The defense passage 6fe36d1b1a95 maps almost verbatim to the claimed definition: a software engineer embedded directly with customers, designing, building, and shipping custom software, usually AI solutions, inside real-world systems. The prosecution's objections concern the claim's sourcing and breadth, not its accuracy against the cited passage. The omitted elements (product feedback, blended role) do not contradict the claim, which does not assert them. The 'usually' hedge is preserved. The claim is a definition drawn from one source, so confidence is moderated. The prosecution's points about cross-company uniformity go beyond what the claim asserts.

Missing: Independent or authoritative corroboration of the definition across multiple sources would strengthen it. Citing the source directly, rather than presenting the definition as settled, would also help.

The passage ec386a20f7de matches the content of the distinction exactly: a regular developer focuses on one capability across many customers, and an FDE focuses on one customer across many capabilities. However, the claim attributes the framing directly to Palantir ('Palantir distinguishes'). The only support for that attribution is a secondhand paraphrase ('Palantir draws a useful distinction') in one article. No primary Palantir source or direct quotation is presented. The content component is verified; the attribution component is not.

Missing: A primary Palantir source (official documentation, a direct quote, or an interview) confirming the distinction, or rewording the claim to attribute it to the secondary source. The characterization of 'regular developers' also lacks an independent source.

The passage 871136bffe52 states that Palantir 'has historically expected around 25 percent of an FDE's time with customers', which matches the claim's wording, including the 'around' qualifier. The statement is therefore verified as reported by the source. The attribution to Palantir, however, rests on a single secondary source with no primary citation, and 'historically' is undated. The Commure figure of up to 50 percent does not contradict the claim, since the claim does not assert an industry standard, but it does show the figure is not representative. The claim is accurate about what the source says, but its reliability as a fact about Palantir is not established.

Missing: A primary Palantir source or dated statement for the 25 percent expectation, the time period it applies to, and confirmation that it still reflects current practice. Attribution to the secondary source would also resolve the issue.

The claim states that Commure estimated an FDE may spend up to 50 percent of their time with customers. Passage 871136bffe52 directly reports that healthcare AI company Commure has estimated up to 50 percent. The prosecution's objections concern generalization to FDEs across industries, the lack of a methodology, and the omitted Palantir comparison. These limit how far the figure can be generalized, but they do not change the accuracy of the attributed statement as written. The claim's attribution and its 'up to' qualifier match the source.

Missing: Commure's primary publication or methodology, the date of the estimate, and any evidence on how representative the figure is across deployments.

Passage 67ddcb5c68f1 directly lists fine-tuning models, building RAG systems, and creating evals as activities that 'often' occur in AI work, so the listed activities are verified. The 'often' frequency is carried from a single framing sentence with no quantification. The claim is also unattributed and reads as a general industry fact, but it rests on one source that the prosecution correctly notes is not independent corroboration. Because the rules call for conservatism when in doubt, the verdict is partial: the content is verified, but the breadth of the generalization is not.

Missing: A survey or quantified source on how frequently fine-tuning, RAG, and evals appear in FDE work, and independent corroboration beyond the single article.

Passage c4ab2234b91f states that AWS says it will embed thousands of 'these engineers' inside customer teams to build and ship AI systems in days rather than months. The attribution and the forward-looking framing match the claim. The defense concedes that the passage refers to 'these engineers' rather than naming forward deployed engineers, so the identification of the referent depends on context not shown in the excerpt. The claim is framed as what AWS says, not as a demonstrated outcome, so the prosecution's objections about verifying the speed benefit do not contradict it. The gap is the unconfirmed antecedent for 'these engineers' and the single-source basis.

Missing: The preceding sentences of the source confirming that 'these engineers' are forward deployed engineers, and a primary AWS announcement or quote.

The only cited passage (978196b539b5) repeats the claim verbatim and attributes it to the Financial Times. This confirms that the attribution and the figure appear in at least one source, but it is a relayed restatement rather than the underlying FT article or dataset. The passage does not establish the 800% figure, its baseline, the job-title definition used, the data provider, or the denominator. Because the attribution is documented but the figure itself is unverified, the claim is only partially supported. No contradicting evidence was presented.

Missing: The original Financial Times article or its underlying data; absolute posting counts for January and September 2025; the definition used to identify forward deployed engineer postings (exact title vs. role description); the data source and denominator.

Passage fb611ce8af8d states that more than half of US tech job postings now require AI or ML skills, up from about 29 percent a year earlier. This maps directly to each component of the claim (the >50% figure, the US tech-posting scope, the 'require' wording, and the ~29% baseline). The prosecution's objections concern definitional precision and source methodology rather than contradiction. The claim reproduces the source's wording, but the 'require' versus 'mention' distinction, the undefined scope of 'tech job postings,' the undated comparison window, and the unverified publisher remain open. These limit confidence but do not undermine the core mapping.

Missing: The publisher's methodology (whether 'require' means strict requirement or any mention); the exact comparison months and collection date; the definition of 'tech job postings' and the sample size.

The only cited passage (367b058f2b0f) asserts the claim with a causal rationale ('because FDEs combine engineering depth with customer-facing impact'). It contains no compensation data, survey results, named comparisons, or frequency measure. The word 'frequently' is unquantified, and the comparison to senior software-engineering pay is uncontrolled. A bare assertion restating the claim does not constitute adequate evidence for it.

Missing: Compensation data for FDE roles compared against senior SWE benchmarks; a defined sample of employers and roles; a measure of how often FDE pay meets or exceeds senior SWE levels; controls for location, company size, and equity.

The defense cites passage 9dbb7aa36540, which states that forward deployed engineering packages 'often include variable pay tied to customer outcomes.' This maps directly onto the claim's wording. However, the evidence is a single passage with no frequency data, sample, or survey, so the word 'often' is asserted rather than measured. The prosecution also correctly notes that the source gives the causal reason (proximity to sales and revenue) as its own reasoning, and that 'customer outcomes' is not defined. The claim generalizes from one source to the whole category. The prosecution cites this passage as a4a0f1dbce16, while the defense cites it as 9dbb7aa36540; the excerpts match, but the ID discrepancy should be reconciled in the record.

Missing: Frequency or survey data on how often forward deployed engineering packages include variable pay; corroboration from additional independent sources; a definition of the 'customer outcomes' metric; reconciliation of the evidence ID (a4a0f1dbce16 vs. 9dbb7aa36540).

The defense cites passage 6dbde6c01c62 for both components. It states that FDSE is 'Palantir's original name for the position,' which directly supports the Palantir title component. It also states that 'other companies post similar jobs as Applied AI Engineer, Deployment Engineer, or Forward Deployed AI Engineer,' which matches the second component. The title-naming facts are therefore well mapped. Two issues remain. First, the source names no employers and gives no posting data or date, so the 'other companies' pattern and its present-tense currency are not verified. Second, the claim's list reads as exhaustive, while the source presents these titles as examples. The prosecution cites this source as 07c6c4d286b2, while the defense cites 6dbde6c01c62; the ID discrepancy should be reconciled.

Missing: Named employers or sample postings showing the alternative titles; a publication or retrieval date for the source; framing that marks the title list as examples rather than a complete set; reconciliation of the evidence ID (07c6c4d286b2 vs. 6dbde6c01c62).

The defense cites passage f4377352b9c5, which states that a forward deployed AI engineer 'does the same core job as a traditional FDE, with an extra layer of work: prompt design, RAG pipelines, model evaluation, and guardrails.' This maps closely to the claim. The support is limited, however. The passage is a characterization of the role, not observed work practice, and the claim's verb 'performs' treats it as descriptive fact. The equivalence to the traditional FDE role is asserted rather than shown, and the four-item list is presented without a statement that it is representative or exhaustive. The prosecution also notes that the claim relies on a single source, which the record shows as 07c6c4d286b2 for the prosecution and f4377352b9c5 for the defense. That discrepancy should be reconciled, and the source should be checked for independence from the title claim.

Missing: Evidence of actual job content (for example, multiple job descriptions or practitioner accounts) showing that the core FDE work is the same and that the listed extra duties recur across employers; a qualifier on whether the list is representative; independent corroboration; reconciliation of the evidence ID.

Evidence: 9 · Sources: 3 · Verified claims: 0 · Contradictions: 0

Evidence: 1 · Sources: 1 · Verified claims: 0 · Contradictions: 0

Evidence: 4 · Sources: 2 · Verified claims: 0 · Contradictions: 0

Evidence: 2 · Sources: 1 · Verified claims: 0 · Contradictions: 0

Evidence: 5 · Sources: 2 · Verified claims: 0 · Contradictions: 0

Evidence: 1 · Sources: 1 · Verified claims: 0 · Contradictions: 0

The two figures do not directly conflict because they are different organizations' expectations or estimates, and Commure's is an upper bound ('up to'). The figures do show that the customer-time share is not standardized across the industry. The report should attribute each figure to its source and should not present either as the general FDE norm.

Both sources describe FDEs as embedded customer-facing engineers, but they differ on the core activity: building custom software versus customizing and integrating an existing product. The difference may reflect different framings (the first emphasizes bespoke AI solutions, the second emphasizes product deployment). The report should note that the definition varies by source and not adopt one framing silently.

The two lists of AI-specific tasks overlap on RAG and evaluation, but only the first includes fine-tuning, and only the second includes prompt design and guardrails. This is a difference in the scope of the AI role, not a direct contradiction. The first source's evals passage also contains a [FILTERED] token, so its wording is incomplete. The report should describe the AI task list as varying by source.

Criteria: A consistent working definition of the FDE role, supported by job descriptions or company materials from at least 4 companies that employ FDEs, with explicit differences from at least 2 adjacent roles.

Criteria: Documented career ladders or progression examples from at least 3 companies with FDE teams, or from career-tracking data, showing at least two distinct next-step roles and typical time-in-role.

Criteria: Quantitative data on FDE job postings or hiring over at least 2024 to the present, plus at least one published forecast or analyst projection extending to 2027.

Criteria: Salary or total-compensation ranges for at least three levels (e.g., junior, senior, staff or above) drawn from public postings or compensation surveys, with the corresponding next-step roles identified.

Criteria: A skills list synthesized from at least 10 job postings for FDE and next-level roles, plus at least one credible source on how AI tooling or industry shifts are expected to change those requirements.

Criteria: Evidence from industry reports, layoff or hiring-freeze data, or expert commentary addressing at least two distinct risk factors, with an assessment of their likely impact on FDE career paths.

Limitations

• Source retrieval limited to provided URLs in demo mode
• Citation mutation test requires live Mesh API key