Receipts, not claims

Every claim on this site is auditable. Including this one.

Straight Edge AI cites primary sources, scores its own confidence, flags contradictions between sources, and says I don't know when the evidence doesn't support an answer. Below: how the comparisons are run, the published head-to-heads, and the exact JSON trace you can export from any answer.

How the head-to-heads are run

1. One hard question — a current event, a contested number, or something with a primary document behind it.
2. The identical prompt goes to Straight Edge AI and to each other assistant in a fresh session on the same day.
3. Answers are quoted verbatim. Nothing is trimmed to make a point.
4. Each competing answer is labelled by what actually happened: waffled, invented, refused, or correct — including when they get it right.
5. Straight Edge's run links to its public share page and sources.

Published head-to-heads

None published yet

No captured transcripts have been published here yet. We won't write comparisons from memory or paraphrase what another model “would probably say” — that would be exactly the behaviour this product exists to refuse. Captured runs appear here verbatim, with dates and sources, as they are recorded.

In the meantime, run one yourself: ask the same hard question here and anywhere else, then share your Straight Edge answer — the score and sources travel with the link.

Run your own comparison

The exportable reasoning trace

Every answer has an Export reasoning trace button. It downloads JSON containing each tool call with its exact inputs and raw outputs, the answer text, and the full verdict block — confidence, what it's based on, whether a knowledge gap remains, what would change the answer, the ranked sources, and any contradictions found between them. This is the real schema, so you know what you're getting before you sign up for anything:

{
  "exported_at": "2026-08-08T09:14:02.118Z",
  "message_id": "msg_7f3c…",
  "answer": "…",
  "verdict": {
    "confidence": 62,
    "basis": "One primary filing, one secondary report that disagrees on the date.",
    "knowledge_gap": true,
    "would_change": "The signed filing itself, or a correction from the issuer.",
    "sources": [
      { "title": "…", "url": "https://…", "tier": "primary", "why": "…" }
    ],
    "conflicts": [
      { "claim": "…", "positions": ["…", "…"], "discrepancy": "…" }
    ]
  },
  "steps": [
    { "tool": "webSearch", "state": "output-available", "input": { "query": "…" }, "output": "…" },
    { "tool": "fetchSource", "state": "output-available", "input": { "url": "https://…" }, "output": "…" }
  ]
}

Field values above are elided with “…”; the structure is exactly what the app exports.

Three questions. No account. Sources and score on every answer.

Ask something it could plausibly get wrong. If the evidence isn't there, it'll tell you that instead of guessing.

Start the free trial Straight Edge vs ChatGPT