> ## Documentation Index
> Fetch the complete documentation index at: https://docs.toorow.com/llms.txt
> Use this file to discover all available pages before exploring further.
# Renders, Dossiers and Sharing
> How an answer in toorow becomes a frozen figure, an addressable document, and a revocable link a colleague can open.
An analysis is only finished when someone else can read it. **toorow** has one
governed path from a question to a page a colleague opens, and every step of it
is addressable, versioned and revocable.
```
Query Spec -> Result -> Visualization -> Render -> Dossier version -> Share
```
Nothing on this path is a screenshot, and nothing is a live re-query. A reader
opening a link sees the exact figure that was drawn, from the exact data that
was executed, with the runtime build that drew it.
***
## 1. The objects, and what each one owns
| Object | What it is | What it pins |
| ------------------ | -------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **Query Spec** | A governed question — measures, dimensions, filters, grain, window — with immutable versions | The Semantic View and the governed names it asks for |
| **Result** | One execution of a Query Spec version. Immutable. | The data, the definitions, the quality findings, the provenance, the freshness, and the reasoning path that produced it |
| **Chart Template** | A reusable, validated presentation starting point with immutable versions | Compatibility predicates and optional governed defaults — never data |
| **Visualization** | A saved presentation for a compatible Result shape | Visual type, bindings, presentation intent |
| **Render** | An immutable presentation snapshot | The exact Result, the exact Visualization Spec version, the renderer build, the theme, the formatters, the responsive profile and the evidence manifest |
| **Dossier** | An ordered composition of Renders and narrative blocks, immutable per version | The exact Render identities it shows, and the author of every narrative block |
| **Share** | A revocable grant of access to exactly one target | Either one Render **or** one Dossier version — never both, and never *the latest* |
A Render is a snapshot, not a rerun instruction. A Share never follows the
latest anything: what you granted is what the reader sees, until you revoke it.
***
## 2. Where this lives in the console
Everything on this path is reached from **Analyze**.
* **Analyze › Results** — the executions, each opening on its data, its
definitions, its quality and its reasoning path.
* **Analyze › Renders** — the frozen figures, plus the **Dossiers** panel. A
project with no Dossier yet says so, and names the gesture that creates one.
* **Reports › Templates** — the Chart Template catalog, and its workbench, where
a template is edited, versioned, archived and restored. An archived version
stays readable, and a pin onto an archived version is named as such rather
than silently broken.
The console, the MCP App surface and the public share page all draw through the
**same rendering runtime**. They may choose different responsive profiles; they
may not change analytical meaning. A figure that reads one way in the console
cannot read another way in the link you sent.
***
## 3. Composing a Dossier from an AI host
A model working over MCP does not hand you a picture — it keeps what it drew.
`compose_dossier` freezes one immutable Render per figure and writes the Dossier
version in a single transaction.
```python theme={null}
compose_dossier(
project_id: str,
label: str,
blocks: list[dict], # render blocks and narrative blocks, in reading order
description: str = "",
dossier_id: str = "" # omit to create; pass to append a new version
) -> dict
```
Three properties are worth knowing before you call it:
* **All or nothing.** A block referencing a figure from another project, or from
another organization, fails the whole call. There is no partial Dossier.
* **Authorship is recorded, not inferred.** Every narrative block carries
`authored_by`. The MCP gate stamps `model` on every block it stores —
*whatever the caller sent* — because a figure is governed and a narrative is
generated, and the page has to show which is which. `human` is what a person
writing in the console leaves.
* **A version is immutable.** Appending to a Dossier creates a new version. The
version someone is reading does not change under them.
Each figure keeps its reasoning path, resolved at read time, so a reader can ask
of any chart in the document: *which question produced this, against which
definitions?*
***
## 4. Sharing: two people, one link
Nothing leaves the platform on one person's signature.
1. **Request.** Someone with the project capability requests a Share, from the
Sharing tab of the Render Workbench or from the Dossier. The response
deliberately carries **no link**.
2. **Confirm.** A *second* role holder authorizes the exit and receives the only
link. The confirmation is stamped: who confirmed, and when. The requester and
the confirmer cannot be the same person — the database refuses it.
3. **Read.** The recipient opens the link. The share page draws every figure
through the same runtime as the console, with each figure's provenance stated
beside it, and states plainly when a fact is absent rather than omitting it.
4. **Revoke.** Revocation is a single act, available at any time, and it does not
need a second holder. A revoked link stops working.
The grant freezes every Render in the pinned version at the moment of
confirmation. If one of them cannot be frozen, the request is refused and the
offending figure is named — you never get a link that half works.
**PDF export is not available yet.** A Dossier version is shareable as a page
today. Export to PDF is a stated target for this surface and is not built: it
will render through the same path as the share page, so the document on paper
shows exactly what the link shows.
***
## 5. Evidence travels with the figure
Every Render carries an evidence manifest, and it is not decorative. A reader —
person or agent — can go from a drawn point back to:
* the **Result** that produced it, and the Query Spec version that asked for it;
* the **definitions** used, at the Semantic View version that was in force;
* the **quality findings** that applied to the underlying data;
* the **freshness** and the pull provenance of each source;
* the **reasoning path**, when a model produced the answer, or an explicit
*No AI path* when none exists.
Feedback attaches to that same anchor: `submit_analyze_feedback` records a rating
against the exact observation a reader was looking at, not against the page as a
whole.
***
## Next Steps & Cross-References
The MCP tools that execute, render and compose along this path.
The governed measures and dimensions a Query Spec is allowed to ask for.
The findings a Result carries with it.
Connect an AI host and compose a Dossier from a conversation.