# Create an optional private evidence-garden site

## Decide whether a site is useful

A website is not required to use the evidence-garden method. Markdown files, a spreadsheet or a notes system may be enough.

A private site becomes useful when the collection needs:

- several ways to browse the same records;
- full-text search;
- clear links between episodes and sources;
- stable URLs or reference handles;
- selective access for trusted reviewers;
- a connected ChatGPT plugin;
- a calmer review experience than reading a directory of files.

## Keep the site private by design

Professional evidence may include client work, personal information and details that are accurate but unsuitable for publication.

Use access controls provided by the hosting platform. `noindex` metadata can discourage search-engine indexing but is not access control.

Before inviting anyone, check:

- which account owns the site;
- whether invited viewers can only read;
- whether the site exposes downloads or original documents;
- whether search results can reveal confidential text;
- whether the plugin applies a separate owner check to write operations;
- whether screenshots or previews created by the platform contain sensitive information.

## Suggested information architecture

### Front page

Keep the overview compact. Useful elements include:

- a short explanation of the collection;
- a prominent “start with a question” section;
- search;
- topic cards;
- recently added or currently reviewed records.

Stable IDs do not need to appear everywhere. They are most useful on record pages and in copy-reference actions.

### Record page

Display the core fields in a scannable list:

- **Contribution**
- **Context**
- **The problem**
- **How the work connected**
- **Outputs**
- **Possible capability links**

Additional extracted detail can remain collapsed. Sources should be visible without overwhelming the main account.

### Source register

List each source with its identifier, title and a short explanation of what it adds. Do not assume the site needs to provide the original files to every viewer.

### Recovered leads

Keep a separate area for incomplete activities and possible episodes. This prevents the system from forcing every fragment into a confident finished record.

### Question-led views

Questions can make the collection more useful than a conventional portfolio. Examples include:

- How has this person turned an ambiguous problem into workable research?
- Where have they helped a team use unfamiliar knowledge?
- What evidence exists for commercial judgement?
- How have they worked across disciplinary or organisational boundaries?

A question retrieves relevant episodes. It should not create a new claim by itself.

## Build process using ChatGPT Sites

The exact interface may change, but the working sequence is:

1. Prepare a small reviewed dataset and source register.
2. Ask ChatGPT to create a private site for browsing and reviewing the collection.
3. Specify the record fields, filters, stable-ID rules and confidentiality boundaries.
4. Review the first version on both desktop and mobile.
5. Correct information architecture and wording before importing more material.
6. Add persistent storage if records will be created after deployment.
7. Add a small MCP toolset if ChatGPT needs to add, retrieve or search records.
8. Invite named viewers only after the access policy has been checked.
9. Keep the site private unless individual records have been separately reviewed for publication.

## Brief for creating the site

> Create a private evidence-garden site for reviewing structured records of professional work. This is an internal evidence system, not a public portfolio.
>
> The site should support search, topic-based browsing, question-led views, individual record pages, a source register and a separate area for incomplete leads. Each episode and source has a stable identifier. Hide episode identifiers from the main overview when they add clutter, but display them prominently on individual records with a copy-reference action.
>
> Present each episode as a scannable list with bold labels for contribution, context, the problem, how the work connected, outputs and possible capability links. Keep additional extraction detail collapsible. Make uncertainty, draft status and source provenance visible.
>
> The design should feel like a calm research collection rather than a performance dashboard. It must work well on mobile. Use real private access controls and prevent public indexing. Do not expose original confidential documents unless that is explicitly requested.
>
> Start with the smallest useful version and preserve the record identifiers exactly. Do not infer missing outcomes or rewrite cautious evidence as promotional copy.

## Publishing a public project page without exposing the garden

The private system and its public case study should remain separate.

For a public portfolio page:

- use cropped or redacted screenshots;
- replace confidential records with a fictional demonstration record where necessary;
- describe the information architecture and method rather than publishing the evidence itself;
- avoid showing source filenames, private links, client quotations or access details;
- explain that the live site remains private;
- provide generic downloadable guides that contain no personal archive data.

The project can therefore demonstrate systems thinking, interaction design and AI-supported building without turning the underlying career archive into public material.