mark on every clip · cell WATCHING · gate NO-GATE-DECLARED

Contract	`vp_mark` · executor Organ · 1.5 u · stands on SP8
Definition of done	Every synthesised clip carries an imperceptible mark naming the grant id and the engine build, extractable by a sovereign reader after PCM16 and MP3 round trips; the gate proves extraction survives a planted re-encode and an unmarked clip reads unmarked
Matrix cell	Provenance mark and attribution on every clip · `runtime/nx_voice_mark.nx` · WATCHING decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_voice_mark`; declare `nx_voice_mark TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	no log row names this rung yet

SP10 Licensed voice ledger · cell WATCHING · gate NO-GATE-DECLARED

Contract	`vl_grant` · executor Organ · 1 u · stands on SP8
Definition of done	Alpha and beta licensed voices are rows in the licence ledger (nx_acquire_license lineage) with term, scope and revocation; synthesis in a voice with no live grant is refused with the rule named; the gate proves the refusal and a positive control
Matrix cell	Licensed voice ledger (no live grant, no synthesis) · `runtime/nx_voice_license.nx` · WATCHING decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_voice_license`; declare `nx_voice_license TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	no log row names this rung yet

SP12 Long-form reliability · cell WATCHING · gate NO-GATE-DECLARED

Contract	`tts_longform` · executor Organ · 2 u · stands on SP6
Definition of done	Detect an error onset during generation, roll back and regenerate under a temporary constraint; worst-of-N WER over 10 seeds on a 1500-word chapter printed beside the rung; accept below 5 percent and zero catastrophic generations
Matrix cell	Long-form reliability: detect, roll back, regenerate · `runtime/nx_tts.nx` · WATCHING decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_tts`; declare `nx_tts TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	no log row names this rung yet

SP13 Streaming synthesis with word timestamps · cell WATCHING · gate NO-GATE-DECLARED

Contract	`tts_stream_words` · executor Organ · 1.5 u · stands on SP6
Definition of done	Audio frames stream as generated with a word-start table the reader page consumes for highlight; first-frame latency printed on the spine
Matrix cell	Streaming synthesis with word timestamps · `runtime/nx_tts.nx` · WATCHING decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_tts`; declare `nx_tts TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	no log row names this rung yet

SP14 Style by natural language · cell WATCHING · gate NO-GATE-DECLARED

Contract	`tts_style_nl` · executor Organ · 2 u · stands on SP6
Definition of done	Emotion and duration control from a text instruction as a post-training stage, judged by the MOS oracle beside the neutral read
Matrix cell	Style control by natural language · `runtime/nx_tts.nx` · WATCHING decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_tts`; declare `nx_tts TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	no log row names this rung yet

SP15 Cross-lingual transcript-free cloning · cell WATCHING · gate NO-GATE-DECLARED

Contract	`tts_xling` · executor Organ · 3 u · stands on SP12
Definition of done	A voice enrolled in English reads Spanish scripture with no transcript of the prompt; speaker similarity printed against the English read
Matrix cell	Cross-lingual transcript-free cloning · `runtime/nx_tts.nx` · WATCHING decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_tts`; declare `nx_tts TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	no log row names this rung yet

SP17 Register the dark VOICE-CLONE-001 lane on the MCP surface · cell WATCHING · gate NO-GATE-DECLARED

Contract	`sp_lane_registered` · executor Organ · 1 u · stands on SP0
Definition of done	OPERATOR 2026-09-21: every capability an MCP verb. nx_voice_read, nx_voice_clone_synth, nx_voiceprint, nx_pitch and nx_phoneme_synth get program wrappers, promote, /api/tools/register, nx_toolreg_reconcile and a minted cap so the floor is callable from any seat today, not only importable; their five gate sources run on the roster; nx_catalog reads REGISTERED and AUTHORISED for each
Matrix cell	The dark lane registered on the MCP surface · `runtime/nx_voice_read.nx` · WATCHING decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_voice_read`; declare `nx_voice_read TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	no log row names this rung yet

SP16 Every rung an MCP verb, one workflow plane, an agent that runs it · cell WATCHING · gate NO-GATE-DECLARED

Contract	`sp_mcp_surface` · executor Organ · 1 u · stands on SP4
Definition of done	Every organ this board ships is registered, reconciled and capped; a seeded plan_run plane speech-read runs normalise, g2p, synthesise, mark and bench in ONE call and journals each step, so an AI seat, a human or the beat can drive it as an agent; the gate proves every rung organ resolves by nx_catalog to REGISTERED and AUTHORISED and that the plane runs end to end on a planted sentence
Matrix cell	Every rung an MCP verb, one workflow plane, an agent · `runtime/nx_speechlab.nx` · WATCHING decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_speechlab`; declare `nx_speechlab TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	no log row names this rung yet

dossiers=18 cells_joined=17 no_matrix_row=1 (sum 18) gates_recorded=0 gates_unrecorded=0 no_gate_declared=17 (sum 17 of cells_joined) evgates=UNSTAMPED

Capability comparisons

Compare the field, search individual capabilities and open their source and adoption evidence. Documented presence does not establish comparative quality.

Position map — centrality and distinctiveness

The four-quadrant map the field uses for brand strategy (Dawar and Bagga, HBR June 2015), re-derived from this matrix on every publish. Centrality is the share of the category's feature mass a player covers, each feature weighted by how many hold it; distinctiveness is the average lead over each rival on the rows the player holds; breadth is the depth-weighted share of the whole matrix (the bubble); depth is how deeply the rows held are held; momentum is the day-over-day move off the spine (green rising, red falling, grey until day two); the dashed path runs first day → previous day → today. Dividers are the category means. Axes are fitted to the field of play, so read the tick numerals, not the frame. Rival marks are documented presence, so a rival's position reads the record, never its quality. The picture grades its own readability below; the table beside it is the same data for a screen reader or a second method.

Rows. 24 feature rows mapped of 35 matrix rows; 11 substrate rows (kernels, bounds and instruments the plan's rungrole marks substrate) are the estate's scaffolding, not category features, and stay off the map

Views. 2D cut: centrality, distinctiveness · 3D cube: centrality, distinctiveness, breadth · axes registered: centrality, distinctiveness, breadth, depth, momentum · a board picks its own in knowledge/compare/speech.cdmap (cut|x|y, cube|x|y|z); absent = the HBR defaults

bubble area = breadth · ring = momentum (green rising, red falling, grey until day two) · dashed = category means · axes fitted to the field of play: centrality 400–1000, distinctiveness 150–400 of 0–1000 permil (the full range put every player in one corner)
readability of the 2D cut, self-graded by the layout ruler: label overlaps 0 · labels over marks 0 · off-canvas 0 · unresolved labels 0 · mark overlaps 0 (a fact of the data: two players that close are that close) · data spread 576 permil of the plot · quadrant words unseated 0
text contrast, measured with wcag2-ratio (floors from contrast.conf), light theme: labels 16.24 (floor 4.50) · notes 16.24 (floor 4.50) · quadrant words 5.89 (floor 4.50) · tick numerals 5.89 (floor 4.50) · axis titles 5.59 (floor 4.50) · dark theme: labels 13.78 (floor 4.50) · quadrant words 5.55 (floor 4.50) · axis titles 5.97 (floor 4.50) · dark classes under their floor 0 (one figure serves both themes: a dark shortfall is a token to fix, never a class to hide) · classes refused under their floor 0 (a refused class is not drawn; the scale classes are measured, never hidden) · export: SVG PNG (receipt, rendered by the estate's own rasteriser from this page)

axes fitted: centrality 400–1000 · distinctiveness 150–400 · breadth 250–650 permil · farther bubbles are painted first, nearer ones over them · the floor shadow is each bubble's (x, y) at height 0
readability of the 3D cube, self-graded by the layout ruler: label overlaps 0 · labels over marks 0 · off-canvas 0 · unresolved labels 0 · mark overlaps 0 (a fact of the data: two players that close are that close) · data spread 373 permil of the plot

Nishi Kokoro-82M Chatterbox Kyutai TTS StepAudio 3 Gen
10 panels over 5 registered axes, every pair once (the lower axis on x, the higher on y) · each panel fitted to its own field of play, first and last tick numerals shown · no labels in a small cut, the legend names the colours; the grade below reports the mark terms only (MM summed over the panels, spread averaged)
readability of the small multiples, self-graded by the layout ruler: label overlaps 0 · labels over marks 0 · off-canvas 0 · unresolved labels 0 · mark overlaps 10 (a fact of the data: two players that close are that close) · data spread 550 permil of the plot

Player	Quadrant	centrality	distinctiveness	breadth	depth	momentum	Rows held	Days on spine	First seen
Nishi	Unconventional	557	375	388	666	0 since 2026-09-23	14	4	2026-09-21
Kokoro-82M	Peripheral	628	212	375	818	0 since 2026-09-23	11	4	2026-09-21
Chatterbox	Mainstream	857	216	500	800	0 since 2026-09-23	15	4	2026-09-21
Kyutai TTS	Mainstream	857	244	527	844	0 since 2026-09-23	15	4	2026-09-21
StepAudio 3 Gen	Aspirational	842	266	527	844	0 since 2026-09-23	15	4	2026-09-21

Day	Matrix rows reviewed	Players recorded
2026-09-21	20	5
2026-09-22	24	5
2026-09-23	24	5
2026-10-08	24	5

How to read this evidence. A Nishi source mark records whether the named symbol is declared in the selected source file. It does not establish working behavior or accepted delivery. A watching tag names an expected source entry; discovering it updates source coverage only. A dark tag means the file exists but the named symbol was not found: inspect the source and contract before assuming the capability is missing or renaming code. Peer codes are authored assessments; unassessed means no comparison is established, not absence. Qualification, deployment identity, runtime observations and acceptance must carry their own evidence. Adoption records the catalog's observed path through source, build, promotion, registration and invocation; libraries use importer evidence and gates use declared execution surfaces. A full catalog path does not establish quality, freshness or a successful user outcome. Partial and unknown states retain their stated next action. Census stamps: importers asof 1790637268, gate census asof 1790104131 (unix seconds; -1 = census absent).

Capability matrix — measured against source

◉ authored stronger claim; source present● present◐ partial○ absent? unassessed; excluded from relative comparison · click any capability for its evidence

Peer cell coverage: authored assessed 140, unassessed 0. Authored assessments are not verified outcomes.

Capability	Nishi	Kokoro-82M	Chatterbox	Kyutai TTS	StepAudio 3 Gen
Text to phoneme IDs (rule G2P over lowercase ASCII) Measured: `nx_g2p` exists in `runtime/nx_g2p.nx`, verified at emit. Present: 27-phoneme rule converter, no lexicon. Kokoro runs the misaki G2P library as a separate stage [kokoro82m]; Chatterbox, Kyutai and StepAudio are tokenizer-driven codec language models with no separate G2P stage, coded 0 as not-a-stage [chatterbox] [kyutai_tts] [arxiv_tts_2026_09] Adoption: `LIB-GATE-ONLY importers=1` — PARTIAL: imported only by validation organs (gates, tests, benches): wire it into a shipping program. not adopted: imported only by validation organs (gates, tests, benches): wire it into a shipping program Pros Implementing source symbol found; runtime acceptance separate; ahead of Chatterbox, Kyutai TTS, StepAudio 3 GenCons not adopted yet: imported only by validation organs (gates, tests, benches): wire it into a shipping program	●	●	○	○	○
Phoneme sequence to time-varying speech (formant articulation) Measured: `nx_phoneme_synth` exists in `runtime/nx_phoneme_synth.nx`, verified at emit. Present at the FLOOR: two-pole resonators per phoneme record, intelligible words. Every rival synthesises at neural quality; ours is the arm that triangulates the bench (SP4) rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them Adoption: `LIB-WIRED importers=2 nonval=1` — fully adopted (top of its ladder). Pros Implementing source symbol found; runtime acceptance separate; fully adopted on the estate ladderCons behind Kokoro-82M (leads), Chatterbox (leads), Kyutai TTS (leads), StepAudio 3 Gen (leads)	●	◉	◉	◉	◉
Sovereign F0 detector (integer YIN) Measured: `nx_pitch_f0` exists in `runtime/nx_pitch.nx`, verified at emit. Present substrate; no rival exposes a pitch tracker as a product surface Adoption: `LIB-WIRED importers=5 nonval=3` — fully adopted (top of its ladder). Pros Implementing source symbol found; runtime acceptance separate; ahead of Kokoro-82M, Chatterbox, Kyutai TTS, StepAudio 3 Gen; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)	●	○	○	○	○
Speaker identity descriptor and distance (PARCOR voiceprint) Measured: `nx_voiceprint_extract` exists in `runtime/nx_voiceprint.nx`, verified at emit. Present: LPC reflection-coefficient print, pitch-invariant by construction. Chatterbox clones from a reference clip [chatterbox], Kyutai from a repository of voices [kyutai_tts], StepAudio by zero-shot prompt and voice design [arxiv_tts_2026_09]; Kokoro ships fixed voices [kokoro82m] Adoption: `LIB-WIRED importers=6 nonval=1` — fully adopted (top of its ladder). Pros Implementing source symbol found; runtime acceptance separate; ahead of Kokoro-82M; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)	●	○	●	●	●
Speech in a captured voice at any pitch (LPC source-filter) Measured: `nx_voice_clone_synth` exists in `runtime/nx_voice_clone_synth.nx`, verified at emit. Present: glottal impulse train driving the voiceprint's filter. Rivals clone at neural quality from seconds of audio rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them Adoption: `LIB-GATE-ONLY importers=1` — PARTIAL: imported only by validation organs (gates, tests, benches): wire it into a shipping program. not adopted: imported only by validation organs (gates, tests, benches): wire it into a shipping program Pros Implementing source symbol found; runtime acceptance separate; ahead of Kokoro-82MCons behind Chatterbox (leads), Kyutai TTS (leads), StepAudio 3 Gen (leads); not adopted yet: imported only by validation organs (gates, tests, benches): wire it into a shipping program	●	○	◉	◉	◉
A cloned voice reading words (vocal-tract warp over articulation) Measured: `nx_voice_read` exists in `runtime/nx_voice_read.nx`, verified at emit. Present: rung 5-FUSE, words in a chosen speaker's tract length and F0. This is the reader the floor rung SP4 drives over a whole chapter rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them Adoption: `LIB-GATE-ONLY importers=2` — PARTIAL: imported only by validation organs (gates, tests, benches): wire it into a shipping program. not adopted: imported only by validation organs (gates, tests, benches): wire it into a shipping program Pros Implementing source symbol found; runtime acceptance separate; ahead of Kokoro-82MCons behind Chatterbox (leads), Kyutai TTS (leads), StepAudio 3 Gen (leads); not adopted yet: imported only by validation organs (gates, tests, benches): wire it into a shipping program	●	○	◉	◉	◉
LPC-residual voice codec (NishiVoice v2) Measured: `v2_encode` exists in `runtime/nx_voice_codec_v2.nx`, verified at emit. Present: a classical codec for the wire. Kyutai's Mimi and StepAudio's 12.5 Hz RVQ tokenizer (16x2048 codes, semantic and acoustic jointly) are neural token codecs the decoder speaks [kyutai_tts] [arxiv_tts_2026_09]; Chatterbox READ 2026-09-21: its README credits the S3Tokenizer speech tokenizer, so it is codec-token driven, coded 1 [chatterbox]; Kokoro is not codec-based Adoption: `LIB-GATE-ONLY importers=1` — PARTIAL: imported only by validation organs (gates, tests, benches): wire it into a shipping program. not adopted: imported only by validation organs (gates, tests, benches): wire it into a shipping program Pros Implementing source symbol found; runtime acceptance separate; ahead of Kokoro-82M, ChatterboxCons behind Kyutai TTS (leads), StepAudio 3 Gen (leads); not adopted yet: imported only by validation organs (gates, tests, benches): wire it into a shipping program	●	○	◐	◉	◉
RIFF/WAVE writer (one of six in the tree) Measured: `nx_wav_write_file` exists in `runtime/nx_wav.nx`, verified at emit. Present six times over (nx_wav, nx_wav_header, nx_wav_writer, wav_header, wav_writer, nx_audio_wav measured by glob 2026-09-21): the duplicate-ruler defect; SP1 consolidates onto this one rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them Adoption: `LIB-UNIMPORTED` — PARTIAL: nothing imports it: wire it into a consumer or retire it. not adopted: nothing imports it: wire it into a consumer or retire it Pros Implementing source symbol found; runtime acceptance separateCons not adopted yet: nothing imports it: wire it into a consumer or retire it	●	●	●	●	●
Part-of-speech tagger (prosody and homograph substrate) Measured: `pt_tag_stream` exists in `runtime/nx_postag.nx`, verified at emit. Present from the intelmine lane (931 on held-out EWT); nothing on this board consumes it yet Adoption: `LIB-WIRED importers=6 nonval=4` — fully adopted (top of its ladder). Pros Implementing source symbol found; runtime acceptance separate; ahead of Kokoro-82M, Chatterbox, Kyutai TTS, StepAudio 3 Gen; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)	●	○	○	○	○
Search core the reader will call Measured: `bks_search` exists in `runtime/_hdl_build/nx_book_search_lib.nx`, verified at emit. Present from the library lane; the scripturestudy board wires it Adoption: `LIB-WIRED importers=1 nonval=1` — fully adopted (top of its ladder). Pros Implementing source symbol found; runtime acceptance separate; ahead of Kokoro-82M, Chatterbox, Kyutai TTS, StepAudio 3 Gen; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)	●	○	○	○	○
Any English text to a WAV (dictionary plus rules G2P into a formant cascade, the older lane) Measured: `g2p` exists in `runtime/nx_tts_speak.nx`, verified at emit. Present SOURCE-ONLY, never built: R2.3 supersedes the fixed-word demos nx_tts_word and nx_tts_seq; float arithmetic and a laptop output path hard-coded, so SP17 wraps it before it can register; the second generator the floor rung SP4 benches rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them Adoption: `SOURCE-ONLY` — PARTIAL: source exists, never compiled: /api/build it. not adopted: source exists, never compiled: /api/build it Pros Implementing source symbol found; runtime acceptance separateCons behind Kokoro-82M (leads), Chatterbox (leads), Kyutai TTS (leads), StepAudio 3 Gen (leads); not adopted yet: source exists, never compiled: /api/build it	●	◉	◉	◉	◉
Reference-free naturalness ruler (F0 mean and spread as proxies) Measured: `main` exists in `runtime/nx_voice_eval.nx`, verified at emit. Present SOURCE-ONLY: the estate's own voice-realism census V0, one main; SP11 extracts it into the lab as the sovereign axis beside the outside MOS oracle Adoption: `SOURCE-ONLY` — PARTIAL: source exists, never compiled: /api/build it. not adopted: source exists, never compiled: /api/build it Pros Implementing source symbol found; runtime acceptance separate; ahead of Kokoro-82M, Chatterbox, Kyutai TTS, StepAudio 3 GenCons not adopted yet: source exists, never compiled: /api/build it	●	○	○	○	○
Rights and consent ledger predicate (grant, revoke, state) Measured: `nx_rights_is_licensed` exists in `runtime/nx_rights_ledger.nx`, verified at emit. Present LIB from the talent-rights lane: default-deny with expired and revoked states; SP8 and SP10 compose it rather than write a second ledger Adoption: `LIB-WIRED importers=10 nonval=3` — fully adopted (top of its ladder). Pros Implementing source symbol found; runtime acceptance separate; ahead of Kokoro-82M, Chatterbox, Kyutai TTS, StepAudio 3 Gen; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)	●	○	○	○	○
Licence-tier wall for imported artefacts Measured: `nx_license_extract` exists in `runtime/nx_license_check.nx`, verified at emit. Present LIB: the tier wall SP7 composes so an unlicensed weight file is refused by the incumbent rule Adoption: `LIB-WIRED importers=2 nonval=1` — fully adopted (top of its ladder). Pros Implementing source symbol found; runtime acceptance separate; ahead of Kokoro-82M, Chatterbox, Kyutai TTS, StepAudio 3 Gen; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)	●	○	○	○	○
One WAV writer, five retired reversibly DARK — `runtime/nx_wav.nx` EXISTS but does not declare `nx_wav_write_pcm16`: something shipped at this path under another name, and this mark would read open forever. Repoint the contract to the real entry point (the plan rung AND this row) or rename the function; the flip follows on the next beat, and the `comparewatch-` plane row reads DARK until then. WATCH SP1. Six writers cannot all be the ruler; retire by nx_retire_path, prove absence by nx_absent rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them dossierwatching nx_wav_write_pcm16 Pros none measured yetCons behind Kokoro-82M, Chatterbox, Kyutai TTS, StepAudio 3 Gen; open contract; expected source declaration not established	○	●	●	●	●
Text normalisation for reading aloud Open — watching `runtime/nx_textnorm.nx` : `tn_normalize`, re-measured on every compare beat. Ship that symbol and this mark flips itself; the `comparewatch-` plane row flips with it. WATCH SP3. Verse references, numbers, ordinals and scripture names as data rows rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them dossierwatching tn_normalize Pros none measured yetCons behind Kokoro-82M, Chatterbox, Kyutai TTS, StepAudio 3 Gen; open contract; expected source declaration not established	○	●	●	●	●
Lexicon G2P with scripture names as data DARK — `runtime/nx_g2p.nx` EXISTS but does not declare `g2p_lex_lookup`: something shipped at this path under another name, and this mark would read open forever. Repoint the contract to the real entry point (the plan rung AND this row) or rename the function; the flip follows on the next beat, and the `comparewatch-` plane row reads DARK until then. WATCH SP2. Kokoro's misaki is a lexicon-plus-rules G2P [kokoro82m]; the codec LMs learn pronunciation from audio; a dictionary-constrained CRF G2P is the month's published shape [arxiv_tts_2026_09] dossierwatching g2p_lex_lookup Pros none measured yetCons behind Kokoro-82M (leads); open contract; expected source declaration not established	○	◉	○	○	○
The floor
a whole chapter through the formant reader DARK — `runtime/nx_voice_read.nx` EXISTS but does not declare `vr_read_chapter`: something shipped at this path under another name, and this mark would read open forever. Repoint the contract to the real entry point (the plan rung AND this row) or rename the function; the flip follows on the next beat, and the `comparewatch-` plane row reads DARK until then. WATCH SP4. The first selectable voice on gospelofwest.com and the number every neural rung must beat rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them dossierwatching vr_read_chapter Pros none measured yetCons behind Kokoro-82M (leads), Chatterbox (leads), Kyutai TTS (leads), StepAudio 3 Gen (leads); open contract; expected source declaration not established	○	◉	◉	◉	◉
Speech lab spine (WER, similarity, MOS per run) Open — watching `runtime/nx_speechlab.nx` : `sl_run`, re-measured on every compare beat. Ship that symbol and this mark flips itself; the `comparewatch-` plane row flips with it. WATCH SP11. One row per run, oracle fields mandatory rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them dossierwatching sl_run Pros none measured yetCons behind Kokoro-82M, Chatterbox, Kyutai TTS, StepAudio 3 Gen; open contract; expected source declaration not established	○	●	●	●	●
Neural audio codec twin Open — watching `runtime/nx_neural_codec.nx` : `nc_encode`, re-measured on every compare beat. Ship that symbol and this mark flips itself; the `comparewatch-` plane row flips with it. WATCH SP5. The RVQ tokenizer at 12.5 Hz frames in integer NishiLang; StepAudio's shared 16x2048 code space is the bar [arxiv_tts_2026_09] dossierwatching nc_encode Pros none measured yetCons behind Chatterbox, Kyutai TTS (leads), StepAudio 3 Gen (leads); open contract; expected source declaration not established	○	○	◐	◉	◉
Codec-token TTS decoder on the integer engine Open — watching `runtime/nx_tts.nx` : `tts_generate`, re-measured on every compare beat. Ship that symbol and this mark flips itself; the `comparewatch-` plane row flips with it. WATCH SP6. The state of the art is an autoregressive LM over codec tokens; Qwen3-TTS and VoxCPM2 hold the short-form WER and SIM bar and StepAudio 3 Gen the unified one [arxiv_tts_2026_09]; accept rule pre-declared on seed-tts-eval [seedtts_eval] dossierwatching tts_generate Pros none measured yetCons behind Kokoro-82M (leads), Chatterbox (leads), Kyutai TTS (leads), StepAudio 3 Gen (leads); open contract; expected source declaration not established	○	◉	◉	◉	◉
Open-weights voices as data read by our engine Open — watching `runtime/nx_voice_weights.nx` : `vw_import`, re-measured on every compare beat. Ship that symbol and this mark flips itself; the `comparewatch-` plane row flips with it. WATCH SP7. Kokoro-82M is apache-2.0 [kokoro82m], Chatterbox MIT [chatterbox], Kyutai TTS weights CC-BY 4.0 [kyutai_tts]: data assets with the licence beside them, executed by our runtime, never their runtime; StepAudio's weights were not read for licence and code 0 dossierwatching vw_import Pros none measured yetCons behind Kokoro-82M (leads), Chatterbox (leads), Kyutai TTS (leads); open contract; expected source declaration not established	○	◉	◉	◉	○
Family voice enrolment with consent and revocation Open — watching `runtime/nx_voice_enrol.nx` : `ve_enrol`, re-measured on every compare beat. Ship that symbol and this mark flips itself; the `comparewatch-` plane row flips with it. WATCH SP8. Rivals clone from any clip; ours refuses without an ACCEPT row and stops on a revocation row rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them dossierwatching ve_enrol Pros none measured yetCons behind Chatterbox, Kyutai TTS, StepAudio 3 Gen; open contract; expected source declaration not established	○	○	●	●	●
Provenance mark and attribution on every clip Open — watching `runtime/nx_voice_mark.nx` : `vp_mark`, re-measured on every compare beat. Ship that symbol and this mark flips itself; the `comparewatch-` plane row flips with it. WATCH SP9. Chatterbox embeds PerTh neural watermarks that survive MP3 compression and ships the extractor [chatterbox]; GenTraceBench and zero-shot attribution are the month's forensic bar [arxiv_clone_2026_09] dossierwatching vp_mark Pros none measured yetCons behind Chatterbox (leads); open contract; expected source declaration not established	○	○	◉	○	○
Licensed voice ledger (no live grant, no synthesis) Open — watching `runtime/nx_voice_license.nx` : `vl_grant`, re-measured on every compare beat. Ship that symbol and this mark flips itself; the `comparewatch-` plane row flips with it. WATCH SP10. Alpha and beta licensed voices are grant rows with term and scope; refusal names the rule dossierwatching vl_grant Pros none measured yetCons open contract; expected source declaration not established	○	○	○	○	○
Long-form reliability: detect, roll back, regenerate Open — watching `runtime/nx_tts.nx` : `tts_longform`, re-measured on every compare beat. Ship that symbol and this mark flips itself; the `comparewatch-` plane row flips with it. WATCH SP12. LACI takes worst-of-N WER on 1500-word prompts from 35.2 to 3.4 percent [arxiv_tts_2026_09]; no open model ships it, coded 0 for all four dossierwatching tts_longform Pros none measured yetCons open contract; expected source declaration not established	○	○	○	○	○
Streaming synthesis with word timestamps Open — watching `runtime/nx_tts.nx` : `tts_stream_words`, re-measured on every compare beat. Ship that symbol and this mark flips itself; the `comparewatch-` plane row flips with it. WATCH SP13. Kyutai TTS is a decoder-only streaming model [kyutai_tts]; word-start tables for a highlight are ours to add dossierwatching tts_stream_words Pros none measured yetCons behind Kyutai TTS (leads); open contract; expected source declaration not established	○	○	○	◉	○
Style control by natural language Open — watching `runtime/nx_tts.nx` : `tts_style_nl`, re-measured on every compare beat. Ship that symbol and this mark flips itself; the `comparewatch-` plane row flips with it. WATCH SP14. Post-training zero-shot control of emotion and duration by instruction is published this month [arxiv_tts_2026_09]; StepAudio's voice design and vibe speech are the bar; Chatterbox exposes an exaggeration control [chatterbox] dossierwatching tts_style_nl Pros none measured yetCons behind Chatterbox, StepAudio 3 Gen (leads); open contract; expected source declaration not established	○	○	●	○	◉
Cross-lingual transcript-free cloning Open — watching `runtime/nx_tts.nx` : `tts_xling`, re-measured on every compare beat. Ship that symbol and this mark flips itself; the `comparewatch-` plane row flips with it. WATCH SP15. Cross-Lingual F5-TTS 2 clones across languages without a prompt transcript or forced alignment [arxiv_clone_2026_09]; Kyutai is English and French [kyutai_tts] dossierwatching tts_xling Pros none measured yetCons behind Kyutai TTS, StepAudio 3 Gen; open contract; expected source declaration not established	○	○	○	●	●
The dark lane registered on the MCP surface DARK — `runtime/nx_voice_read.nx` EXISTS but does not declare `sp_lane_registered`: something shipped at this path under another name, and this mark would read open forever. Repoint the contract to the real entry point (the plan rung AND this row) or rename the function; the flip follows on the next beat, and the `comparewatch-` plane row reads DARK until then. WATCH SP17. Five gated rungs importable but not callable; every rival ships a callable model; ours becomes verbs with caps rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them dossierwatching sp_lane_registered Pros none measured yetCons behind Kokoro-82M, Chatterbox, Kyutai TTS, StepAudio 3 Gen; open contract; expected source declaration not established	○	●	●	●	●
Every rung an MCP verb, one workflow plane, an agent Open — watching `runtime/nx_speechlab.nx` : `sp_mcp_surface`, re-measured on every compare beat. Ship that symbol and this mark flips itself; the `comparewatch-` plane row flips with it. WATCH SP16. speech-read as a seeded plan_run plane: normalise, g2p, synthesise, mark, bench in one call; rivals expose one API call each rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them dossierwatching sp_mcp_surface Pros none measured yetCons behind Kokoro-82M, Chatterbox, Kyutai TTS, StepAudio 3 Gen; open contract; expected source declaration not established	○	●	●	●	●
MOS oracle run outside, sovereign twin owed Open — watching `runtime/nx_speechlab.nx` : `sl_mos_twin`, re-measured on every compare beat. Ship that symbol and this mark flips itself; the `comparewatch-` plane row flips with it. WATCH SP11. UTMOSv2 predicts MOS from audio [utmosv2]; VoiceMOS 2026 is the challenge that ranks such predictors [arxiv_tts_2026_09]; it runs outside and only its number enters the spine until a twin exists watching sl_mos_twin Pros none measured yetCons behind Kokoro-82M, Chatterbox, Kyutai TTS, StepAudio 3 Gen; open contract; expected source declaration not established	○	●	●	●	●
WER and speaker-similarity test sets Open — watching `runtime/nx_speechlab.nx` : `sl_seedtts`, re-measured on every compare beat. Ship that symbol and this mark flips itself; the `comparewatch-` plane row flips with it. WATCH SP11. seed-tts-eval's test sets and scripts are the WER and SIM ruler the decoder's accept rule names [seedtts_eval] watching sl_seedtts Pros none measured yetCons behind Kokoro-82M, Chatterbox, Kyutai TTS, StepAudio 3 Gen; open contract; expected source declaration not established	○	●	●	●	●
Human preference arena as a bar Open — watching `runtime/nx_speechlab.nx` : `sl_arena_bar`, re-measured on every compare beat. Ship that symbol and this mark flips itself; the `comparewatch-` plane row flips with it. WATCH SP11. The Artificial Analysis TTS arena page is mirrored; its leaderboard is rendered by script, so the bytes witness the surface and not the numbers, and a numbers read is owed [aa_tts_arena] watching sl_arena_bar Pros none measured yetCons behind Kokoro-82M, Chatterbox, Kyutai TTS, StepAudio 3 Gen; open contract; expected source declaration not established	○	●	●	●	●
Corpus with licence for training and evaluation Open — watching `runtime/nx_speechlab.nx` : `sl_corpus`, re-measured on every compare beat. Ship that symbol and this mark flips itself; the `comparewatch-` plane row flips with it. WATCH SP5. LibriTTS-R is CC BY 4.0 on OpenSLR 141 [libritts_r]; the LibriVox Book of Mormon reading is public domain and is both training audio and the scripture bench (see scripturestudy) watching sl_corpus Pros none measured yetCons behind Kokoro-82M, Chatterbox, Kyutai TTS, StepAudio 3 Gen; open contract; expected source declaration not established	○	●	●	●	●

Rival-claim provenance. Every Yes, Best or Part mark in a rival column is a claim about someone else's product. A cited row carries a reference mark that resolves to a pinned mirror (the refs gate measures that); an uncited row is an observation read and is badged in its evidence until a reference lands. The badge is the receipt, never the proof: a mark proves a mirror exists, not that the mirror supports the code. The rows are counted below.

Rival claims, counted
Class	Count	Note
rows with rival marks	27	rows carrying a Yes, Best or Part in a rival column
cited	15	a reference mark resolves to a pinned mirror
uncited	12	an observation read, badged until a reference lands
rows, summed	27	RECONCILED: the parts sum to the total 27

Delivery and evidence

Inspect risks, technical debt, rendered observations, experiments and references. Read scope and limitations alongside every result.

Release history and work log

The worked plan, on the board. Every leg appends what it measured, landed, retracted, learned and left queued, per rung, so a crash leaves the next seat a ledger here rather than a transcript to mine. Kinds: measure, land, retract, lesson, queue. Newest last.

When	Rung	Kind	Entry
2026-09-21	SP0	measure	nx_catalog 2026-09-21: nx_tts nx_vocoder nx_speech nx_audio ABSENT; nx_g2p LIB 8150 B and nx_wav LIB 5889 B unbuilt; nx_synthroom PROMOTED 35473 B; nx_shelltool glob wav.nx over buildroot/runtime = 9 files of 25822 (corpus_complete=1), six of them writers
2026-09-21	SP0	measure	nx_capsearch text to speech synthesis voice narration: nx_g2p first, then the VOICE-CLONE-001 libs (nx_voice_clone_synth, nx_voice_read, nx_voiceprint, nx_voice_codec_v2); no registered TTS tool exists
2026-09-21	SP0	land	speech.matrix, speech.plan, speech.refs admitted to regen.list; nine mirrors pinned by sha256 from nx_research_fetch provenance lines
2026-09-21	SP0	lesson	A lane with five gated rungs sat in the tree for weeks with no board and no registration, invisible to every seat that asked the catalogue by the obvious names; a capability with no board is dark by construction, which is the operator's 2026-09-06 order arriving from the audio side
2026-09-21	SP0	retract	This board's first pos row said nx_tts was ABSENT by catalogue. Within the hour the ranker's incumbent index named nx_tts_speak, nx_tts_seq, nx_tts_word, nx_tts_formant, nx_voice_eval, nx_rights_ledger, nx_license_check and nx_lipsync, and nx_catalog confirmed all eight in source (SOURCE-ONLY or LIB, none built). A name-keyed catalogue query answers about the NAME asked, never the capability; the incumbent index over 21,294 organs is the instrument that answers the capability question. The claim is withdrawn; the pos row and four matrix rows now carry the eight, and SP4, SP10, SP11 and SP17 compose them.
2026-09-21	SP0	lesson	Two refs rows raised the fleet's open-redirect ratchet because both cited URLs redirect (the arena page to /models, the YouVersion page to its canonical); the canonical pages were re-mirrored and repinned the same hour. Cite the canonical URL the fetch lands on, never the one typed.
2026-09-21	SP0	queue	treat rows (mcp= api= prim= wf= agent=) land per rung as each verb registers, starting with SP17; declaring them before the verbs exist would refuse the ranker on a board with no callable organ yet, the permanently-red detector; the opt-in is deliberate and dated here.
2026-09-21	SP0	measure	RANKER CHECK 2026-09-21 (operator: see if the ranker agrees). Seat order: SP17 then SP1, SP2, SP3, SP4, SP11. Ranker order before the fix: SP4, SP1, SP11, SP6, SP3, SP2, SP17. Evidence per pair: SP4 before SP3 and SP2 violates SP4's own dep chain; SP6 before SP5 violates SP6 deps SP5 and prices a 6 u rung at 12 tenths against a 4 u rung at 40 because five contracts share nx_tts.nx; SP17 at eighth is the ranker's honest reading of a rung whose only dependent is SP4 through a missing edge, so SP4's deps row now names SP17 as well. The ranker was improved (pmdash PR13) rather than overridden; after the fix it re-decides
2026-09-21	SP0	measure	Ranker re-run on the shipped binary with SP4's missing edge declared (deps SP3,SP17): SP17 rose from twelfth to fifth, tied with SP3 at priority 7000, directly above SP4 that needs it; order now SP0 SP1 SP2 SP3 SP17 SP4 SP11 SP5 SP6 SP7 SP8 SP12; zero open rungs above a direct open prerequisite. The seat and the ranker agree, and the agreement came from an edge on the board, not a hand override
2026-09-21	SP0	measure	nx_depbench speech (sovereign dependency-order ruler, shipped ef3a5048): rungs=18 ranked=18 edges_checked=18 violations=0 verdict=GREEN on the ranker shipped at 71b07255

On these two registers. Rows are declared in the domain's plan file and carry the debt id, which is the join key back to the sovereign debt plane — that plane, not this page, is the authority on state. Reconciling them automatically (the regen reading the plane and refreshing these rows) is a named, owed rung; until it lands, treat an id here as a pointer to look up, not a status to trust.

Honest verdict. The coverage above records source-symbol presence. Stronger comparisons are authored claims, not measured runtime superiority or accepted outcomes. A watch contract names an expected source symbol; finding it changes source discovery only. Qualification, deployment and runtime acceptance require their own evidence.

Person · product · place — not yet measured for this domain

Every compare carries this layer. Declare knowledge/compare/speech.ppp (rows surface|nishi or c1..c4|label|url|connect naming OUR live surface and each rival's front door), run nx_ppp_probe domain speech, and this section fills itself on the next beat: the same ruler on both sides — privacy and CX (third-party hosts, tracker classes, cookies, security headers), design and longevity (design hygiene, computed WCAG contrast, render-blocking resources, unsized media, script weight, theme and motion queries), findability (landmarks, skip link, on-site search, breadcrumb, headings, internal links).

Exclusive and shared capabilities

One capability is graded on one board; every other board references it. The fleet duplicate census (nx_rivalcensus dupes, on the fleetdupes clock row) reads every board's matrix and names each capability graded as a row on two or more boards, with a home derived from evidence: rung when exactly one board's plan contracts the symbol (that board is starred), gates when no plan contracts it and exactly one board's executed gate map runs the organ's gate (that board is plussed), ambiguous when several boards carry the deciding evidence (a decision for their seats), and none when no evidence decides it, with the reason beside it: unmapped (no board of the duplicate declares a gates map; declare the maps) or unproven (a map exists and runs none of the organ's gates; gate the organ on its proving board, or declare it a shared library by policy). A home elsewhere means this board should carry an @xrow reference to the home board's row rather than a graded row of its own; a home here means the other boards should. The table is this board's slice of one estate-wide census: the shape is shared, the rows are the board's.

This board grades 35 capabilities: 32 exclusive to it and 3 shared with other boards. Shared rows on this page: 3 = home here 0 + home elsewhere 0 + ambiguous 0 + undecided 3 + refused 0 (partition 3); undecided = unmapped 3 + unproven 0 + unevidenced 0; homonyms 1; since the previous census: new 0, resolved 0. Census FRESH age_s=7232.

Capability	Organ	Boards grading it	Home	Rule	Evidence	Standing on this board
main	`-`	browser:runtime/_hdl_build/nx_uiq_perf.nx, graphviz:runtime/_hdl_build/nx_eco_graph_forest.nx, graphviz:runtime/_hdl_build/nx_eco_graph_png.nx, graphviz:runtime/_hdl_build/nx_eco_graph_forest.nx, graphviz:runtime/_hdl_build/nx_eco_graph_wiki.nx, familyatlas:runtime/_hdl_build/nx_eco_graph_png.nx, familyatlas:runtime/_hdl_build/nx_eco_graph_card.nx, ecosysdesign:runtime/_hdl_build/nx_eco_graph_arch.nx, typography:runtime/nx_font_ttf_emit.nx, koikatsu:runtime/_hdl_build/nx_retarget_emit.nx, pmdash:runtime/nx_dr_ocm_cli.nx, failmodes:runtime/nx_compare_refs_gate.nx, ecosystem:runtime/nx_compare_rank_fleet.nx, gen:runtime/_hdl_build/nx_gen_orchestrator_daemon.nx, capregistry:runtime/nx_catalog.nx, capregistry:runtime/_hdl_build/nx_coe_intoto.nx, capregistry:runtime/nx_toolreg_reconcile.nx, capregistry:runtime/nx_capsweep.nx, datavis:runtime/_hdl_build/nx_chartlay_gate.nx, speech:runtime/nx_voice_eval.nx	—	homonym	-	homonym
bks_search	`nx_book_search_lib.nx`	sharehub, speech, scripturestudy	—	none	unmapped	undecided
nx_voice_read	`nx_voice_read.nx`	speech, scripturestudy	—	none	unmapped	undecided
g2p	`nx_tts_speak.nx`	speech, scripturestudy	—	none	unmapped	undecided

Gaps from the record — what the estate does that no board carries

The record census (nx_goalmap record) reads the invoked-tool population and every plan queue row and files each organ or directive that NO matrix, plan or gates row names. A row here is a callout the boards missed: adjudicate it onto a board or declare it infrastructure. Census state STALE (age 1396337 s), sources read 4 of 7 declared — a BLIND census is a FLOOR: unread sources can only add rows.

kind	name	board	source	evidence

rows shown 0|this board's directives 0|estate-wide un-boarded organs 452 (listed in full on /compare/ecosystem)|census rows 639|malformed 0 (counted, never rendered)

References

Beyond a link list. Every reference below resolves twice — the publisher's copy and, where banked, the estate's own non-rottable library mirror with a content pin — and carries its evidence class plus the exact claim on this page it grounds. Keyed marks like [key] in the matrix notes jump here. A dash means honestly absent, never assumed.

[arxiv_tts_2026_09] arXiv API listing all:text-to-speech sorted by submission date, 30 entries, newest 2026-09-18 (GenTraceBench, StepAudio 3 Gen, Taming Long-form TTS, VoiceMOS 2026, dictionary-constrained G2P) publisher · read in our library knowledge/fetched/cmp_speech_arxiv_tts_2026-09.xml · pin hc4e39860589157d2d6cbf9cb725cacfc3988c98d0b13131abb81fa2ae9182d7b · accessed 2026-09-21 · mirrored-indexGrounds: Codec-token TTS decoder on the integer engine
[arxiv_clone_2026_09] arXiv API listing all:voice cloning OR zero-shot TTS sorted by submission date, 30 entries, newest 2026-09-15 (Cross-Lingual F5-TTS 2, zero-shot temporal localisation of deepfakes, a survey of AI voices and their detection) publisher · read in our library knowledge/fetched/cmp_speech_arxiv_clone_2026-09.xml · pin h48e70ecc2dec4b83dc0081a13cd4b60914ad5ea281899d397fd050410d3ca8c1 · accessed 2026-09-21 · mirrored-indexGrounds: Cross-lingual transcript-free cloning
[kokoro82m] Kokoro-82M model card on Hugging Face (hexgrad): licence apache-2.0, G2P by the misaki library publisher · read in our library knowledge/fetched/cmp_speech_kokoro82m.html · pin h08e144fe995be179fb8e4fa9143cc8413bab81d54d30662a14571d91d305e70e · accessed 2026-09-21 · vendor-docGrounds: Open-weights voices as data read by our engine
[chatterbox] Chatterbox repository README (Resemble AI): MIT licence, built-in PerTh watermarking with an extractor, zero-shot cloning from a reference clip publisher · read in our library knowledge/fetched/cmp_speech_chatterbox.html · pin he77a4b84dbc0523464f6e45f3b3e54f62df93640e153943cfdd8b3e4bc5656ab · accessed 2026-09-21 · vendor-docGrounds: Provenance mark and attribution on every clip
[kyutai_tts] Kyutai TTS 1.6B en_fr model card: weights CC-BY 4.0, decoder-only streaming text-to-speech over the delayed-streams architecture, English and French, voices from a repository publisher · read in our library knowledge/fetched/cmp_speech_kyutai_tts.html · pin h8584ebeb2943660551930832e5927f559f39d731947ea2eeecd443adc29f4792 · accessed 2026-09-21 · vendor-docGrounds: Streaming synthesis with word timestamps
[libritts_r] LibriTTS-R on OpenSLR 141: licence CC BY 4.0, speech category publisher · read in our library knowledge/fetched/cmp_speech_libritts_r.html · pin h6b0262043e82cadfce2bdb514b09f17d04a20895f0deb4eae8a74e9ade83ba60 · accessed 2026-09-21 · datasetGrounds: Corpus with licence for training and evaluation
[utmosv2] UTMOSv2 repository (sarulab-speech): automatic MOS prediction for synthesised speech publisher · read in our library knowledge/fetched/cmp_speech_utmosv2.html · pin h299e98ce0d7dd89d5c072d412c5b2da3116a583e0235439395967b4e2044e760 · accessed 2026-09-21 · vendor-docGrounds: MOS oracle run outside, sovereign twin owed
[seedtts_eval] seed-tts-eval repository (ByteDance Speech): WER and speaker-similarity test sets and scoring scripts for zero-shot TTS publisher · read in our library knowledge/fetched/cmp_speech_seedtts_eval.html · pin hef8ce10e50abc36c2874141896f1b9e65aea841ae6603ec0e37c89bb3a3dc1b5 · accessed 2026-09-21 · vendor-docGrounds: WER and speaker-similarity test sets
[aa_tts_arena] Artificial Analysis text-to-speech models page, the canonical the arena URL redirects to (script-rendered leaderboard; the mirror witnesses the surface, a numbers read is owed) publisher · read in our library knowledge/fetched/cmp_speech_aa_tts_models.html · pin h18ea42d384d174aa12775eabd22e90a94832500231a06a9c621573feb2e2498b · accessed 2026-09-21 · mirrored-indexGrounds: Human preference arena as a bar

generated by nx_swcompare_matrix (sovereign NishiLang organ) from knowledge/compare/speech.matrix · source checks show implementation presence; runtime and user-outcome evidence are reported separately · JavaScript supports page controls