Evidence, workbench and sharing
One evidence model, rendered honestly in four formats.
The fixed evidence pack renders only the sections you actually have evidence for, and states plainly, on every page, that this is research and paper mode with no live order routing. The Report Workbench builds a different kind of document from the same evidence, tracks whether a figure has gone stale, and never recomputes a number behind your back.
- Export formats
- 4PDF, HTML, DOCX and JSON, all from the same evidence model
- Fixed PDF sections
- 8Each renders only when its evidence exists
- Workbench block kinds
- 8table, keyvalue, matrix, metrics, bullets, text, formula, image
01What you actually do here
Five things, in the order the page presents them.
- 01
Assemble the evidence pack
POST /api/report builds the fixed model from what the project actually computed. Universe and data, absolute risk and return, benchmark-relative risk, portfolio construction, the model, out-of-sample validation, the strategy backtest, and method and limitations — each section renders only when the corresponding evidence exists.
- 02
Read the universe and risk pages
Page one carries the universe table, per-ticker risk and return statistics, benchmark-relative risk and the portfolio weights actually used. An undefined statistic renders as an em dash, never as 0 or NaN — reporting the worst possible value for an undefined quantity is the more damaging of the two errors.
- 03
Read the model and validation pages
Page two states the model, the fold count, mean and standard deviation of the primary metric, the worst fold, and the embargo rows removed, then the strategy backtest against buy-and-hold. The footer on every page reads: research / paper mode - live order routing disabled.
- 04
Build a different document in the Workbench
The evidence registry — id, module, type, category, title, description, tags, route, source, build — feeds a Report made of Sections made of Blocks. Each block carries a fingerprint marked current, stale or unavailable; refresh is manual, and the server never recomputes a figure underneath the author.
- 05
Save, share and audit
Save, duplicate, rename or delete a report; export the Workbench view with the browser's own Print / PDF dialog. Snapshots keep a project's saved results and notes; a share link's token, scope and permission are a stored fact, and revoking it kills the link. Every action lands in the audit log.
02How it works
The method, stated for the specialist.
The fixed pack and the Workbench draw from the same evidence, but they are not the same guarantee: one is a rendered snapshot of what ran, the other is an editable document that tracks whether its sources have moved since it was assembled.
- Rendered, not typed twice
- HTML, DOCX and the JSON payload all come from the same evidence model as the PDF: the Word document carries the same sections and figures as the printed pack, and the JSON payload is that same model returned as data for downstream use.
- A section that has no evidence does not appear
- The report always reflects the work actually run. A missing validation section means validation was not run, not that the page renders a placeholder.
- The no-live-order-routing statement is verbatim, twice
- It appears in the footer of every page and again in the Method & Limitations section: research and paper mode only, this workspace performs no live order routing.
- Fingerprint staleness, not silent drift
- A block's fingerprint reads current when it still matches its source, stale when the source was re-run with different results, and unavailable when the source was never loaded this session. A stale block carries an amber note in both the Workbench and any export — the number is not swapped underneath the author.
- Snapshots are frozen, not editable
- A saved result captures rendered HTML, an optional image, the parameters in force at capture, the underlying figures and provenance. Reopening one shows that frozen record; it does not repopulate a page's controls or re-run anything, and a hard server-side guard refuses any executable content in stored HTML.
- A share link is a database row
- A random token, a scope of project or page, and a permission of view, comment or edit. Revoking the record kills the link immediately. Enforcing who is allowed to reach that link at the account level is the deployment's front door, not something this store claims to do itself.
- Experiment comparison sorts by Deflated Sharpe
- Never by raw Sharpe — sorting by raw Sharpe is how the most over-searched configuration wins. The raw Sharpe still renders directly above the deflated figure so the gap between them is visible.
03What you get out
Outputs, and what they let you defend.
A report is a rendering of evidence that already exists elsewhere in the project — nothing in the export path invents a number.
- A fixed evidence pack in PDF, HTML, DOCX and JSON from one underlying model
- A Workbench report built from evidence blocks with per-block staleness state
- Saved snapshots with parameters, figures and provenance, immutable in substance
- A revocable share link scoped to a project or page with a stated permission
- An audit-log entry for every report, snapshot and share action taken
04What this does not do
Stated plainly, because it matters as much as what it does.
Workbench export uses the browser's print dialog
Print / PDF from the Workbench is not server-rendered. Only the fixed /api/report pack produces a server-rendered PDF.
No CSV or PNG export route for reports
Reports export as PDF, HTML, DOCX or JSON. There is no separate CSV or PNG export path.
Snapshots do not restore application state
A saved result is frozen markup and figures. Opening one does not repopulate the page's controls or re-run the computation that produced it.
A share link does not enforce account access on its own
The token, scope and permission are a stored fact. Whether the person opening the link is who they claim to be is the deployment's authentication layer, not the share-link store.
Turn a project's results into a document someone can check.
Beta researchers export the fixed evidence pack from a real run, then build a Workbench report around the figures they want to present.