# Adjacent’s AI Disclosure Policy

## Purpose

The following AI Disclosure Policy exists to provide transparency and insight into the role AI plays in Adjacent’s content development processes. We publish a wide assortment of content including prediction market-driven news, governance, and methodology documentation. The following explains our guiding principles when it comes to employing AI to assist with content creation. This policy is informed by our intent to optimize for transparency and to maximize customer value.

## Principles

- **Transparency** — Transparency and trust are critical as a neutral third-party index provider. It is especially critical for writing rules-based index and reference rate methodologies that are reliable, reproducible, and robust. We believe it is important to make clear the roles AI plays and does not play in our methodology creation and publication processes.
- **Human-to-human** — There are categories of content we create that are optimized for human-to-human interactions and experiences. These include our prediction market-driven news, weekly look ahead articles, and social media posts. This content will always be written by humans, for humans. AI may be used as a supplemental tool to assist throughout the editorial process.
- **Thematic methodologies** — We build thematic event-contract indices that make themes measurable, reproducible, and tradable. They provide actionable insights and signals that both humans and agents can act upon. Breadth is critical when it comes to thematic indices. The more breadth, the more flexibility one has to express a particular view. We therefore use AI to accelerate methodology and index development, all in an effort to ship more value to our customers.
- **Index ideas and concepts** — Humans are ultimately behind the thematic ideas and concepts that drive our event-contract index products. They are best equipped to relate to and understand what thematic index products other humans will find valuable. Agents assist in taking human-driven thematics and creating specifications that help realize those thematic exposures.
- **Templated and repeatable tasks** — Agents are exceptionally good at executing against templated, rules-based, and repeatable tasks. Our index methodologies are templated documents. We’ve developed a methodology construction kit that enables agents to rapidly generate methodologies after the initial thematic concepts and index specifications have been identified (e.g., return type, constituent universe, weighting, reconstitution and rebalancing schedules, etc.). Further, agents are particularly good at identifying inconspicuous edge cases and gaps that humans tend to miss. They help mitigate those risks when used to generate our methodology documentation.
- **Never one-shotted** — Agents help generate index methodologies we publish using our rules-based construction kit and methodology building blocks. However, they do so with humans in the loop. Methodology documents are never generated with single “one-shotted” prompts. The process is highly iterative. Humans are intimately involved in methodology construction and refinement as well as attesting to publication readiness.
- **Governance** — The Adjacent Index Committee (AIC) is responsible for the governance of Adjacent index and reference rate products. Agents are used to augment AIC governance processes and help support with documentation, scheduling AIC reviews, and constituent management. However, human discretion drives AIC governance decision-making.

## Labels and Definitions

We have the following labels that can be applied to content artifacts in order to provide transparency into the role AI played in their creation.

| Label | Meaning | Example uses |
| --- | --- | --- |
| `Human-written` | Humans authored the piece for other humans to consume. AI may have been used in the editorial process (e.g., spellcheck, formatting, etc), but the ideas, concepts, and voice originated from humans with the intent to share with other humans. | Commentary; Opinion pieces; Daily prediction market-driven news; Weekly roundups; SEC and CFTC request for comment submissions |
| `Agent-assembled` | Produced primarily by agents using Adjacent rules-based methodology construction kits and building block templates. Optimized for reproducibility, efficiency, reliability, and robustness. Not a claim of zero human involvement. | Draft methodologies; Draft methodology edits; Templated specifications and construction kits; Governance documentation (e.g., decision register, corrections and restatements, constituent update notices, etc.) |
| `Human-attested` | Originated as an agent-assembled artifact that involved multiple cycles of human review and iteration. A human ultimately attests to and signs off on the artifact’s publication upon determining the artifact fits its stated purpose and satisfies its objectives. | Published methodologies; Methodologies for products that may be traded; Artifacts going from agent-assembled draft to “official” |

## Methodology Creation Workflow

The following provides a high-level summary of our index methodology creation process including the roles humans and agents play:

| # | Step | Description | Label |
| --- | --- | --- | --- |
| 1 | **Theme generation** | The Adjacent team identifies a theme for a potential event-contract index worth developing. | `Human-written` |
| 2 | **Index type selection** | Given the theme, the Adjacent team identifies the characteristics the index should have so as to best express the theme. E.g., total return, price return, weights, rebalancing/reconstitution schedules, universe selection, constituent eligibility, etc. | `Human-written` |
| 3 | **Draft methodology** | After the index type and the attributes best-suited to express the theme have been selected, we task a team of agents to draft a methodology using our internal methodology construction kit. It is comprised of the building blocks necessary to build an event-contract index. | `Agent-assembled` |
| 4 | **Backtesting** | Backtesting is done in order to evaluate how the draft methodology’s proposed set of rules performs using historical data. | `Agent-assembled` |
| 5 | **Review and iterate** | Upon receiving the initial draft and backtest data, an Adjacent team member reviews the methodology in depth and iterates upon it. This involves a back and forth dialogue in which an Adjacent team member is prompting the agent to make edits to the draft. Additional backtesting is done following any changes. | `Agent-assembled` |
| 6 | **Publish** | Adjacent’s AIC finalizes and publishes the methodology after multiple cycles of human review and iteration. In doing so, Adjacent attests to the methodology’s readiness when it comes to satisfying its stated purpose. | `Human-attested` |

## Adjacent’s Perspective

We share this policy to provide as much transparency as possible. We at Adjacent use AI detection tools like Pangram to review our content as well as the content of others. We think they provide useful data points, especially when content is meant to evoke human interactions and experiences such as our daily prediction market-driven news and weekly look ahead articles. We also believe generative writings are particularly useful and well-suited for certain types of content, such as templated, rules-based index methodologies. That said, we acknowledge that AI detectors today do have false positives and are not perfect. It is for those reasons and nuances why this AI Disclosure Policy exists.