01. The Problem

Ask any capable language model to analyze a person and it returns something that reads like expertise. The trouble is that it is pattern-completing what a profile sounds like, not reasoning from evidence about this specific person. The output is fluent and confident, and it is frequently wrong.

The failure is invisible, which is what makes it dangerous. A fabricated profile is internally consistent by construction, so it looks exactly as authoritative as an accurate one, and nothing about reading it tells you which parts came from evidence and which were filled in. People act on profiles. They end relationships, make hires, and decide how to treat someone based on a read, so a confident wrong profile is worse than no profile.

A fabricated profile is internally consistent by construction. It looks exactly as authoritative as a correct one.

02. The Iron Law

The whole system is built on one rule, applied without exception. Every factual or interpretive claim must be backed by a verbatim quote from the material gathered this run. A claim that cannot be quoted is cut, not softened, not hedged, not kept because it is probably true.

Two disciplines follow from it. Frameworks are used as lenses rather than labels, so the output reads consistent with high Detachment rather than is detached, because that is an honest description of what reading behavior can establish. And behavior outranks self-report. What a person did counts for more than what they call themselves, and when the two diverge, that gap is one of the most useful findings in the profile.

A reading that cannot point to a quote is cut, not softened.

03. The Pipeline

A single prompt asked to read everything and write a profile collapses gathering evidence, judging it, and writing into one pass, and in that pass the model can invent a quote, apply a framework to a self-label, or inflate a weak signal with nothing to catch it. The pipeline splits those jobs so they check each other.

The synthesists never read the raw material. They can only build on quotes the extractors pulled and the gate already checked, so a claim with no evidence behind it has nothing to stand on. Models are assigned by job, a fast one for extraction and the gate, the strongest for the synthesis where judgment lives.

04. The Frameworks

The knowledge is in the skill, not improvised from training data. A librarian agent turns each framework into a checklist of specific observable behaviors, and the synthesist applies those only against quoted evidence. The models it reads through:

• Big Five (OCEAN): the most validated model of normal personality, read at the facet level rather than as broad scores
• HEXACO: the sixth factor, Honesty-Humility, that the Big Five misses, which matters most for the influence layer
• Attachment theory: how a person relates in close bonds, read from relational behavior across multiple relationships
• Schema therapy: the early maladaptive schemas that recur across contexts, detected by repetition rather than single events
• The DSM-5 dimensional trait model: where a person sits on clinical trait continua, used to describe tendencies, never to assign a disorder
• Cognitive and decision style: how the person processes information, decides, and weighs evidence

A validity ladder governs all of it. The well-supported models outweigh weaker typologies like Enneagram and MBTI, which are used only as motivational hypotheses with their low validity stated plainly. When two frameworks disagree, the better-validated one wins and the profile says so.

05. Influence and Persuasion

The second layer reads how the person is wired around influence. It is the same knowledge whether you mean to persuade honestly or to recognize when someone is working on you, because there is no separate manipulation circuitry. Every technique runs through normal psychology, and what separates persuasion from manipulation is intent and transparency, not the mechanism.

Built from the verified profile, it produces four things: a susceptibility map rating the person against each principle of influence with the trait or schema that drives it, what actually persuades them when the approach is honest, the points where they are hard to move, and the handful of levers most likely to be used against them with the defense for each.

The difference between persuasion and manipulation is not the mechanism. It is whether it discloses or conceals.

It maps the lock and how locks work. It does not write the script to pick this person’s lock to harm them.

06. Where It Refuses

An accurate profiler can do real damage in the wrong hands, so the limits are built into the tool rather than left to good intentions.

• It does not diagnose. It describes tendencies on a spectrum, read from behavior. A categorical verdict of the kind a disorder name implies is off-limits, because diagnosis needs a clinician, history, and instruments this tool does not have.
• Crisis is routed, not analyzed. If the material surfaces self-harm or acute crisis, the tool stops profiling that thread and surfaces real resources, because a profiler is the wrong instrument for a person in crisis.
• Influence stays defensive. It maps susceptibilities and defenses, and it refuses to write a targeted plan to coerce a named, non-consenting person.
• Third-party reads stay calibrated. A profile from a handful of one-sided messages is labeled a sketch, not a verdict, and the limits of partial, context-stripped data are named.

These are not disclaimers bolted on at the end. They change what the tool will and will not produce, which is the difference between a responsible instrument and a merely clever one.

07. Outcome

Profiler is an evidence-bound analysis tool where fabrication is structurally hard rather than discouraged by instruction. The Iron Law, the extract-gate-synthesize separation, the two verification passes, and the calibrated confidence each close a different way a profile drifts from its evidence.

It is public and self-contained. All the framework knowledge lives inside the skill, it reads no private database, and it works on anyone from their writing or an interview. It began as a private tool that profiled a single person, and the generalization kept the anti-fabrication machinery while replacing the personal data sources with a general intake and a self-contained body of knowledge.

The honest framing is the point. It is an analytical aid that thinks out loud with its evidence shown, so a reader can check its work, rather than an oracle that hands down a verdict on who someone is.

It is not an oracle. It is an argument with its evidence attached, and you are meant to check it.