Skip to content

methods.md

Latest commit

 

History

History
89 lines (72 loc) · 5.15 KB

methods.md

File metadata and controls

89 lines (72 loc) · 5.15 KB
Raw

Methods and evidence boundaries

Choose a workflow

Browse the methods hub or the reviewed integration catalog. uv run symposium organize methods prints the same catalog locally; --format json supplies machine-readable records. Each handoff states its purpose and limits and pins public repository evidence to a reviewed commit. Only the native Template adapter is an installed cross-repository render integration. Other links direct separate workflows, without automatic data transfer, compatibility claims or scientific validation.

Use credentials to select optional services and discovery to understand what is indexed. Catalog changes follow the refresh procedure there.

Approved public content

Public dossiers and participant briefs are deterministic derivatives of an approved organizer snapshot. Presentations retain supplied abstracts and attribution; participant briefs retain affirmative-consent self-reports and supplied resources. They do not perform external identity verification or infer competence, rankings, collaborations or biographies. Missing abstracts and unconfirmed session times must remain visibly pending. Only explicit sessions produce calendar entries.

The current approved snapshot lives in data/public/2026/. The separate data/inputs/aif_2026/ research scaffold may remain empty. Validate the former with organize check-public and the latter with edition validate; neither substitutes for the other. Structural validity does not establish truth, current consent outside the recorded source, source freshness, or event readiness.

Next intake and change review

  1. Pull a fresh source snapshot into a private workspace outside the public checkout. Verify mapping/schema and preserve raw responses and transport evidence privately.
  2. Project only allowlisted public fields. Review affirmative consent, identity ambiguities, missing text and exclusion warnings. Do not reuse historical consent.
  3. Create dossiers, participant briefs and explicit-session calendar artifacts. Stage and render the native book separately; inspect every final PDF/DOCX surface.
  4. Bind reviewed artifacts to the candidate bundle and stage a fresh public tree.
  5. Verify integrity, inspect content readiness and compare the candidate with the previous approved tree before an organizer decides whether to integrate it.
uv run symposium organize readiness --public-dir "$CANDIDATE_PUBLIC_DIR"
uv run symposium organize readiness --public-dir "$CANDIDATE_PUBLIC_DIR" --require-complete
uv run symposium organize review-update --previous-public-dir data/public/2026 \
  --candidate-public-dir "$CANDIDATE_PUBLIC_DIR" \
  --output "$SYMPOSIUM_PRIVATE_DIR/update-review.json" --public-root .

Set both directory variables to organizer-owned locations outside the public checkout. Normal readiness reports pending content without failing; strict readiness returns nonzero for incomplete content. Integrity errors fail both modes. Readiness reports source_freshness: not_checked: hashes do not establish whether the source has changed since capture.

Update review verifies both public trees and records additions, removals, changed fields and artifact differences. The report is private and refuses an existing output. It does not promote a candidate, delete a record, identify renamed people, or infer that a removal means revoked consent. Resolve removals and ambiguous identity changes with organizers. A confirmed withdrawal requires an explicit publication decision; earlier downloads and publication history may persist.

Optional research analysis

The analyze and generate commands use external LLM providers and can incur charges. Profiles, methods, curricula, summaries and proposals are unverified drafts. Public-sharing consent is not itself authorization for every third-party processing purpose. Verify authorization and minimize transmitted data before making requests. Keep drafts private and review material claims against primary sources, including identity, provenance, quotations and dates. No generated draft is automatically published by the organizer pipeline.

CSV schema and preflight cover structural input checks; the API guide covers actual configuration and error handling. Prompt requests for citations or comprehensive research are not guarantees of accuracy.

Artifacts and interpretation

A manifest or artifact index establishes byte/bundle bindings, not semantic or visual quality. Native rendering uses the separate Template Python 3.14 pipeline; a render exit status and extracted text do not replace page-by-page visual review. Website verification needs actual browser, keyboard, responsive and download checks.

Text visualization is exploratory corpus analysis. Retain input hashes/order, options, lockfile, actual method, warnings and outputs. Distances and clusters describe a representation of text, not validated scientific relationships or assessments of people. No publication, test result, rendering or visualization in this pipeline independently validates scientific claims.