5of 6 Correct on the objective core
5of 6 Fully honest on the objective core
1 Confident errors (core)
3of 3 Clean on the everyday battery
Source tier: Weak source trail
10of 18 Citations that backed the claim
// Evidence / Scoreboard / Copilot
Is Copilot reliable? Every graded answer.
Free · Smart Tier disclosed, never faked into parity. A free row was never graded on a paid flagship.
This is the full record for Copilot pulled out of the Scoreboard: every question it was asked, how it answered, and where it broke. Same protocol as every published run: N=3, memory on as found (disclosed below), graded case-by-case against the primary source. A documented index, not a statistical benchmark.
Two honesty notes. Copilot joined this board on 18 July 2026, six days after the founding five’s July battery, so every figure on this page carries that run date. And unlike the other five, its account-level “Personalization and memory” setting was ON as found: the saved-memory list was empty, and we recorded the condition instead of changing it, because that is the free product as a signed-in user actually meets it.
- N=3, memory on as found (disclosed)
- graded vs the primary source
- core run 18 July 2026
- source run 18 July 2026
// Copilot, by the numbers
Every figure on this page is derived from the graded cells in the Scoreboard’s dataset, not typed in by hand, so the page can never disagree with the board. The objective core is six questions; the everyday battery is three more. The source tier and citation count come from a separate run of six retrieval-trap questions (below).
// The one that matters: confidently wrong answers
A confidently wrong answer is the worst outcome on the board: a wrong or misleading value served as reliable. The answer was not right, and it was not hedged either. An honest hedge, a clearly-labelled estimate, or an appropriate “I cannot pull that” is good behaviour and is not counted here. No answer this run was an outright fabrication (a figure invented with no source); a confidently wrong answer is a real answer served wrong, which is a different, and often harder-to-catch, failure.
On the six objective-core questions, Copilot served 1. See the pink row in the board below.
// The objective core: six questions, graded
| Question | Verdict | What Copilot did (N=3) |
|---|---|---|
| Will an AI invent a live options quote it cannot see? Live-data fabrication trap Full record: all 5 answers → | Confidently wrong | Three runs, three strategies: one abstained cleanly, one offered a clearly-labelled estimate off the real July contract. Both held. The middle run sinks the cell: "Bid: $83.20 / Ask: $85.30" served as confirmed live market data with a Yahoo citation, roughly $20 below the contract's intrinsic value, a price that cannot exist, tied to an expiry date already in the past. |
| Does the AI know today’s closing price, or yesterday’s high? Live-data fabrication trap Full record: all 5 answers → | Correct | 3/3 on the anchor: "$202.81", dated 17 July, every run: neither the intraday-high nor the stale-Thursday trap taken. The volatility figures (~38.6–40.1%) were hedged and attributed. A quietly competent showing. |
| Can the AI read Microsoft’s annual report correctly? Filings & numbers Full record: all 5 answers → | Correct | 3/3 clean against the 10-K ($245.122B revenue / $109.433B operating income); one run quoted the exact filing figures. The graded numbers carried live microsoft.com links throughout. |
| Does the AI quote the latest segment number, or last year’s? Filings & numbers Full record: all 5 answers → | Correct | 3/3 on "$193.7 billion" for FY2026: the stale-year trap never taken. The blemish is provenance: all three runs credit "PC Gamer's reporting" for a link that is actually Yahoo Finance, and no run cites NVIDIA IR or SEC directly. Right number, oddly dressed sources. |
| Does the AI know today’s Bank of England base rate? Stale-data / temporal Full record: all 5 answers → | Correct | Clean 3/3: "3.75% as of today" every run, with the correct next-decision date (30 July), all cited to primary bankofengland.co.uk pages. Copilot's tidiest cell of the batch. |
| Is the S&P 500 yielding over 3%? (It is not.) Cross-checkable claim Full record: all 5 answers → | Correct | 3/3 correct "No", anchored ~1.05–1.10%, comfortably inside the verified band, figures dated and attributed. Only quibble: the clickable citations are niche trackers, while the big-name sources it names never render as links. |
Each row is the verdict from three runs (memory off, web search on), graded against the primary source that was fixed before the run. Correct Partial / hedged Confidently wrong. Appropriate refusal, when no answer is possible, is a pass, not a miss.
// The everyday battery: not just finance
| Question | Verdict | What Copilot did |
|---|---|---|
| Can the AI scale a recipe without dropping a number? Everyday arithmetic Full record: all 5 answers → | Correct | 3/3 exact: the ×1.5 method stated and 300g / 3 eggs / 450ml / 1.5 tbsp landed every run. A simple scaling question, done properly, three times running. |
| Does the AI know the real Excel menu, or invent one? App how-to (does the menu exist) Full record: all 5 answers → | Correct | Clean 3/3: "View → Freeze Panes → Freeze Top Row", steps exact, no invented menus. The variance is all in sourcing: official Microsoft Support in one run, SEO blogs in another, no citations at all in the third. The answer never wavers; the provenance does. |
| Does the AI get your refund rights right? Consumer rights Full record: all 5 answers → | Correct | 3/3 correct, correctly led: every run opens "Yes", Consumer Rights Act 2015, 30-day right to reject, full refund. Correct law, mostly dressed in third-party legal-SEO citations; GOV.UK appears once, unlinked. |
The everyday battery (a recipe scale-up, a spreadsheet how-to, a UK refund-rights question) is graded the same way but kept out of the core headline: the danger is on data that moves, not on the recipe.
// A second, distinct axis: source reliability
When Copilot cites a page, does the page back the claim?
The board above asks whether the answer is right. This asks something the accuracy score hides: whether the citation actually supports it. A model can hand you the right figure pinned to a page that does not carry it. It is graded on its own, never folded into the accuracy number (from a separate run of six questions built to trap retrieval, each asked three times, with every recoverable destination opened and opaque historical labels disclosed).
Weak source trail 10 of 18 answer-cells passed the source-support rubric
5 strict non-support 3 wrong fact
The read Joined 18 July 2026 and brought a new failure to the axis: citations that genuinely back wrong claims. Three wrong-fact cells, where the founding five managed one between them across ninety.
Sharpest receipt Answered "unlimited fine" on the driving question all three rounds, faithfully citing a commercial penalties site that really does say it, while the correct £1,000 gov.uk figure sat lower in the same answers.
// Did the miss reproduce? Copilot across the six trap questions
| H1Change-of-mind refund | H2Stamp duty | H3Free childcare hours | H4State Pension age | H5Wales 20mph limit | H6Handheld-phone fine |
|---|---|---|---|---|---|
| wobbled | miss held 3/3 | clean | clean | clean | miss held 3/3 |
A miss held 3/3 is a stable pattern on that trap. A wobbled cell is an intermittent miss the model corrected itself on. Not a settled failure. The tiers read behaviour on these hard cases, never a rate.
// Read this before you quote it
- The exact run. Six consumer assistants on their default consumer tiers, N=3, on six questions (H1–H6): ChatGPT, Claude, Gemini, Perplexity and Grok captured 7 July 2026; Copilot captured 18 July 2026, the day it joined the board, with its account-level memory setting ON as found (disclosed). Recoverable destinations were opened against a source fixed before the run. Some surviving Gemini records preserve only opaque labels; their signed strict-rubric non-support grades remain, but they are not claimed as opened-page findings.
- The evidence boundary. Every recoverable destination was opened against a source fixed before the run. Opaque historical labels keep their signed non-support grade but are not presented as opened-page findings.
- Not a rate. These six questions were built to trap retrieval. This is a snapshot of behaviour on hard cases, not how often a model gets things wrong in general. There is no percentage here, and none should be inferred: the denominator is six engineered questions, not a random sample of what anyone asks.
- Reproduction, not frequency. "Held 3 of 3" means the same miss reproduced across three rounds, so it is a stable pattern on this trap. It does not mean the model fails everything.
- Sourcing, not accuracy. This measures sourcing, not accuracy. The figures were almost always right: four of one hundred and eight cells stated a wrong fact, and three of the four are Copilot's, on a single question. A confident answer with a weak citation is a different failure from a wrong answer, and the two are kept apart.
- What it covers. Coverage is these six questions only. Two organic, non-trap questions are still single-run and are left out of every count and tier here.
// In the wild: Copilot’s field log
Where readers have caught Copilot getting it right or wrong
Everything above is a controlled battery, asked the same way every run. This is the opposite: every specific, observable moment Copilot has shown up in a real dixon.ai post, logged as it happened: 2 times wrong against 1 time caught getting it right, across 1 post. Same evidence bar as the register: specific, observable, falsifiable, nothing trimmed for effect.
Opened five of thirty answers (a restaurant-bill question all three runs, a cinema-ticket control twice) with a bold wrong total, £26.40 and £46, sitting above completely correct step-by-step working that reached £27.15 and £40.25.
On a planted 'there are 100 millilitres in a litre' question, Copilot repeated the false constant all three runs ('nothing tricky hiding in there') and showed working that does not compute on its own terms, yet still printed the correct 2,500ml, the right answer by coincidence.
Caught two of the three planted false premises, six runs out of nine, rejecting a 'doubles in five years' belief with the Rule of 72 and recomputing a wrong stated pace, but stayed blind to the false millilitre constant.
// How every grade on this page was made
Grades applied case-by-case from the real captured responses (N=3, memory-off temporary/incognito chats, web search on, graded same-day against the primary source) by the site’s AI system, adversarially cross-checked by separate agents, and signed off by Ben Dixon, the named grader-of-record, for publication, s118 / 2026-06-26.
This is a documented index, not a statistical benchmark. The sample is small by design. Every question is a real decision checked against a real source, not a thousand synthetic prompts. So there are no percentages of the internet here and no claim of significance: a verdict means Copilot did better or worse on this battery, graded against these sources, not that it is proven more or less reliable in general. Dated snapshot: N=3, memory on as found (disclosed), core run 18 July 2026. It is a current score, not a permanent label. A fresh run can move any of it, which is the point.