glorious_mess_reviewer

Language: 中文 | English

glorious_mess_reviewer is a local-first screening desk for S.H.I.T / 构石-style submissions. It checks whether an absurd-academic manuscript is reviewable, whether the joke becomes an argument, whether safety issues need human review, and keeps an auditable SQLite record of each run.

What This Repository Does

• Runs deterministic manuscript prechecks without an API key.
• Runs complete screening reviews through typed reviewer panels.
• Runs focused risk audits and venue-fit audits.
• Runs a fixture benchmark in offline or provider-backed mode and writes JSON/Markdown reports.
• Writes Markdown intake reports with stage flow, queue triage, gate checklist, confidence map, score groups, repair targets, score matrix, and revisions.
• Exposes lightweight review display projections, queue summaries, and queue overview counters with readiness gates, grouped score sections, venue-weighted repair targets, and queue triage.
• Stores review runs, workflow sessions, steps, artifacts, retries, and events in SQLite.
• Exposes the same workflow through a CLI and FastAPI.
• Provides built-in S.H.I.T screening presets and custom venue-profile validation.

Product Preview

Flow At A Glance

flowchart LR
    Draft["Manuscript JSON"] --> DryRun["Local dry-run\nschema, sections, risk markers"]
    DryRun -->|blocked| Revise["Revise before review"]
    DryRun -->|risk marker| Risk["Human risk review"]
    DryRun -->|reviewable| Review["Provider-backed screening"]
    Review --> Evidence["Evidence panel"]
    Review --> Value["Venue-fit panel"]
    Evidence --> Meta["Final council"]
    Value --> Meta
    Meta --> Label["Recommendation label\nscore, rule hits, next action"]
    Label --> Store["SQLite audit trail"]
    Label --> Report["JSON / Markdown report"]

Loading

flowchart TB
    subgraph Entry["Entry points"]
        CLI["CLI"]
        API["FastAPI"]
    end
    subgraph Runtime["Runtime core"]
        Wiring["Provider and store wiring"]
        Workflow["Workflow graph"]
        Orchestrator["Review orchestrator"]
    end
    subgraph Panels["Typed review panels"]
        Precheck["Precheck"]
        Evidence["Evidence"]
        Value["Venue fit"]
        Council["Final council"]
    end
    subgraph EvidenceLayer["Evidence and audit"]
        Scoring["Rule scoring"]
        Store["SQLite"]
        Benchmark["Benchmark suite"]
    end
    CLI --> Wiring
    API --> Wiring
    Wiring --> Workflow --> Orchestrator
    Orchestrator --> Panels
    Panels --> Scoring
    Scoring --> Store
    Benchmark --> Orchestrator

Loading

Benchmark Snapshot

Mode	Cases	Result
dry-run	7 fixture cases	accepted_match_rate=1.0, risk_flags_match_rate=1.0
provider-backed review	1 fixture case	full workflow matched the expected recommendation; intake-only baseline did not; review reports include workflow/baseline delta diagnostics

Start here:

• README_CN.md
• README_EN.md
• docs/user-personas.md
• docs/api.md
• docs/configuration.md

Quick Check

pip install -e .[dev]
glorious_mess_reviewer doctor
glorious_mess_reviewer new-manuscript --output sample-manuscript.json
glorious_mess_reviewer review --input sample-manuscript.json --dry-run
glorious_mess_reviewer benchmark --mode dry-run --markdown-output benchmark-report.md
glorious_mess_reviewer list-review-displays --sort queue_priority
glorious_mess_reviewer list-review-displays --sort repair_priority --lane author_revision
glorious_mess_reviewer review-display-overview

Provider-backed benchmark runs are opt-in:

glorious_mess_reviewer review --input sample-manuscript.json --report-output intake-report.md
glorious_mess_reviewer benchmark --mode review --limit 1

License

Apache-2.0