# LCP Experiment Designs — Legal Commons and Multi-Context Convergence

**Bundle channel:** djat-poc-20260602  
**Emerged in:** this PoC collaboration context  
**Registry:** [`lcp-pathways/data/lcp-experiment-registry.v1.json`](../lcp-pathways/data/lcp-experiment-registry.v1.json)  
**Pathways package:** [`lcp-pathways/`](../lcp-pathways/)  
**Cross-register:** [`convergence-poc-lcp-mikeoss-cross-register.md`](convergence-poc-lcp-mikeoss-cross-register.md)

---

## Purpose

This sidecar pre-registers **LCP-forward experiments** that are compelling for insurers, neutral dispute providers, and Pathways implementers. They extend the convergence analysis (bundle × LCP × PoC) into **executable hypotheses** with Pathway templates, not prose alone.

Design principle (P0, from convergence sidecar): **never merge** `bundle_root_hash`, LCP `atrHash`, and behavior/run-manifest hashes. Link them with explicit cross-reference fields; verify each domain independently.

---

## Experiment suite overview

| ID | Hypothesis | Pathway | Status |
|----|------------|---------|--------|
| EXP-LCP-1 | H-LCP-1 | `Legal.LCP.TripleHashEvidenceSandwich@v1` | design_pending |
| EXP-LCP-2 | H-LCP-2 | `Legal.LCP.InfuseLegalContext@v1` | design_pending |
| EXP-LCP-3 | H-LCP-3 | `Legal.LCP.ConvergeMultiContext@v1` | design_pending |
| EXP-LCP-4 | H-LCP-4 | `Legal.Commons.LongitudinalOutcomeLearn@v1` | design_pending |
| EXP-LCP-5 | H-LCP-5 | `Legal.Commons.LongitudinalOutcomeLearn@v1` | design_pending |
| EXP-CONVERGE-MO-LCP-1 | H-CONVERGE-MO-LCP-1 | `Legal.LCP.ConvergeMultiContext@v1` | design_pending |

---

## EXP-LCP-1 / H-LCP-1 — Triple-hash evidence sandwich

**Demonstration:** One agentic commerce or dispute-handling transaction carries three hashes in a small cross-reference JSON:

```json
{
  "legal_context_ref": {
    "atr_hash": "<LCP terms artifact>",
    "lcp_level": 2,
    "terms_url": "https://example.org/terms.json"
  },
  "context_anchor_ref": {
    "bundle_root_hash": "<this or upstream collaboration bundle>",
    "channel": "djat-poc-20260602"
  },
  "behavior_evidence_ref": {
    "run_manifest_sha256": "<AttestedRun or model_update_provenance>"
  }
}
```

**Success:** Each hash verifies offline in its own toolchain; no unified “super-hash” claim.

**Why it matters:** Insurers and AAA-style neutrals can audit “terms + context + behavior” without forcing one protocol to swallow the others.

---

## EXP-LCP-2 / H-LCP-2 — LCP discoverability infusion

**Demonstration:** Add optional `legal_context` block to `collaboration-manifest.yaml` via `Legal.LCP.InfuseLegalContext@v1`:

- Discovery URL (`.well-known/legal-context.json` pattern)
- Terms URL + `atrHash`
- `lcp_level` (1–4 per LCP draft)

**Success:** Agent or assessor resolves terms before acting; bundle seal unchanged; cross-ref documented in sidecar.

**PoC link:** Scenario B (licensed fine-tune) and Scenario H (policy gate) in Dynamic Update Provenance Brief.

---

## EXP-LCP-3 / H-LCP-3 — Authorization gate meets LCP level

**Demonstration:** Map PoC **Authorization** domain `gate_profile` rows in [`poc-gap-matrix.v1.json`](../poc-pathways/data/poc-gap-matrix.v1.json) to LCP trust levels:

| LCP level | PoC gate_profile analogue | Non-delegable acts |
|-----------|---------------------------|-------------------|
| 1 informational | A0 advisory | none |
| 2 provable | A1 attest-only | terms publication |
| 3 signed | A2 constrained execute | outcome recording |
| 4 integrated | A3+ with human break-glass | dispute escalation |

**Success:** Convergence matrix row in updated sidecar; no normative claim that PoC subsumes LCP.

---

## EXP-LCP-4 / H-LCP-4 — Mediation process as LCP terms artifact (flagship thread)

**Narrative:** Traditional mediation and arbitration playbooks are PDFs and institutional habit. This experiment treats a **context-specific process** as a **hashable LCP terms JSON**:

- Phases (intake, caucus, settlement window, escalation to arbitration)
- Neutral role and evidence hooks
- Agentic handoff points (where an AI mediator assistant may act vs. must defer)
- Fork lineage: contributors publish variants; `atrHash` identifies which process version governed a cohort

**Pathway:** `Legal.Commons.LongitudinalOutcomeLearn@v1` phase `publish_process_terms`.

**Outcome signals (not AI quality metrics):**

- `settlement_reached` (boolean)
- `days_to_resolution`
- `terms_compliance_score` (post-settlement adherence to agreed terms)
- `dispute_category` (employment, consumer, IP, cross-border, healthcare, agentic_commerce)

**Privacy:** Cohort statistics only; no party PII in commons aggregate.

**Success (design phase):** Example process JSON + placeholder `atrHash` + PathwayRun design record.  
**Success (run phase):** At least one synthetic cohort row in commons registry keyed by `process_atr_hash`.

---

## EXP-LCP-5 / H-LCP-5 — Cross-category longitudinal learning (legal commons)

**Narrative:** Employment, consumer, and IP disputes share **structural** process features (cooling-off periods, mandatory disclosure steps) but differ in **substantive** terms. The legal commons learns **which structural variants correlate with faster fair settlement** across categories without merging confidential case facts.

**Mechanism:**

1. Each category publishes process templates (EXP-LCP-4) with shared Term registry entries (`Term.LegalCommons@v1`).
2. PathwayRuns record outcome signals per session.
3. Aggregate layer computes cohort metrics by `(process_atr_hash, category, time_window)`.
4. Fork proposals cite parent template PathwayRun + LCP atrHash lineage (marketplace-style attribution).

**Continuous collaborative refinement:** Contributors iterate process variants in the open; provenance shows who proposed what and under which hypothesis. This is **longitudinal** and **cross-category** learning of **legal best practices for social process**, not model weights.

**Success:** Registry shows ≥2 categories pointing at distinct `process_atr_hash` values with shared structural Term IDs; fork lineage documented in PathwayRun ledger.

---

## Relationship to MikeOSS and PoC downstream work

| Downstream context | What convergence needs | Hypotheses |
|--------------------|------------------------|------------|
| **MikeOSS** (`djat-mikeoss-20260528`) | Attested legal research runs, citation assay, run manifests | H-MO1, H-MO2, H-MO3, H-MO6, H-MO8 |
| **PoC** (this bundle) | Context anchor, Authorization/gate_profile, dynamic update loop | H-POC-VALIDATE, H-POC6, H-POC-REV-* |
| **LCP** (infused here) | Terms discoverability, dispute-layer binding | H-LCP-2, H-LCP-4 |

**Cross-registered experiment:** H-CONVERGE-MO-LCP-1 — see dedicated sidecar. **Tri-context extension:** H-CONVERGE-TRI-1 — [`convergence-tri-context-weave-poc-mikeoss.md`](convergence-tri-context-weave-poc-mikeoss.md).

---

## Execution order (recommended)

1. `Legal.LCP.CrossRegisterUpstreamBundle@v1` (MikeOSS anchor)
2. `Legal.LCP.TripleHashEvidenceSandwich@v1` (minimal demo)
3. `Legal.LCP.InfuseLegalContext@v1` (manifest block)
4. `Legal.Commons.LongitudinalOutcomeLearn@v1` (flagship design or pilot)
5. `Legal.LCP.ConvergeMultiContext@v1` (merge dimensions when downstream proofs exist)
6. Re-seal bundle; update `proof-results/index.yaml`

---

## References

- [`lcp-pathways/APPLICATION_PLAYBOOK_LCP.md`](../lcp-pathways/APPLICATION_PLAYBOOK_LCP.md)
- [`sidecars/convergence-analysis-bundle-x-lcp.md`](convergence-analysis-bundle-x-lcp.md)
- [`sidecars/context-analysis-legal-context-protocol.md`](context-analysis-legal-context-protocol.md)
- [`collaboration-pathways/test/HYPOTHESES.md`](../collaboration-pathways/test/HYPOTHESES.md) — H-LCP-* and H-CONVERGE-MO-LCP-1

*Encoded by pathway run `pr_lcp_experiment_designs` · emerged in djat-poc-20260602*