Caught well today
Internal math & arithmetic
A number that does not follow from its own inputs. The engine recomputes checkable values and flags the ones that do not foot — the same class as a worked example whose total contradicts its line items.
Accuracy is the product you sell. exclam.ai is a pre-publication QA layer that re-reads your material and flags the errors that become errata notices — wrong arithmetic, answer-key mismatches, broken cross-references — and routes each candidate to your editors with the evidence attached.
Or write directly to errata@exclam.ai.
By the time an erratum is published, the damage is already done: candidates studied the wrong number, support tickets piled up, and a brand whose entire promise is precision took a visible hit. For a prep provider, accuracy is not a feature — it is the reason people pay you instead of using a free PDF.
The same errors are far cheaper to catch one step earlier. exclam.ai points a deterministic checking layer and document-wide reasoning at your material before it ships, surfaces the checkable error classes as a triage queue, and hands each candidate to a human with the evidence attached. Reviewing a flagged item is minutes; a published erratum is refunds, churn, and a correction notice with your name on it.
Run over one published CFA Level III volume with no access to the answer key, the engine surfaced internal-inconsistency and numeric errors — including candidates that were not on the official errata notice.
We lead with the error classes a machine can verify against ground truth, where precision is defensible. The judgement calls stay advisory, with a subject-matter expert always making the final decision.
A number that does not follow from its own inputs. The engine recomputes checkable values and flags the ones that do not foot — the same class as a worked example whose total contradicts its line items.
A solution that re-solves to a different answer than the one printed, or a rationale that contradicts its own key. Re-solve, then compare — checkable without a human first pass.
Broken "see Exhibit 7" pointers, equation numbers that drifted after an edit, mislabeled exhibits, and structural formula typos. Pattern- and structure-based, so precision is high.
The same concept given two conflicting values across volumes or editions. Needs document-wide reasoning; we surface candidates for an editor to adjudicate rather than auto-correct.
A subtly wrong explanation. There is no internal signal to verify these deterministically, so they stay advisory and a subject-matter expert always makes the final call.
You do not have to share anything sensitive to see whether this works. We start on content you have already published, prove the number, then move earlier in your pipeline.
We run the engine over an edition you have already published and score every flag against your own errata list — your ground truth. A fixed-scope, low-risk first engagement that proves the precision number before anything else.
Point the same engine at the next edition before it ships. Flagged candidates go to your SMEs as a triage queue — they confirm or dismiss. You catch the embarrassing erratum before a candidate does, not after.
Track drift across editions, volumes, and derivative products as the official curriculum changes. Protects the accuracy promise that is the core of your brand.
On-prem / VPC deployment for unreleased editions and live exam items, with data handling scoped in writing up front. See our security overview.
Every flag is a candidate with its evidence, not an auto-edit. Your editors confirm or dismiss; the engine never changes your curriculum on its own.
Any team shipping high-stakes, numerically dense curricula — where a single wrong number is a support fire and a refund request.
No. For unreleased editions and live exam items we deploy on-premise or into your own VPC, so the content never leaves infrastructure you control. Data handling is scoped in writing before any material is shared.
No — it is a triage layer in front of them. The engine surfaces candidate errors with the evidence for each; your subject-matter expert always makes the final call. It is built to save reviewer time on the mechanical, checkable error classes, not to make editorial judgements.
We report precision and recall honestly and per error type, scored against your own published errata as ground truth. We optimise for precision first: a tool that reliably catches the checkable errors with few false alarms is worth more to a reviewer than one that claims high recall and floods the queue with noise. The retrospective audit is where you see the real number on your own material.
Internal arithmetic that does not foot, answer-key and solution mismatches, and broken cross-references, figure/table numbering, and formula typos. In a blind retrospective over a published CFA Level III volume the engine surfaced internal-inconsistency and numeric errors — including candidates that were not on the official errata notice — without ever seeing the answer key. Label-only and "delete this paragraph" errata have no internal signal and stay human-only.
Any high-stakes, numerically dense curriculum: CFA, CPA, FRM, CAIA, actuarial (SOA/CAS), and the textbook publishers behind them. The engine is content-agnostic — it analyses material you already own or license, in a clean-room workflow.
With a fixed-scope retrospective audit on an edition you have already shipped. It de-risks the decision for you, needs no access to unreleased content, and gives both sides a concrete precision number before any subscription conversation.
Tell us the program and the edition you ship. We will set up a short call, agree the scope, and run a fixed-price retrospective audit on material you have already published — so you see the precision number on your own content before committing to anything.
exclam.ai is not affiliated with or endorsed by CFA Institute or any prep provider. CFA Institute, CFA, and Chartered Financial Analyst are trademarks owned by CFA Institute. The engine analyses only material you own or are licensed to use, in a clean-room workflow; a subject-matter expert reviews every flag.