escrow: Aurora autonomous flywheel thesis — defined home + expiration + one-test gate + no-multi-reviewer rule (cross-AI converged refinement, Aaron 2026-04-29)

[AceHack]

Summary

• Establishes the escrow at its canonical home: docs/research/escrowed/aurora-autonomous-flywheel-thesis-2026-04-28.md
• Per Aaron's clarification: thesis is escrowed, not rejected. Bounded preservation, not deferral.
• Per cross-AI converged refinement (second-opinion-reviewer + deep-research collaborator), the three load-bearing additions: defined home, expiration trigger, one-test reopen gate. Plus the no-multi-reviewer-loop rule.

Aaron's clarification

"This is not rejected. It is escrowed. The thesis is important enough that we do not let it land sloppily. Don't bury it, don't bloat it, and don't rush it. Put it in escrow and make it earn its way into the system."

The harsh pushback from external-AI reviewer across two consecutive rounds was protecting the thesis, not killing it.

What landed (vs original B-0094 backlog row)

This PR replaces the original backlog-row form with the cross-AI-recommended escrow shape:

Change	Why
docs/backlog/P3/... → docs/research/escrowed/...	Escrow is research-grade preservation, not a work-row. Single canonical location prevents loss.
Drop backlog frontmatter (id/priority/status); add §33 archive header (Scope/Attribution/Operational status/Non-fusion disclaimer)	Escrow is research, not backlog work
Add status header block at top	Fast-path read in 5 seconds
Add expiration trigger (4 substantive rounds)	Prevents "escrowed indefinitely" failure
Clarify ONE-test reopen gate	Six-test bar would be impossibly hard
Add explicit no-multi-reviewer-loop rule	Directly addresses carrier-laundering surface

Status of escrow

Status:            Escrowed until first prototype result
Gate:              Candidate-count scanner self-destruct test
Reopen condition:  one meaningful prototype result (pass OR fail with useful signal)
Multi-AI loop:     paused until prototype result
Expiration:        4 substantive rounds without prototype run → surface for explicit decision

What this file does NOT authorize

• Does NOT advance the thesis
• Does NOT integrate thesis components into operational substrate
• Does NOT extend the Aurora bridge
• Does NOT mutate Aurora core math
• Does NOT authorize forwarding the thesis to multi-AI loop

What this file DOES authorize

• Preservation of the thesis across session boundaries
• Reading the thesis as deferred candidate substrate
• Bilateral clarification with Aaron (single-channel, not multi-reviewer)
• Implementation work on the falsifier-gate prototype

Test plan

• Single file at canonical-home location
• §33 archive header on first 5 lines (research-grade hypothesis, not operational)
• Status header block at top (Status / Gate / Reopen condition / Multi-AI loop / Expiration / Created / Last surfaced)
• Expiration trigger explicit (4 substantive rounds; minute ticks excluded)
• One-test reopen gate explicit
• No-multi-reviewer-loop rule explicit
• Composition with already-landed substrate named (PR research(aurora-immune-governance-bridge): minimal first artifact — three immune translations + falsifier + prototype (Aurora converged + Ani falsifier-first + multi-AI consensus 2026-04-28) #707, PR docs(AGENT-BEST-PRACTICES): land session-closure rule as small operational artifact (cross-AI converged stance) #712, bead-promotion memory, Beacon-promotion memory)
• Conversation-log lineage pointer for full thesis content
• Cathedral-shape avoided: bounded scope, recursive escrow application

The keeper line (preserved verbatim from cross-AI converged stance)

If the thesis is real, it will survive waiting. If it cannot survive waiting, then the theory was already describing its own capture.

Pausing IS the test of the thesis, not an obstacle to it.

[Copilot]

Pull request overview

Adds a new P3 backlog row to “escrow” (bounded preservation) the Aurora Immune Governance flywheel thesis until the stated prototype produces a measurable result, keeping the thesis discoverable without integrating it as active substrate.

Changes:

• Adds backlog row B-0094 with escrowed status and explicit constraints on what the escrow does/does not authorize.
• Defines a falsifier-gated set of conditions required to transition out of escrow.
• Records preservation pointers and composition links to related rows/PRs.

[Copilot]

P1: status: escrowed introduces a new status value that isn’t in the backlog schema docs (tools/backlog/README.md currently documents open / closed / superseded-by-B-NNNN / deferred). Either (a) map this to an existing documented status (e.g., deferred) and keep “escrow” as a tag/body concept, or (b) update the schema documentation/tooling in the same change set so readers and future linters don’t treat this as drift.

[Copilot]

P1: composes_with is documented as a list of B-NNNN IDs, but this frontmatter includes PR-707. If you want to keep PR cross-links, consider moving them into the body (where other rows already use “PR #…”), and keep composes_with limited to backlog IDs for consistency and easier future linting/indexing.

[Copilot]

P2: Status naming/casing is inconsistent: this section says “ESCROWED” → “ACTIVE-RESEARCH”, but later the target is written as backticked active-research. Recommend using the exact literal status tokens consistently (prefer backticked lowercase) so it’s unambiguous what frontmatter value a future edit should use.