# Week 7 proposal: spillovers and the actual treatment unit

**Length:** two pages plus an exposure map.  
**Name / setting / date:** …  
**Status:** research proposal; label illustrative synthetic numbers.

## 1. Question and four ledgers

Define membership, workflow, interaction and outcome boundaries. Distinguish planned rosters from realized attendance and a unique deliverable from member-level copies. Draw directed edges with the convention **source $j$ affects receiver $i$**. Give timing, overlap weights, the no-peer rule and links that cross nominal team boundaries.

## 2. Intervention, support and estimand

Name actual assignment units and stages, eligible allocations, saturation/take-up and the baseline population. Define the complete exposure state including own treatment and peers. List state probabilities and the population with positive support for each contrast. Distinguish recipient, spillover and policy targets; state when a recipient comparison changes peers too. Explain what a treatment-induced graph change would mean for the proposed target.

| Exposure state | Own assignment | Peer exposure / source identity | Timing | Probability / support | Proposed comparison |
|---|---|---|---|---|---|
| … | … | … | … | … | … |

## 3. Two mechanisms and a discriminating prediction

| Mechanism | Observable prediction | Counterprediction / possible falsifier | Additional information |
|---|---|---|---|
| Mechanism 1: … | … | … | … |
| Mechanism 2: … | … | … | … |

Distinguish complementarity, pressure, learning, shared demand and selection. Explain why correlated output or more messages would not alone establish the favored mechanism.

## 4. Outcomes, costs and missingness

Define unique joint output, individual contributions, independent quality, sender opportunity costs, recipient benefits, person-time and worker welfare. Include exposure/observation coverage and missing interactions or outcomes. State how unlike output units would be valued and how duplicated deliverables would be prevented. Preserve eligible nonusers and absent workers in the appropriate assignment ledger.

## 5. Identification, inference and limits

State the counterfactual comparison and its assumptions. Identify assignment/dependence units, cross-team leakage, the null or estimand being tested, and limits from few groups or unsupported exposures. A clustering correction alone does not justify causal peer effects. If proposing contribution scores, specify independent construction and held-out validation. Give one justified conclusion and one mechanism, portability or welfare claim the design cannot support.

**Data and reproducibility note:** record accessible versus proposed inputs, identifiers/join keys, timestamps, field definitions and permissions. Distinguish executed synthetic code from future empirical work. Record command, seed and modifications. No provider contact or restricted-data application is required.