Adaptive Delta execution, sub-PV, and Delta Learning

Every executable Plan row uses one recursive Delta contract: exact entry, bounded work, optional mid-query refire, verified exit, auto-admitted row work and learning, then either the next Delta or a separate full-PV dual-HIL boundary.

Current counts are derived release facts, not permanent ceilings.

Complete recursive flow

flowchart TB
    Prompt["Prompt or steer"] --> Intake["Entry Slip + Source Intake + ChatLineage"]
    Intake --> Plan["Sole active Plan row + linked Deltas"]
    Plan --> Entry["Adaptive Delta entry"]
    PrevPV["Immutable full-PV baseline PV(n-1)"] --> Entry
    PrevSub["Latest verified predecessor sub-PV"] --> Entry
    PrevLearn["Auto-admitted predecessor Delta Learning"] --> Entry
    Entry --> Query["Bounded current-authority query across lanes and authorities"] --> Work["Current dirty implementation"]
    Work --> Mid["Mid-query / no-hit refresh and one refire"] --> Validate["Exact task acceptance + source/test/install disposition"]
    Validate --> Exit["Verified adaptive Delta exit"]
    Exit --> Refresh["Changed-only refresh: sectors, Learning, Canon, Memory, Universe, connector"]
    Refresh --> SubPV["AUTO_ACCEPTED_DELTA_ROW_WORK"]
    Refresh --> DLearn["AUTO_ACCEPTED_DELTA_LEARNING"]
    SubPV --> Next["Next Delta entry"]
    DLearn --> Next
    Exit -->|ordinary row| Next
    Exit -->|full-PV HIL row only| Candidate["Unaccepted full-PV candidate + bounded learning weave"]
    Candidate --> PHIL["Project HIL"]
    Candidate --> LHIL["Consolidated Learning HIL"]
    LHIL --> Weave["Learning decision"]
    PHIL --> Fuse["Exact Fuse after matching Learning weave"]
    Weave --> Fuse --> Accepted["Accepted pointer + root ZIP + Project Overlay"]
    Validate -. mismatch .-> Fail["Fail closed; row remains active"]

Entry contract

DELTA_ENTRY_AND_CURRENT_AUTHORITY_QUERY fires on ACTIVE_PLAN_ROW_ENTRY. It binds the active Plan row, linked steers, immutable full-PV baseline, latest verified sub-PV, accepted Delta Learning, live sector lanes, Agent Learning, Canon, Project Memory, Project Universe, connector integrity, AGENTS.md, and host MEMORY.md.

Hooks are optional for explicit entry actions. Accepted archives are never queried. Re-entry reuses the existing entry formula; it does not append a second entry or rebuild the Plan from chat.

Mid-query contract

IN_DELTA_BOUNDED_QUERY_AND_NO_HIT_REFIRE is triggered only by TASK_SCOPE_CHANGE, BOUNDED_NO_HIT. Its no-hit order is LIVE_SECTOR_FALLBACK → AI_LEARNING → CANON → PROJECT_MEMORY → PROJECT_UNIVERSE → CONNECTOR_BRAIN. It refires at most once and never substitutes an accepted ZIP.

Exit contract

DELTA_EXIT_APPEND_REFRESH uses task_confirm_source_update, adaptive_delta_exit, pv_task_transition and refreshes the exact changed authorities only after targeted validation passes. Ordinary rows cannot refresh Project Overlay.

Auto-admitted states versus full-PV HIL

Sub-PV row work and Delta Learning are verified continuing-work inputs, not miniature full-PV approvals. They are automatically admitted at Delta exit, reused by the next Delta, and retain exact receipts. They create no individual HIL, candidate, Overlay, accepted ZIP, or full-PV pointer movement.

FULL_PV_DUAL_HIL is distinct and requires two explicit human decisions. The Fuse contract declares individual_delta_learning_hil=false, project_and_learning_decisions_separate=true, and acceptance authority PLAN_ROW_DUAL_HIL_STAMP.

The accepted ZIP is post-approval snapshot storage only and is never an entry, query, Learning, or State Travel source. Project Overlay is created only by the full-PV HIL path after the exact Project decision permits it.

Source-bound workflow map

This page is projected from the same current executable snapshot as the rest of the documentation set. The map is deliberately two-directional: each horizontal district shows peer stages while vertical edges show ownership and state progression.

flowchart TB
    subgraph InputDistrict["Input and classification"]
      direction LR
      A["Prompt or execution-changing steer"] --> B["Entry Slip and Source Intake"] --> C["Adaptive Delta entry"]
    end
    subgraph ExecutionDistrict["Selection and execution"]
      direction TB
      D["Active Plan row plus predecessor sub-PV"] --> E["Bounded work and mid-query refire"] --> F["Exact acceptance and installed-behavior checks"]
    end
    subgraph EvidenceDistrict["Evidence and outcome"]
      direction LR
      G["Verified Delta-exit receipt"] --> H["Auto-admitted sub-PV plus Delta Learning"]
      G -. mismatch .-> I["No full-PV pointer, Overlay, ZIP, or inferred HIL effect"]
    end
    C --> D
    F --> G

Contract and readback

Canonical source owners

• sdk/delta/entry-mid-exit.v1.json
• sdk/delta/entry.workflow.v1.json
• sdk/delta/exit.workflow.v1.json

Exact backend readback

Cross-surface invariants

• The current snapshot contains 91 public actions, 26 skills, 11 hook events / 44 handlers, 119 tool requirements, 18 sector lanes, and 11 named authorities. These are derived counts, not fixed ceilings.
• Executable ownership stays one-way: skills select, MCP exposes, the outer SDK routes, the internal SDK executes, ENV selects, UOP governs, tools perform bounded work, hooks emit receipts, and the owning authority validates effects.
• Any missing identity, schema, grant, capability, dependency, receipt, or authority proof must fail closed at its owning phase; a later green check cannot retroactively authorize the skipped boundary.
• A changed route refreshes every dependent schema, manifest, generator, test, diagram, and documentation reference; the superseded executable route is directly purged in the same Delta.
• Tests, Git, CI, installation, restart, deployment, discussion, or a rendered page never imply Project HIL, Learning HIL, Goal completion, or pointer movement.

This page is a Git-tracked documentation projection. Executable source, SQLite authorities, installed-runtime receipts, and explicit human gates remain the governing evidence.