with LLM-assisted feature extraction
The target is most consistent with Reporter C across all measured stylistic dimensions, with strong positive matches on first-person framing, intensifier usage, register, and rhetorical structure. The signal is meaningful but not conclusive given the short sample length and the tendency of high-emotion writing to compress individual style. Recommend treating this as a strong directional indicator pending corroboration through other investigative means.
Reporter C
Primary attributionThe target shares Reporter C's most distinctive features: emotional directness with first-person framing, sentence-final intensifiers ("I've watched," "I am tired"), and a willingness to use unmoderated declarative judgments ("I do not accept that"). Both samples open with a self-justifying statement of stakes ("I'm not writing this lightly" / "I'm sorry but this is genuinely not okay") before moving to a specific complaint.
- ›First-person singular pronouns per 100 words: target ≈ 9.0, Reporter C ≈ 8.2, others ≤ 1.5
- ›Sentence-initial "I" frequency: target 5×, Reporter C 4× in 50 words, others 0–1×
- ›Use of intensifiers "genuinely," "actually," "frankly" — present in both target and Reporter C, absent in others
- ›Both samples deploy a rhetorical concession before escalation ("I am tired of pretending" / "I don't blame them")
- ›Both avoid corporate registers ("per the discussion," "flag," "OKR") that dominate Reporters A, B, and D
Reporter A
Reporter A's sample has direct language and willingness to flag friction, but the register is consistently lowercase-informal ("hey just wanted to flag," "can we look at") with frequent use of softening hedges ("feels like," "happy to draft") that do not appear in the target. The target's escalation tone is not characteristic.
- ›Reporter A uses lowercase sentence-initial casing throughout — absent in target
- ›Hedges ("feels like," "happy to," "could be") frequent in sample, absent in target
- ›Reporter A uses no emotional intensifiers; target uses several
Reporter D
Reporter D's writing is procedural and management-coded ("per the discussion," "operationally feasible," "formalize the dependency"). The target document contains none of these markers and operates in a fundamentally different register.
- ›Corporate-register vocabulary in sample, absent in target
- ›Sample uses passive voice and nominalization heavily; target is direct active voice
- ›No first-person emotional framing in sample; pervasive in target
Reporter B
Reporter B's sample is structured and enumerated ((1), (2), (3)) with explicit calendar coordination. The target is unstructured prose with no enumeration, no logistics, and no concrete asks. Stylistic profiles diverge across nearly every measured dimension.
- ›Enumerated lists in sample, absent in target
- ›Concrete logistics ("Thursday after 2") in sample, absent in target
- ›Sentence-final periods only; target uses sentence-final periods plus declarative judgments
- ⊘Samples are short (~50 words each). Above 300 words per sample, signal strength roughly doubles.
- ⊘Anonymous complaints often involve deliberate stylistic masking. Compared sample writing was likely written without that intent.
- ⊘Emotional content compresses individual style — when people are upset, they often write more like one another than like themselves at baseline.
- ⊘Do not confront the suspected author on the basis of this report alone. Use as one input in a broader investigation.