Tab to a result · Enter to open · Esc to close

DOCUMENTATION / VERIFICATION

The first live run

Evidence from a read-only team running in native Conductor sessions.

Adapted from toolkit · f5911f6

The read-only pilot passed on November 16, 2026 (UTC). A real 00-Agent in a separate Conductor workspace received an assignment, sent progress and completion reports, and returned an immutable source check. The Secretary kept routine progress quiet, checked the result independently, and put the verified result in the Queen inbox.

The public receipt records the versions, profile, message identity, source hash and observed results. It contains no raw transcripts, account identities, credentials or host configuration. Workspace and session identifiers are fingerprinted.

What ran

ItemValue
Monarchy source9118e3927218cd248c5bc2f99e70005cca985c55
Conductor CLI0.1.0; exact CLI artifact hash retained in the receipt
Selected profileCodex, gpt-6-astra, ultra, fast mode enabled
Observable native profileModel, effort and fast mode matched; agent type was not exposed by session readback
ProjectPISCES, in a fresh Conductor cloud workspace
AssignmentRead the exact Git revision and README bytes, report their hash, and leave tracked files unchanged
Source checked54df6526125a2377ad8884e4116cee387819aaed

Conductor first rejected a Monarchy workspace because that repository was absent from the machine's repository list. The exact-name readback found no created workspace. The pilot then used an allowed PISCES workspace and cloned the exact Monarchy source under .context. No machine settings were changed.

The 00-Agent read the checkpoint and verification skills and used monarchy.conductor.report_text to validate its reports. This exercises those instructions in a real assignment; it is not a scored evaluation of the skill pack.

The receipts tell the story

  1. Dispatch recorded its intent before sending. Immediate readback did not yet show the message, so the operation stayed unknown.
  2. A fresh process reopened the journal. Readback found the original request ID and confirmed delivery. Recovery did not issue another send.
  3. The progress report arrived with native sender metadata. Ingestion verified both workspace bindings and the active assignment. Triage consumed the routine event and left the Queen inbox empty.
  4. The completion report passed the same sender checks. It created work for the Secretary and left the task incomplete, pending verification.
  5. The maintainer independently read the reported Git object from another checkout. Its README had 11,686 bytes and the same SHA-256 as the 00-Agent's report. A separate review repeated that check and reread the native report.
  6. Verification created a result for the Queen. Reopening the database returned the same unprocessed presentation. Explicit acknowledgment, followed by another reopen, returned an empty inbox.

The 00-Agent reported the same full revision and a clean tracked tree before and after its work. The independent content check confirms the immutable README bytes. It does not independently observe every filesystem action inside the 00-Agent's VM.

Keep the scope of the proof clear

This run qualifies native message dispatch, delayed readback recovery, reports across separate workspaces, quiet routine triage and a reviewed result. It did not exercise a new human decision or prove how a host authenticates owner permission. The full decision-and-answer lifecycle remains covered by the offline integration scenarios.

The session was created through the native workspace command. The adapter's separate create_session path was not exercised live. Explicit fast mode off remains held because this CLI cannot express it during creation. The run also did not inject a native process crash, restore a lost host, run application tests, install a background service, or evaluate the 72 skill scenarios.

The runtime remains opt-in. Native owner input, automatic prompt protection, additional profiles and unattended operation each need their own qualification. Reading one successful receipt should make Monarchy's limits easier to see.