1/ ⚙️ MONARK Under the hood ⚙️
Yesterday: what Ukemi and Bell are for. Today: what they actually do, call by call. Same engine underneath: Shōgen perceives, Hikae gates, Narabi and Ukemi act, Bell is the next reading on that spine 🧵
2/ 🔗 Nothing is read from one source. Every on-chain fact (a fill, a log, a mint, a supply) is fetched from two independent RPC operators and must agree. One provider, or two aliases of the same one? No quorum, no print. The record says "no_quorum" and stops there.
3/ 🔔 Bell, step by step: pull the DEX fills of a tokenized stock over a session, recompute the VWAP from the raw fills (CI recomputes it bit for bit), read the mint itself for its multiplier, pause flag and delegate, then compare to the last cash close of the primary market. Not to a 24/7 token feed.
4/ 📐 The number Bell writes is the log gap between the token and the close, per session: overnight, weekend, holiday, halt. Volume goes in as a ratio to the underlying's average daily volume, never as a raw figure. Rebases and corporate actions are applied before the comparison, or the session is marked, not printed.
5/ 🛡️ Ukemi's recorder, step by step: scan the logs of the redemption contracts across a block range, five keyless providers, bounded retries with a backoff cap, duplicates removed on (block, index, tx), a provider that refuses the range is benched instead of splitting the query. The book it produces is hashed and pinned.
6/ 🚪 Then the gate. Hikae takes each attested reading and conforms it into a coverage region with a stated error budget. No committed calibration for that regime? It abstains by construction. Ukemi acts on the decision, not on the raw number. That is where conformal inference sits.
7/ 🔒 What never enters a record: API keys, provider URLs, the collector's output folder. Journals carry provider domains only, outputs live outside the repository, series committed to the repo are sha-pinned and declared next to their provenance file.
8/ 🧭 Multi-directional by design: the same attest → gate → act chain serves an agent through MCP, and gives birth to new instruments like MONARK Bell. Every handoff is a served path replayed by a non-LLM integration test. Not wired yet? It stays "upcoming". 🧪
Update 🕒
1/ 🏗️ MONARK Building 🏗️
We are building the full Ukemi, and Bell.
Both are measuring instruments, not scores 📏. Each output is an attested record: what was read, from where, with which residual left open. No probability of being right, ever.
2/ 🛡️ Ukemi reads the redemption pipes of stablecoins and wrappers on-chain: mints, burns, supply, redeem queues, reserve attestations, bridge locks. It records what happened during a stress window and publishes a book 📖 that a third party can replay line by line.
3/ 👥 Who needs Ukemi:
⚖️ risk curators setting collateral parameters on lending markets
🏦 DAO and protocol treasuries holding wrapped or yield-bearing dollars
📝 cover writers who need "the peg failed" to match a measurable event
🔍 auditors and researchers reconstructing a run after the fact
4/ 🔔 Bell reads tokenized stocks where they actually trade: DEX fills on Solana, BNB Chain, Robinhood Chain, Base, 24/7 🌙, against the last cash close of the underlying. Overnight, weekend, holiday, halt: Bell records the gap and the volume, session by session, with rebases and corporate actions applied first.
5/ 👥 Who needs Bell:
📈 market makers and LPs quoting tokenized equities out of market hours
🏛️ issuers who want an independent record of their token versus the cash print
🧮 lending markets deciding if a tokenized stock can be collateral, and at what haircut
📐 anyone writing rules for tokenized equities who wants measurements, not anecdotes
6/ 🚦 Status: both under review, gate by gate, with independent verification at every step ✅. Nothing goes live without a served path and an end-to-end test 🧪. We publish when it is proven, not before.