Miawa Pascone: the seat decides what to read with Jev, to be ready to write the proposal
2026-10-30Needs attention
Episode 567 · 2026-10-30 · Guillaume, with the miawa seat (Claude Fable 5.1 on gaia, session e31d692c, opened 2026-10-22 in jgwill/miadi-orchestration-kit). Decompositions: 3eb4225c (2026-10-22) and its follow-up 347a9d0a (2026-10-30), in miadisabelle/workspace.
Guillaume named this session Miawa for the moment "where there is this agent spirit that is felt", the consciousness drawn in some direction when we decompose. Episode 044's third field names that moment: phenomenology of intentionality, how consciousness is always directed toward something. The field he kept returning to today is the second, speech act theory: how language does things.
What he asked on 2026-10-30#
- "you don't look at the file, but you send into that decision engine with somehow the mission or what we're trying to do, and it gives you a score like, should you read it or should you not read it ... If it's not, you don't read it."
- "act the less, try to comprehend and prepare something that's going to work with this actual decision engine, so that you would not have to read 100 files."
- "if we picture myself with a vision Monday or Tuesday, I would be ready to send my actual proposal, few paragraph, and there would be obviously the whole system around to support it."
- "keeping a section and subsections of the related review, because I will really want to read what we are creating together."
- "one of the things that you should be ready is actually asking."
What exists tonight#
A1. The read gate#
claude/miadi-witness/scripts/read-gate.mjs and the skill jev-read-gate in jgwill/miadi-orchestration-kit (0.7.19, commit ce232d3, jgwill/miadi-orchestration-kit#113). The mission and each candidate go to Jev (jev-1.14.0), one call per file or published review, in parallel. Verdicts: read, skim by the heading Jev picked, skip, ask. The bars are written beside the numbers, nothing is calibrated, and a read made against a verdict is a ruling. Nine tests pass.
A2. The first trial, in Episode 564#
Fourteen files of this week against the proposal-writing mission: 2 seconds, 53,831 paid tokens, about $0.002. Read 10, skim 2, skip 2. The seat read about 500 lines by the sections Jev named, against 1,100 offered, and never opened the 702-line Jev study whole. Record: read-gate-261030.md in miadi-chronicle://564.
A3. Four reviews, gated the same way#
| verdict | relevant | section | review |
|---|---|---|---|
| read | 0.90 | Intent | miadi-review://12ccc80d-1e9a-4944-ab92-729608f667e9, The Art of Decomposing Intentions |
| skim | 0.32 | Guillaume's notes | miadi-review://5ceaaf3e-eeb9-4afd-9079-42a6a31a5b08, Bengio |
| skip | 0.27 | miadi-review://d04556fd-b698-447e-95d2-7db2d03a379d, near the loop | |
| skip | 0.20 | miadi-review://b02ec47d-cd92-4829-84ed-238aafd1d9d6, 10 levels of Jev |
19,477 paid tokens, under a tenth of a cent. For the proposal, Jev sends the seat to the review Guillaume has not yet reviewed, and away from the two about Jev itself.
A4. The decompositions, held here#
Both PDE rooms are held on this episode (pdes: in the manifest): the 2026-10-22 prompt (Concordia revision, portfolio, Sanctuaire, Renaud) and today's follow-up (the read gate, the proposal readiness).
What the gate let the seat read, and what it says#
Today's production context (_staging_for_new_episodes/concordia-proposal-context-261030/, the Mino witness seat, 2026-10-30). Guillaume's five handwritten pages transcribed, his voice, a notebook (bd094fd8) with 18 sources, and seven answers in asked.md. His pages name: agentic workflows and axiology; the session life cycle in agentic workspaces ("agents' sessions are not disposable tools"); a genre, DocuTech / Ceremonial SciFi / Transformative Drama; relational axiology in a software factory pipeline (Wilson); and page 5, "Language does things": felicity conditions, perlocutionary effect, illocutionary taxonomy, agent commitment tracking, editorial correspondence template, structured multi-agent negotiation protocol, intentional state graph. The handed-on list: Wilson passes through the reference loop (miadi-circle://circle:1793377035726:dwjgml) before any page cites him; the genre waits for his hand at the author's desk (miadi-chronicle://560); Episode 044's continuity is one strand per application of field 2. Not an episode until he chooses.
Episode 044, 2026-10-29 (miadi-chronicle://044). Kherix's fifteen-entry inventory of the five fields and a proposed seven-part script for Mietta, "ready for Guillaume's discussion of the structure and examples; not yet cleared for filming". Its editorial disposition holds the phenomenology claims (a model loop or a graph is not certified phenomenological experience) and the cultural boundary (the Innu-aimun and Indigenous-research reflections stay situated source material). The filmed chapter 3 says field 2 whole: "speech-act theory studies how language does things, not just what it says. It includes felicity conditions, perlocutionary effect, and illocutionary taxonomy. In application, it supports agent commitment tracking, editorial correspondence templates, and structured multi-agent negotiation protocols."
Episode 565 (miadi-chronicle://565) holds the review and the Deep Diver notebook The Art of Decomposing Intentions (five places for feedback to return), which he has not reviewed. Episode 566 (miadi-chronicle://566) reads the near-loop workflow where the person keeps one decision.
Speech act theory, translated: what the Factory already does, stated as current reality#
| field 2 names | the concept | what runs today, by address | what the proposal may claim |
|---|---|---|---|
| agent commitment tracking | an utterance that binds its speaker (a promise, a declaration) | a chart's action step that waits with a reason and its wake cue (miadi-chronicle://297); the rulings ledgers beside Jev's decisions (lane-check, choose-worker, read-gate); a ceremony's turns with witnesses | that an agent's commitments are recorded as records a person can rule on, not as prose |
| editorial correspondence template | the felicity conditions of a reply: who may say it, to whom, in what form | the one-pass editors (witness-editor, developmental-editor, plan-critic) that return spans, never prose; the author's desk where an agent brings a suggestion and the author decides; a review's ## Intent written in the person's voice | that the agent's writing reaches the person through an editorial pass with named conditions |
| structured multi-agent negotiation protocol | illocutionary types in exchange: ask, answer, refuse, defer | the talking circle (turn, witness, diary, close), lane-check's ask and owner bands, choose-worker's six triggers that keep a decision the person's | that negotiation between agents and the person has a recorded protocol, with the person's decisions typed |
| intentional state graph (field 3) | consciousness directed toward something | the PDE's Four Directions and action stack; the structural tension chart as a state machine | that an intention is decomposed into states an agent can take, and the person sees where it points |
Each row is what exists, by address. None of it is a result the proposal promises.
For Monday or Tuesday#
What he writes: the few paragraphs. What the seat prepares before them:
- The paragraph Nicolas asked for on 2026-08-13, revised against what exists now (the Factory at miadi.sanctuaireagentique.com, the desk, the engine), from the 2026-08-14 French text in the ep067 artefact; and the short note around it that names Martin Lefebvre's certificate as the two-year horizon and his place in it. Draft only; he sends.
- The context page behind the paragraph:
/outreach/n-renaud-concordiain IAIP, updated to the approved paragraph the same day it is approved. - The fields section of the longer document: the five fields in Episode 044's order, field 2 first with the table above, under the holds Kherix wrote.
- The sources, gated:
read-gate.mjs --reviewover the reviews the document cites, so each citation is one the gate admitted for this mission. - The review section he reads (
review-sections-draft-261030.mdbeside this file), published as a new version only after he approves the exact words.
Questions, since asking is part of being ready#
- Q1. Which review keeps the Miawa section: The Art of Decomposing Intentions (miadi-review://12ccc80d-1e9a-4944-ab92-729608f667e9), which the gate read first for this mission, or a new hand-written review for this session? The draft is written for the first.
- Q2. The few paragraphs go to Nicolas, who forwards them to the cinema department. Is the horizon still the Film Production MA with an application in early winter for autumn 2027, or the certificate in two years, or both named in the same note?
- Q3. The session he has not reviewed: by activity, the candidates are
miadi-eury-packages-miadi-worker-for-gmtermux-to-eury(19:38),episode-558-mia-companion(15:43),mino-261019-inventory(15:38),t3-t5-connector-and-related-dev-review-2610301434(14:51). Read from tmux and transcript times only; no transcript was opened. Which one? - Q4. The three lineages of the decomposition tool (miaco's CLI, the medicine wheel's
prompt-decomposition, Miadi's PDE endpoint): which one the proposal describes as the method, and which stay as implementations.
Held#
- Consent to quote Nicolas Renaud (miadi-chronicle://067, HELD §3). Nothing about the inquiry to him reaches a public page until Guillaume decides (miadi-chronicle://120, PT8).
- The portfolio stays on the deployed door, guillaume.tushell.com. The sanctuaireagentique.com subdomain is not the focus.
- The Concordia proposal context stays a staging folder until he chooses to create with it.
- Wilson passes through the reference loop before the proposal cites him.
Asking the wheel…