frontend-patterns/workplans/FEP-WP-0005-research-bench-setup.md
repo-manager 4774e0898a repo.work.assign_missing_identifiers
source: repo-manager
reason: deterministic projection registration

Assistant: codex
Assistant-Model: gpt-6-astra
Assistant-Session: 01a0cffe-206f-7323-89d3-b20ca4cf4bf9
2026-09-24 09:47:23 +02:00

74 lines
2.6 KiB
Markdown

---
id: FEP-WP-0005
type: workplan
title: "Set up the frontend research bench"
domain: infotech
repo: frontend-patterns
status: finished
owner: codex
topic_slug: canon
created: "2026-09-24"
updated: "2026-09-24"
state_hub_workstream_id: "23ce6971-fdd2-50d9-b7f5-35a5fc80354d"
---
# Set up the frontend research bench
Implement the user's accepted infospace-bench integration inside this repository.
The research corpus stays repository-owned; bench executes inspectable source
investigations. Live research generation and canonical promotion are outside setup.
## Install the domain workflow and intake adapter
```task
id: FEP-WP-0005-T01
status: done
priority: high
state_hub_task_id: "d2210752-57b5-5929-b788-08a5684b6db8"
```
Added bibliography-bearing source intake, immutable source digests, copied workflow
profiles and four ordered stages: claims, candidates, relation proposals, critique.
Each review sees the original source and all intermediate results. One source per
workspace avoids the upstream stage-output isolation limitation. Empty proposals
are permitted; no automatic entity or canon admission occurs.
## Verify fixture execution and failure boundaries
```task
id: FEP-WP-0005-T02
status: done
priority: high
state_hub_task_id: "ebe9bd19-7555-5ea6-bb71-0b45132b8905"
```
The synthetic fixture completed all four stages at
`research/bench/infospaces/fixture-smoke/`. `make bench-test` checks the real bench
workflow adapter: source and intermediate prompt context, provenance, source
mutation, multiple-source rejection, call cap, missing bibliography, failure receipt
and refusal to overwrite attempted runs. No live provider calls occurred.
## Document research operation and admission
```task
id: FEP-WP-0005-T03
status: done
priority: medium
state_hub_task_id: "d3fe2d80-438c-5b31-bf8a-66b7f661183a"
```
`research/bench/README.md`, `SourceInput.md` and `Admission.md` describe setup,
fixture/live commands, evidence status, ID mapping and review boundaries. Updated
agent workflow and research program entry points.
## Residual handoff
- Original source acquisition and evidence review: FEP-WP-0002-T02.
- Resolve metadata contract and provisional canonical ID mapping: FEP-WP-0002-T01.
- Cross-source synthesis and VC-01/VC-02 walkthroughs: FEP-WP-0002-T03.
- Typed graph export, metadata validator and retrieval evaluation: FEP-WP-0004.
Known limits: structural validation does not validate research claims; critique is
assisted self-review; custom run supports explicit OpenRouter or fixture adapters;
call caps do not enforce dollar/token budgets; one source per run; failed runs
require a fresh workspace rather than an automatic resume.