Architecture Application Roadmap

Architecture Application Roadmap

> Built on 2026-03-15 from `meta-candidate-mining`, the current `evo-cube-output/` inventory, and the `backlog/code4ai-batch/` findings.

Canonical Docs

Use these as the entrypoint for the current program state before reading historical stage files:

• `meta-evolution-overview.md`
• `meta-evolution-current-state.md`
• `meta-evolution-folder-map.md`
• `meta-evolution-karl-bridge.md`
• `wave2-staging.md`
• `README.md`

Decision

Do not keep generating disconnected cubes. Collapse the research into a linear application program:

1. map completed cubes onto the 25 mega-cube registry,
2. extract only the findings that improve shared architecture,
3. apply those deltas in waves,
4. resume automated cube generation only after the control plane and learning loop are stable.

New research videos are evidence feeds into this program. They should update priority, validate assumptions, or add deltas to an existing mega-cube. They should not automatically become another isolated folder.

Current State

• `meta-candidate-mining` already defined the fleet-wide control problem: 114 candidates collapsed into 25 mega-cubes.
• The original plan stalled because the control-plane files were never bootstrapped: no registry, no priority override file, no cross-synthesis template, and no SFT extraction scaffold.
• Several mega-cubes are effectively backfilled already by existing outputs in `Desktop/evo-cube-output/`.
• The code4AI backlog already extracted the most reusable external research into architecture deltas: 54 videos mined, 10 cubes processed, 334 tasks extracted, and 3 convergent critical bugs around the KARL or Cortex learning loop.

What Counts As Applicable Research

Only apply findings that satisfy at least one of these tests:

1. They improve a shared substrate used by multiple systems.
2. They repair a broken feedback loop, data path, or evaluation path.
3. They remove duplicated architecture by collapsing multiple outputs into one primitive.
4. They unblock revenue-facing or fleet-facing product surfaces.

If a finding is interesting but does not change a shared substrate, it stays as reference material until it finds a real host in the registry.

Linear Waves

Wave 0: Control Plane Recovery

Purpose: turn meta-mining into an operational program instead of a historical plan.

Deliverables:

• `Desktop/evo-cube-output/mega-cube-registry.md`
• `Desktop/evo-cube-output/priority_override.json`
• `Desktop/evo-cube-output/cross-synthesis-template.md`
• `Desktop/voice-corpus/evo-cube-sft/extract.py`
• a current mapping from existing cube outputs to the 25 mega-cubes

Architecture effect:

• new research has a place to land,
• execution can be reprioritized without rewriting the whole plan,
• existing cube output can be exported into training data,
• the system gets one canonical registry instead of chat-memory drift.

Exit gate: complete.

Wave 1: Learning Loop Integrity

Purpose: apply the highest-signal external research to the architecture where the stack is currently blind.

Primary source cubes:

• `backlog/code4ai-batch/` Cube 1: CARL-KARL Trajectory RL
• `backlog/code4ai-batch/` Cube 4: System 3 Reward-Free RL
• `backlog/code4ai-batch/` Cube 5: RL2F Cognitive Twin Training
• `backlog/code4ai-batch/` Cube 9: FlowRL-KARL Integration

Target files and systems:

• `[home-path]`
• `[home-path]`
• `[home-path]`
• `[home-path]`
• `[home-path]`
• `[home-path]`
• `[home-path]`

Why first:

• the code4AI backlog found three convergent critical failures: null outcome signals, missing skill labels, and zero Cortex corrections,
• all later RL, routing, and twin improvements are contaminated if the learning loop is blind,
• this is the cleanest example of external research directly improving the current architecture.

Exit gate:

• corrections are captured,
• skill labels exist on trajectories,
• reward calculation is no longer a single narrow metric,
• evaluation and export paths are coherent enough to trust.

Wave 2: Shared Runtime Primitives

Purpose: apply cube findings that should become shared runtime layers instead of remaining local experiments.

Primary mega-cubes:

• #6 Multi-Agent Protocol
• #8 Comp-Core Production Map
• #11 Observability + Healing
• #13 Mesh Infrastructure
• #17 Pane Orchestration
• #18 Graph + Vector Intelligence
• #20 Security + Credentials

Target outputs:

• unify message transport and protocol boundaries,
• canonicalize duplicate or dual-location infrastructure,
• wire security in shadow mode before hard enforcement,
• give pane orchestration and health loops one runtime model,
• align graph and vector retrieval with actual routing decisions.

Key rule:

When two cubes propose modifications to the same substrate, merge them into a single primitive before touching product surfaces.

Exit gate:

• one transport story,
• one infrastructure map,
• one pane scheduler story,
• one security observation layer,
• one retrieval and memory substrate story.

Wave 3: Product Architecture Roll-Forward

Purpose: apply the shared substrates to the product and revenue surfaces that already have meaningful output anchors.

Primary mega-cubes:

• #2 Creative Content Platform
• #7 App Fleet Lifecycle
• #9 Creative Intelligence Triad
• #10 CRM + Revenue Pipeline
• #12 OpenClawHub Next Gen
• #24 Dream Journal + Garden

Target product surfaces:

• Koatji and Milkmen CRM,
• app fleet lifecycle and payment activation,
• creative content and Serenity production systems,
• OpenClawHub control surfaces,
• Spore and idea-garden capture loops.

Key rule:

No product wave should invent a new substrate that Wave 1 or Wave 2 was supposed to own. Product waves consume shared primitives; they do not replace them.

Exit gate:

• product outputs are using shared primitives,
• revenue surfaces have clearer telemetry,
• creative systems stop duplicating orchestration logic.

Wave 4: Factory Activation

Purpose: only after Waves 1-3 are real, resume automation around meta-mining.

What gets activated:

• `evo_cube_factory.py` or equivalent flow,
• scheduled cross-synthesis reviews,
• export of approved cube outputs into the SFT corpus,
• priority overrides driven by new research intake.

Key rule:

Factory output is draft quality until it passes the same cross-synthesis and architecture-fit checks as manual outputs.

Exit gate:

• new research can be ingested without creating sprawl,
• new cubes land in the registry automatically,
• the factory is improving the architecture instead of multiplying folders.

Intake Rule For New Research Videos

For any new video or transcript:

1. ingest the transcript into the research archive,
2. map the insight to an existing mega-cube,
3. if it changes execution order, write an entry in `priority_override.json`,
4. if it changes shared architecture, record it in the next cross-synthesis review,
5. only create a brand-new cube if no existing mega-cube can absorb it cleanly.

This makes external research a routing input, not a folder generator.

Immediate Execution Targets

These are the next architecture changes worth doing after this bootstrap.

1. Repair KARL or Cortex correction capture and skill labeling from the code4AI backlog critical section.
2. Reconcile all proposed `reward_engine.py` changes into one weighted reward model.
3. Build the unified KARL evaluation layer before more training or export work.
4. Collapse protocol and pane orchestration findings into shared runtime primitives.
5. Canonicalize duplicate infrastructure surfaces called out by meta-candidate mining before adding more automation.
6. Use the new SFT extractor on the already backfilled mega-cube outputs instead of waiting for hypothetical future cubes.

What Gets Deferred

• full manual execution of all 25 mega-cubes,
• daily autonomous cube generation,
• new external research cubes that do not change a shared substrate,
• low-signal specialized tracks until the shared learning and runtime layers are repaired.

Working Principle

The meta-mining program should now behave like an architecture governor:

• research comes in,
• the registry absorbs it,
• cross-synthesis converts it into deltas,
• waves apply those deltas to real systems,
• only then does the factory generate more material.

That is the correct collapse. The goal is not more cubes. The goal is a better architecture.