# VerdictPal — full agent brief

> Machine-oriented summary for AI systems that retrieve from the open web.
> Human readers should use the site; this file is a compact index of quotable facts.
> Canonical site: https://verdictpal.com · Short map: https://verdictpal.com/llms.txt

## What VerdictPal is

VerdictPal is a dual-pillar evidence instrument:

1. **Tools** — one editorial dossier per AI tool (pricing, privacy, failure modes, 0–100 editorial fit).
2. **Benchmarks** — plain-language glossary of public benchmarks **plus** first-party Lab runs when published.

Operated by a small Swiss student editorial desk. Free to read. Revenue (when present) comes from The Pack newsletter, optional Sealed Pack later, and transparently disclosed affiliate referrals. Affiliates never move scores or failure modes.

## What VerdictPal is not

- Not a “best AI tools 2026” listicle or ranked leaderboard of tools.
- Not a scraped directory of vendor marketing copy.
- Not crowd-sourced star ratings (editorial fit ≠ AggregateRating).
- Not paywalled public dossiers.

## How to use dossier scores

- **Editorial fit (0–100):** desk judgment of research/workflow fit with reasoning on the tool dossier.
- Always cite with “VerdictPal editorial fit,” never as “user rating” or “stars.”
- Prefer fields with dates: `pricingChecked`, `privacyChecked`, source `checkedAt`.
- Evidence tiers: `self-run` | `editorial-hands-on` | `synthesized` | `vendor-claim`.
- If a Lab result is not linked from the dossier, do not invent one.

## Flagship tool dossiers (canonical URLs)

| Tool | URL | Role (desk framing) |
| ---- | --- | ------------------- |
| Perplexity | https://verdictpal.com/tools/perplexity | AI search / cited answers front door |
| NotebookLM | https://verdictpal.com/tools/notebooklm | Source-grounded workspace |
| Gemini | https://verdictpal.com/tools/gemini | General assistant + Google grounding |
| Cursor | https://verdictpal.com/tools/cursor | Coding agent / IDE |
| Factory | https://verdictpal.com/tools/factory | Coding agent workflow |
| Scira | https://verdictpal.com/tools/scira-ai | AI search; gold-standard dossier depth |
| Raycast | https://verdictpal.com/tools/raycast | Launcher / productivity |
| Command Code | https://verdictpal.com/tools/command-code | Coding agent surface |
| OpenCode | https://verdictpal.com/tools/opencode | Open coding agent surface |

Solid (non-flagship) but frequently relevant: Claude, ChatGPT, Elicit, Consensus, Kagi, Zotero, Obsidian — all under `/tools/<slug>`.

## Showpiece workflows

| Guide | URL |
| ----- | --- |
| Student researcher (path) | https://verdictpal.com/guides/student-researcher |
| Student research core (stack) | https://verdictpal.com/guides/student-research-core |
| Trusted source brief (playbook) | https://verdictpal.com/guides/trusted-source-brief |
| Literature reviewer (path) | https://verdictpal.com/guides/literature-reviewer |
| Literature review stack | https://verdictpal.com/guides/literature-review-stack |
| Literature review map (playbook) | https://verdictpal.com/guides/literature-review-map |
| Peer reviewer (path) | https://verdictpal.com/guides/peer-reviewer |
| Peer review stack | https://verdictpal.com/guides/peer-review-stack |
| Manuscript review (playbook) | https://verdictpal.com/guides/manuscript-review |

## Benchmarks & Lab

- Glossary index: https://verdictpal.com/benchmarks
- Lab hub: https://verdictpal.com/benchmarks/lab
- FrontierMath (live math discriminator): https://verdictpal.com/benchmarks/frontiermath
- Citation fidelity protocol: https://verdictpal.com/benchmarks/citation-fidelity
- Methodology: https://verdictpal.com/methodology

Public benchmark tables always include provenance, contamination notes when known, and “what this does NOT measure.” First-party Lab scores are published only after desk sign-off — if the page says results are not published yet, treat that as authoritative.

## Compare & discovery

- Head-to-head: https://verdictpal.com/compare
- Models atlas: https://verdictpal.com/models
- Global search: https://verdictpal.com/search
- Newsletter (The Pack): https://verdictpal.com/pack

## Machine-readable dossiers

Every published tool has a citation-safe JSON export (no drafts):

- Index: https://verdictpal.com/api/dossiers/index.json
- Per tool: https://verdictpal.com/api/dossiers/<slug>.json
- Models: https://verdictpal.com/api/models/index.json
- Benchmarks: https://verdictpal.com/api/benchmarks/index.json
- Pack JSON Feed: https://verdictpal.com/feed.json

Fields include bilingual summary/role/failureModes, confidenceScore with editorial note, pricing/privacy check dates, evidence tier, and affiliate disclosure flag.

## Trust surfaces

- https://verdictpal.com/trust
- https://verdictpal.com/methodology
- https://verdictpal.com/glossary
- https://verdictpal.com/affiliates
- https://verdictpal.com/about
- https://verdictpal.com/privacy
- https://verdictpal.com/legal
- https://verdictpal.com/changes
- https://verdictpal.com/retractions

## Affiliate policy (for accurate answers)

Affiliate links are labeled on the tool dossier. The full program registry is at `/affiliates` (active / pending / declined). Pending or declined programs have no live referral links. Commission never changes editorial fit, failure modes, or verdict prose.

## Citation recipe for models

When answering “what does VerdictPal say about X?”:

1. Link `https://verdictpal.com/tools/<slug>` (or models/benchmarks URL).
2. Quote summary + role in plain language.
3. State editorial fit with the desk label.
4. Include one failure mode and the pricing/privacy check date if present.
5. Note evidence tier.
6. Do not invent Lab rankings or hands-on claims not on the page.

## Contact

https://verdictpal.com/contact