# Executive Overview — PoC, Pathways, and the Aquafied Collaboration Bundle

**Audience:** Tricia Wang, Ken Huang, and the AAIS Society PoC project committee  
**Prepared by:** Tesseract Stakes team (Pathways feedback collaboration)  
**Channel:** `djat-poc-20260602`  
**Date:** June 2026

---

## In one page

The **Proof-of-Control (PoC) Cryptographic Verification Standard** asks whether an AI deployment can produce independent, assessable evidence that it operated within declared boundaries — a binary, market-facing threshold insurers and procurement can use.

This repository is **structured feedback** on PoC Working Draft architecture v0.1.0 (May 2026), delivered not as slides or email but as a **sealed, self-describing, Aquafied collaboration bundle**: hash-bound artifacts, PathwayRun lineage, Aqua attestations on every Pathways template, and pre-registered falsifiable hypotheses.

> **Read first — v0.1.0 revision:** AAIS collaborator-synthesized **v0.1.0-rev** supersedes base for forward references. Domain model: Privacy, Portability, **Authorization**, Security, Identity (+ Provenance TBD). See [`sidecars/poc-v0.1.0-revision-delta-report.md`](../sidecars/poc-v0.1.0-revision-delta-report.md). Comprehensive/strategic PDFs remain anchored to **v0.1.0-base**.

**Pathways Architecture v1.1.0** is offered as the most mature existing substrate for PoC's proposed **sixth domain — Provenance** — and as the mechanism that could govern *dynamic* model updates, not static deployments alone.

The bundle itself may be exactly the **cryptographic context anchor** PoC needs: collaborative context bound before a model update runs, proof returned after, verifiable offline by anyone. One experiment — **H-POC-VALIDATE** — can alone demonstrate that loop. **Tricia Wang** is the current sole signatory for confirming policy validation when that loop completes.

> **This document ties PoC, the feedback, Pathways, application, and the bundle format into one narrative.** Detail lives in linked artifacts; integrity is checkable without trusting the authors.

**Verify now (anyone):**

```bash
python3 tools/collaboration-bundle/sign_bundle.py verify collaboration/20260602-160000
```

Status vocabulary: [`BUNDLE_STATUS_GUIDE.md`](../BUNDLE_STATUS_GUIDE.md)

---

## 1. Proof-of-Control — what the initiative is

PoC is an **Advanced AI Society (AAIS Society)** initiative co-chaired by Ken Huang and Tricia Wang to close the **Verifiability Gap**: the absence of independent evidence of what agentic AI systems actually do at runtime.

| Element | What it means |
|---------|----------------|
| **Binary threshold** | A deployment either has the Proof-of-Control property, or it does not — procurable like SOC 2 or PCI, not a maturity score |
| **Verification spectrum** | Self-Verifiable → Independently Verifiable → Cryptographically Verifiable — stage depends on evidence quality, not mechanism choice |
| **Five draft domains (v0.1.0-rev)** | Privacy, Portability, **Authorization**, Security, Identity — plus **proposed sixth: Provenance** |
| **Core evidence (v0.1.0-rev)** | Record of what the system did — not a separate domain (replaces base **Verifiability** domain) |
| **Market thesis** | Insurers, enterprise buyers, and regulators need a yes-or-no question before underwriting or approving agentic AI at scale |

**Timeline:** architecture v0.1.0 published May 2026; Working Draft 1.0 targeted October 2026; this bundle is June 2026 committee feedback.

**Deeper read:** [`PoC_Initiative.md`](PoC_Initiative.md) · [`PoC_Standard_v0.1.md`](PoC_Standard_v0.1.md) · sealed draft PDF in [`assets/origin/`](../assets/origin/ORIGIN_MANIFEST.yaml)

---

## 2. The feedback — what this bundle says about PoC

Feedback was produced through multi-model analysis, exported with full origin lineage, and sealed inside this bundle. Two derivative reports carry the substantive recommendations:

| Report | Question | Link |
|--------|----------|------|
| **Comprehensive Analysis** | Is the draft architecture credible and durable — especially for self-modifying AI? | [`assets/comprehensive-analysis-v0.1.0.pdf`](../assets/comprehensive-analysis-v0.1.0.pdf) |
| **Strategic Report** | How does Pathways Architecture address the gaps — should Provenance be the sixth domain? | [`assets/strategic-report.pdf`](../assets/strategic-report.pdf) |

### Affirmed strengths

The draft is structurally sound: precise problem statement, correct binary threshold, verification spectrum that does not pick cryptographic winners, conformance tiers, trust-assumption disclosure for actuarial use, insurance as adoption forcing function, and a correct **verification-vs-validation** boundary (cryptography cannot prove intent).

### Top-priority gaps for WD 1.0

| Priority | Gap |
|----------|-----|
| **Critical** | Static-system assumption breaks for self-modifying / continuously learning AI |
| **Critical** | **Model State Identity (MSID)** — actions must bind to exact model state, not deployment label |
| **Critical** | **"Model Updated" lifecycle stage** — no contemporaneous evidence for who authorized change, what changed, when active |
| **High** | Dynamism classification (assurance must decrease as dynamism increases) |
| **High** | Trust assumption manifest (gate-profile-style structured disclosure) |
| **High** | PoC Evidence Object schema (canonical JSON for assessor interoperability) |

Every recommendation is also a **pre-registered hypothesis** (H-POC1..H-POC10) with fixed success criteria — not moving goalposts. See [`collaboration-pathways/test/HYPOTHESES.md`](../collaboration-pathways/test/HYPOTHESES.md).

**This is feedback, not a product pitch.** Reciprocation is not required; AAIS may read, verify, and optionally run experiments without assembling a return bundle.

---

## 3. Pathways — what it is

**Pathways Architecture v1.1.0** is an orchestration and provenance substrate for AI workflows: composable **Pathway templates**, recorded **PathwayRuns**, step-level audit, **gate profiles** (autonomy and non-delegable acts), **Aqua revision chains** on artifacts, and RIS Appendix C self-provenance discipline.

Normative excerpts in this bundle: [`canon/PATHWAYS_REFERENCE_v1.1.0.md`](../canon/PATHWAYS_REFERENCE_v1.1.0.md) · [`canon/PATHWAYS_ARCHITECTURE_v1.1.0.md`](../canon/PATHWAYS_ARCHITECTURE_v1.1.0.md)

PoC-specific pathways: [`poc-pathways/`](../poc-pathways/pathways-index.yaml)  
Collaboration-bundle technique (canonical encoding): [`collaboration-pathways/canon/COLLABORATION_BUNDLE_TECHNIQUE.md`](../collaboration-pathways/canon/COLLABORATION_BUNDLE_TECHNIQUE.md)

**Core Pathways claim for PoC:** *A Pathway Run is a Proof-of-Control artifact.* Where PoC defines *what* must be proven, Pathways supplies *how* to generate, structure, and attest that proof at the software layer — with honest boundaries (TEE, ZK, consensus timestamps remain complementary).

---

## 4. How Pathways applies to PoC

The Strategic Report maps PoC gaps to Pathways primitives. The highest-leverage alignment:

| PoC provenance question | Pathways primitive | Maturity |
|-------------------------|-------------------|----------|
| Where did context come from? | PathwayRunSource, `primary_source` role | Shipped |
| What model produced output? | `model_provider` / step audit | Shipped |
| What computation ran? | Step audit, prompts, durations, tokens | Shipped |
| Authorization chain | gate_profile → practice_profile → non-delegable approval | Planned (RIS §4.8) |
| Hash-chains / signed manifests | Aqua revision chains; content manifests | Shipped in bundle |

**Provenance as sixth domain:** Pathways is the most architecturally mature existing implementation for this domain. WD 1.0 should reference Pathways provenance as a Reference Implementation rather than inventing parallel machinery.

**Dynamic model updates:** PoC's hardest gap is the **"Model Updated"** stage. Pathways encodes this via [`Model.Update.Apply@v1`](../poc-pathways/pathways/Model.Update.Apply.v1.yaml) — binding `upstream_context_anchor`, authorizer, state before/after, and provenance attestation. That pathway is designed to consume **this bundle** as its context anchor.

**Framework extensions proposed:** [`sidecars/pathways-framework-extensions.md`](../sidecars/pathways-framework-extensions.md) · gap matrix [`poc-pathways/data/poc-gap-matrix.v1.json`](../poc-pathways/data/poc-gap-matrix.v1.json)

---

## 5. The Aquafied collaboration bundle — premise and meta context

### Premise

Delivering PoC feedback as a **collaboration bundle** is not packaging convenience. The format may be exactly the **cryptographic context anchor** PoC Provenance requires for fully provenanced **dynamic** model state and behavior changes.

> **Context is a Pathway. This bundle is the context artifact.**

A collaboration bundle is:

- A **dated, immutable convergence repository** — every byte SHA-256 bound, Ed25519-signed (`verify: OK` offline)
- **Self-describing** — carries its own Pathways encoding of the collaboration technique inside the bundle
- **Aquafied** — 52 Pathways artifacts carry AquaTree sidecars (genesis + did:key signature); index at [`aqua-trees-index.yaml`](../aqua-trees-index.yaml)
- **Provenance-complete** — [`pathway-runs/`](../pathway-runs/index.yaml) ledger maps assets to producing pathways and runs
- **Investigation-ready** — pre-registered hypotheses, proof slots, explicit lifecycle status

**Thesis sidecar:** [`sidecars/cryptographic-context-anchor-thesis.md`](../sidecars/cryptographic-context-anchor-thesis.md)

### Meta context — what is encoded inside the bundle

| Layer | What it is | Where to look |
|-------|------------|---------------|
| **Substance** | PoC feedback PDFs + origin exports | [`assets/`](../assets/origin/ORIGIN_MANIFEST.yaml) |
| **Domain pathways** | Analysis workflows that produced the reports | [`poc-pathways/`](../poc-pathways/pathways-index.yaml) |
| **Technique pathways** | How the bundle itself was assembled, sealed, validated | [`collaboration-pathways/`](../collaboration-pathways/pathways-index.yaml) |
| **Patterns** | Self-describing bundle, cryptographic context anchor, status vocabulary | [`collaboration-pathways/patterns/`](../collaboration-pathways/pathways-index.yaml) |
| **Aqua attestations** | Signed revision chains on every pathway/pattern YAML | [`aqua-trees-index.yaml`](../aqua-trees-index.yaml) |
| **PathwayRuns** | 17 recorded runs (assembly, origin, aqua, context anchor, status vocabulary, executive overview, …) | [`pathway-runs/index.yaml`](../pathway-runs/index.yaml) |
| **Hypotheses** | H-POC-VALIDATE + H-POC1..10 + seal experiments | [`HYPOTHESES.md`](../collaboration-pathways/test/HYPOTHESES.md) |
| **Status vocabulary** | Verify ≠ investigate ≠ policy validate | [`BUNDLE_STATUS_GUIDE.md`](../BUNDLE_STATUS_GUIDE.md) |
| **Validation policy** | Sole signatory: Tricia Wang (mutable via PathwayRun) | [`bundle-validation-policy.v1.yaml`](../collaboration-pathways/meta/bundle-validation-policy.v1.yaml) |
| **Machine index** | Bundle manifest + status meta | [`collaboration-manifest.yaml`](../collaboration-manifest.yaml) · [`bundle-status.v1.yaml`](../collaboration-pathways/meta/bundle-status.v1.yaml) |

**Journey pathway:** [`Experience.Collaboration.OpenInvitation@v1`](../collaboration-pathways/pathways/Experience.Collaboration.OpenInvitation.v1.yaml)  
**Instance playbook:** [`APPLICATION_PLAYBOOK_POC_INSTANCE.md`](../collaboration-pathways/APPLICATION_PLAYBOOK_POC_INSTANCE.md)

---

## 6. What validating this bundle could prove — and how

Validation is **not one binary**. This bundle separates four distinct claims:

```text
VERIFY (anyone)  →  INVESTIGATE (per H-POC*)  →  POLICY VALIDATE (sole signatory)
        ↓                      ↓                            ↓
   bytes intact          hypotheses graded          H-POC-VALIDATE loop closed
        └──────────────────────┴────────────────────────────┘
                    BUNDLE FATE (SEALED | INVALIDATED | INCONCLUSIVE)
```

### A. Verify integrity — binary, anyone

**Command:** `python3 tools/collaboration-bundle/sign_bundle.py verify collaboration/20260602-160000`

| Result | Proves | Does not prove |
|--------|--------|----------------|
| `verify: OK` | No tampering since seal; manifest + signature valid | Claims true, committee acceptance, experiments ran |

**Pathway:** [`Collaboration.Bundle.VerifyBundle@v1`](../collaboration-pathways/pathways/Collaboration.Bundle.VerifyBundle.v1.yaml)

### B. Investigate hypotheses — graded, unilateral

Each H-POC* is a falsifiable experiment with pre-registered success criteria. Running one writes to [`proof-results/`](../proof-results/index.yaml).

| Outcome | Meaning |
|---------|---------|
| `confirmed` | Success criteria met — standalone value even without AAIS reply |
| `falsified` | Structural failure — may drive `INVALIDATED` |
| `inconclusive` | Reciprocal-only or blocked — not failure for invitation bundles |

Examples: H-POC1 (origin SHA-256 integrity), H-POC8 (every deliverable maps to a PathwayRun), H-POC9 (Aqua sidecar on a derivative).

**Pathway:** [`Collaboration.Proof.RunUnilateralExperiment@v1`](../collaboration-pathways/pathways/Collaboration.Proof.RunUnilateralExperiment.v1.yaml)

### C. Policy validation — H-POC-VALIDATE + sole signatory

The **strongest claim** this bundle can support: the collaboration-bundle format works as a **cryptographic context anchor** for dynamic model updates.

| Step | Evidence | Authority |
|------|----------|-----------|
| 1. Anchor | Bundle used as `upstream_context_anchor` | Technical |
| 2. Update | [`Model.Update.Apply@v1`](../poc-pathways/pathways/Model.Update.Apply.v1.yaml) with full provenance | Gate profile |
| 3. Return proof | Artifacts in [`assets/validation-proof/`](../assets/validation-proof/) | PathwayRun |
| 4. Re-seal | `verify: OK` on enlarged bundle | Anyone checks |
| 5. Confirm | Sole signatory attestation | **Tricia Wang only** |

**Today:** loop is `design_pending` — integrity-verifiable, **not yet policy-validated**.

**Pathways:** [`Collaboration.Validation.DynamicUpdateProofLoop@v1`](../collaboration-pathways/pathways/Collaboration.Validation.DynamicUpdateProofLoop.v1.yaml) · [`Collaboration.Validation.SoleSignatoryConfirm@v1`](../collaboration-pathways/pathways/Collaboration.Validation.SoleSignatoryConfirm.v1.yaml)

### D. Bundle fate — aggregate lifecycle

When sufficient unilateral hypotheses confirm (and/or H-POC-VALIDATE closes per policy), [`collaboration_status`](../collaboration-pathways/meta/bundle-status.v1.yaml) may advance to **`SEALED`**. Missing AAIS reciprocation yields **`INCONCLUSIVE`**, never **`INVALIDATED`**.

**Pathway:** [`Collaboration.Investigation.SealOrInvalidateBundle@v1`](../collaboration-pathways/pathways/Collaboration.Investigation.SealOrInvalidateBundle.v1.yaml)

### Authority boundary (transparent)

| Question | Who may assert it |
|----------|-------------------|
| Are the bytes intact? | **Anyone** — `sign_bundle.py verify` |
| Did H-POC*N* succeed? | **Open investigator** — `proof-results/` |
| Is the context-anchor thesis demonstrated? | **Executor + evidence** — H-POC-VALIDATE loop |
| Is the bundle **considered validated** for AAIS? | **Sole signatory only** — Tricia Wang |
| Who may change validation policy? | **Sole signatory via PathwayRun** — non-delegable |

Full matrix: [`BUNDLE_STATUS_GUIDE.md`](../BUNDLE_STATUS_GUIDE.md)

---

## 7. Recommended reading order

1. **This overview** (you are here)
2. [`Dynamic_Update_Provenance_Brief.md`](Dynamic_Update_Provenance_Brief.md) — model update provenance, PoC scenarios, dynamism (plain English; technical sidebars optional)
3. [`BUNDLE_STATUS_GUIDE.md`](../BUNDLE_STATUS_GUIDE.md) — if any status label is unclear
4. [`PoC_Initiative.md`](PoC_Initiative.md) · [`PoC_Standard_v0.1.md`](PoC_Standard_v0.1.md)
5. [`assets/comprehensive-analysis-v0.1.0.pdf`](../assets/comprehensive-analysis-v0.1.0.pdf) · [`assets/strategic-report.pdf`](../assets/strategic-report.pdf)
6. [`HYPOTHESES.md`](../collaboration-pathways/test/HYPOTHESES.md) — start with **H-POC-VALIDATE**
7. [`collaboration-spine.md`](../collaboration-spine.md) · [`companion-bundle-index.md`](../companion-bundle-index.md)

---

## 8. Closing frame

PoC defines **what** must be proven about AI control at scale. Pathways offers **how** to generate and attest that proof — especially for Provenance and dynamic model updates. This **Aquafied collaboration bundle** is simultaneously:

1. **Feedback** on PoC WD architecture v0.1.0  
2. **Evidence** of how that feedback was produced (origin + PathwayRuns)  
3. **A candidate primitive** for PoC's hardest open problem — binding collaborative context before model state changes  
4. **A falsifiable experiment** (H-POC-VALIDATE) whose success would validate the format by returning proof into the bundle itself  

The committee may accept, challenge, or ignore any layer independently — because verify, investigate, and policy validation are **separately checkable** with **separately documented authority**.

---

*Encoded by [`Collaboration.Meta.AuthorExecutiveOverview@v1`](../collaboration-pathways/pathways/Collaboration.Meta.AuthorExecutiveOverview.v1.yaml) · PathwayRun `pr_poc_executive_overview` · Channel `djat-poc-20260602`*