The method, with its receipts.

The framework lives as files with distinct jobs: direction, current state, an append-only event record, public artefacts and release receipts. A new session can reconstruct the work from the folder.

AGENTENV-SITE/
├── CLAUDE.md          # rules and deploy pipeline
├── BRIEF.md           # purpose, decisions and claims boundary
├── STATUS.md          # current operational truth
├── LOG.md             # append-only event record
├── site/              # public framework and this evidence room
├── resources/claims/  # direct authority for public wording
└── reports/           # QA, deployment and domain receipts

Status is deliberately separate from the event log. It records what is true now, what changed and what remains unresolved.

STATE: FRAMEWORK-RELEASE V5 DEPLOYED AND VERIFIED
SEAT: Kimi — framework-release seat
CONTENT: five mechanisms + five-step operating loop
CTA: Discuss the framework
SALES / PAYMENTS: none
DNS CHANGES IN THIS RELEASE: none

This is a faithful excerpt from the release state, condensed for publication.

The log is append-only. It preserves the direction change, deployment and verification as separate events rather than rewriting them into a polished launch story.

14:18  FRAMEWORK-V5-DEPLOYED
       agentenv.pages.dev HTTP 200
       exact source match verified

14:19  CUSTOM-DOMAIN-VERIFIED
       agentenviroment.ai HTTP 200
       framework title / operating loop / CTA present

14:21  REVIEW-DISPATCH-BLOCKED
       no independent PASS claimed

Times are AEST on 29 October 2026. Entries are condensed; the meaning and unresolved review state are retained.

The release candidate was checked structurally, rendered in Chrome, deployed and fetched again. The staging response matched the reviewed source byte for byte.

STRUCTURAL CHECKS: PASS
MECHANISMS: 5
OPERATING STEPS: 5
WITHDRAWN VALUATION: absent
OLD WALKTHROUGH CTA: absent
STAGING HTTP: 200
CUSTOM-DOMAIN HTTP: 200
SOURCE SHA-256:
160074323c73f3eb9559e93709688b8d5773234d569f8a2b2064fa1a2a5c725a

These checks verify the release process and page integrity. They do not measure whether the framework improves company performance.

The public-copy review package was prepared for an independent BLACK/Codex gate. Its dispatch was refused because no supported BLACK process was running.

The failed dispatch remains in the record. The release proceeded under Jesse's explicit direction; that is human authority, not independent adversarial approval.

The framework is a field method released for inspection and challenge. It is not presented as settled science.

• No controlled evaluation yet. Twelve months of operating use is Jesse's account, not a benchmark or measured ROI.
• No invention claim. Filesystem memory, context engineering and multi-model review all have substantial prior art. The contribution here is their integration into a company operating method.
• Records can go stale. Persistent files need ownership, timestamps, supersession and correction. A ledger entry is a claim until checked against an outcome.
• Reviewers can share blind spots. Different models are not automatically independent. Deterministic checks and human judgement still matter.
• Permissions require real enforcement. “Bounded tools” cannot mean instructions alone. Consequential actions need external authority, allowlists and auditable receipts.
• This is not a runnable reference implementation. The public release currently provides the operating model and evidence room, not a packaged software product.

Publish repeated comparisons against a single-agent and simpler-harness baseline. Measure task success, cost, latency, human interventions and failure severity. Include cases where the extra protocol did not help.

Until then, the strongest honest claim is specific: this is the actual environment-heavy method Jesse uses, released with its records and limits open to inspection.