Prosoche v2 · Adaptive Analytical Intelligence
The Prosoche Methodology
Apatheia Labs runs on Prosoche v2 — an adaptive analytical method. You bring a thesis or a question; it builds the fullest defensible picture around it and tells you what is known, what is contested, and what cannot yet be known. It has one shape: a spine of seven stages that always runs, a not-capturable coreit is bound to obey — the Method Constants Block — and a toolkit of operations selected to fit the task. The finish line is honesty, not accuracy: when the grounding will not carry a verdict, it abstains and says why.
The method began in adversarial-forensic work — reading a hostile institutional record against the grain, hunting the contradictions buried in volume. That origin is the why, not the what: the same discipline serves any field where a narrative is assembled to steer a conclusion — political investigation, scientific audit, a market read. Earlier versions (developed under the working names Systematic Adversarial Methodology (S.A.M.) and the Sovereign Analyst Framework) hard-coded that adversarial mode as the whole method; v2 demotes it from the spine to one selectable operation and builds the complete picture around the question instead. See the full story or explore the intellectual foundations.
The spine
The seven stages
This is the method: seven stages that always run, from framing a question to an honest release. Every analysis moves through them in order. What changes from task to task is not the stages but the operations selected inside them.
Stage 1 — Frame & self-locate
Fix the question — and a rival framing of it. Disclose your priors against an objective signal, tag the target type, draft the set of operations the task looks like it needs, and stamp the version of the constants block that will govern the run.
Stage 2 — Perspective inventory & contextual sweep
Census every viewpoint, stakeholder, and ownerless frame against a fixed roster-template floor. An absent but materially-affected party is a logged finding, never a silent drop.
Stage 3 — Gather, grade & bind
Collect the evidence, trace each load-bearing claim to a reproducible anchor, and grade every source for reliability before it is relied on. Score each finding for how much the question rests on it — then reconfirm the operations now that the material has been read.
Stage 4 — Symmetric scrutiny
One battery, at the same depth, on every perspective — including the analyst’s own prior, entered as row #0 and tested like any other. This is the always-on anti-capture pass. A view may be down-weighted only for a named, cited evidential reason, never for who holds it.
Stage 5 — Post the ledger
The report is three standing sections in fixed order: what is established, what is contested, and what is unknown. The verdict is barred by structure from standing as a headline — it is invalid without the per-item table behind it.
Stage 6 — Context-blind external check
An always-on independent floor that no party can switch off, scaling up on objective triggers. It can force a dropped perspective or operation back into the analysis before the method is allowed to assert.
Stage 7 — Determinability verdict & honest release
The release decision is computed from a fixed table over the findings that actually bear the question. Then the method releases — or abstains and names the next sharper question. An abstention is a result, not a failure.
None of this is a fixed pipeline. The seven stages always run; which operations run insidethem is selected to fit the task — and justifying that selection is the discipline. Those operations are the toolkit, below.
Two stances, one discipline
It argues your case, or it holds the center line
The stance is chosen when the question is framed (Stage 1). The same seven-stage spine and the same constants run underneath either one — only the entry posture differs.
Thesis-grounded
Build the case for a position and let it survive scrutiny. The method does not set out to defeat your thesis — but thesis-grounded is not thesis-assuming: the case is held to the same standard the method uses to expose obfuscation in the material, turned on itself. For advocacy- and publication-grade work.
Center-line
No privileged thesis. The question is only the way in; every perspective — including the analyst’s own prior, entered as row #0 and tested by the identical Stage-4 battery — is weighed by one rule, and the output is the fullest defensible picture and its distribution, not a side. The stance for not being captured by a viewpoint.
Governing principles
Two disciplines that guard both flanks
One guards against cynicism, the other against overconfidence. Together they keep a compelling story from standing in for a checked one.
Patterns are hypotheses, not conclusions
A compelling pattern is exactly how judgment gets misled — by accident, and by design. So every pattern is tested for provenance: is it a real recurrence across the fullevidence, or instances cherry-picked and labelled “a pattern” to steer toward a predetermined conclusion? The tells are consistent — a selective versus a complete set, base rates ignored, who benefits from the labelling, and what false impression it seeds. Naming a manufactured pattern is the blunt translation the method exists to perform, whether in a contested record, a political narrative, or a market story.
The unknown gets equal billing
What is not known is a first-class section of the report — Stage 5’s ledger gives it a standing column, not a gap left quiet. The method states what is warranted, what is uncertain, and what is genuinely unknown — and which kind: not-yet-found, unknowable, or contested. This is “done = honesty, not accuracy” made operational: an honest “we cannot yet know” outranks a confident answer that turns out wrong.
What keeps it honest
The Method Constants Block
The levers a motivated party would reach for are taken out of the investigation and fixed once, in the open, in a single versioned object — the Method Constants Block (MCB-1.0.0). It is authored before any material is read, stamped in every output, and never edited mid-run. Symmetry stops being a promise and becomes a mechanism. It has four parts.
REW Rule — holder-blind weighting
Evidence is weighted by its properties — establishment, provenance, corroboration, how much the question rests on it. Who holds a view, whether it flatters the analyst, and how loudly it is repeated are forbidden inputs. A down-weight with no cited evidential reason is a defect.
Roster Templates — the census floor
One fixed stakeholder template per target type, keyed to the abstract type and never to the real actors, so it can neither pre-name nor pre-drop a party. A template is a floor, not a ceiling: it can only add roles to check for. An absent but materially-affected party is a logged finding — and a missing one can force the verdict to withhold.
Calibration Constants — the confidence ladder
The confidence words are defined before any material is read — established, supported, contested, speculative, not-yet-established — with the independence test for what counts as corroboration. The bar cannot be moved mid-run to fit a preferred conclusion.
Determinability Table — the computed verdict
The release decision is a table over the load-bearing findings — withhold, indeterminate, insufficient, or supportable, ties breaking toward withholding. It is published with the per-item table behind it, so it can never stand as a lone headline and any reader can re-run it.
One roster template keys the census floor for each target type — single document · multi-source corpus · decision propagated through a chain · broadcast / persuasion / multimodal · contested public dispute — declared when the task is framed. Templates are floors, not ceilings: they can only add parties to check for.
One operation in the kit
The toolkit — and its most-developed operation
Inside the stages, the method selects operations to fit the task. Some always run — decomposition, the perspective census, grading, the symmetric-scrutiny pass. Others are conditional: a propagation trace when a decision was inherited through a chain; a historical-depth pass when when did this begin is load-bearing; a framing-and-language pass for broadcast or persuasion material; and the adversarial-scrutiny mode when a hostile or self-serving record must be read against the grain.
One caution on a shared word: the Stage-4 symmetric scrutinyabove is the always-on pass that tests every perspective — including the analyst’s own — at equal depth. The adversarial-scrutiny mode below is a different thing: a conditional operation, selected only when the material is hostile or self-serving.
The adversarial-scrutiny mode
The operation behind the published audits
The most-developed operation in the toolkit. Its own internal procedure runs nine phases — from an epistemic self-audit through source rating, a fallacy firewall, contradiction analysis, a propagation trace, and a constitutional adversarial review. It is the right instrument for reading an institutional record against the grain rather than accepting its account of itself. Earlier versions treated this mode as the whole method; v2 keeps it as one operation. The one relationship worth stating: the spine’s single Stage-4 scrutiny pass replacedthis mode’s former triple audit — it does not sit inside it.
Foundations
What it is built on
Four established disciplines underwrite the method — and it applies theirapparatus, not a model’s generic reasoning. Each analytical move is traced to its source in the canon (author, work, page); each finding to its source in the record (cited, checkable, graded). Each foundation below is named, traced to where it operates, and linked to the source that documents it — so the claim can be checked, not taken on authority.
NATO Admiralty Code
What it is
A two-axis intelligence rating that scores a source’s reliability (A–F) separately from a claim’s credibility (1–6). A1 is independently confirmed by a reliable source; F6 means neither the source nor the claim can be judged.
In Prosoche
Stage 3 (gather, grade & bind) rates every source A–F and every proposition 1–6, and the REW Rule ties a claim’s weight to that grade — which is what stops five outlets tracing to one weak origin from reading as five-fold corroboration.
Structured Analytical Techniques
What it is
The intelligence-community toolkit (Heuer & Pherson) built to defeat confirmation bias by externalising reasoning so someone other than the analyst can audit it — competing hypotheses, key-assumptions checks, structured adversarial review.
In Prosoche
The lineage of the symmetric-scrutiny pass (Stage 4) and of the adversarial-scrutiny operation — its steel-man, its council review, its integrity audit. Each is a structured check against an analyst settling on a narrative and then gathering support for it.
Formal Logic & Evidence Theory
What it is
Two adjacent disciplines: the classical apparatus of valid inference and its catalogue of fallacies, and the formal study of how evidence accumulates, conflicts, and updates belief — Bayesian inference, Dempster–Shafer, and legal evidence doctrine.
In Prosoche
The Fallacy Firewall within the adversarial-scrutiny operation catches invalid inference dressed as reasoning. Evidence theory is the tradition behind the MCB’s calibration ladder and the named-falsifiability tests — so absence of evidence is never quietly counted as evidence of absence.
Informational Cascade Theory
What it is
The economics of how a weak or false claim acquires authority: each actor rationally defers to the apparent consensus of those before them, so the error propagates while looking more confirmed at every step (Bikhchandani; the institutional-failure literature — Vaughan, Perrow, Reason).
In Prosoche
The problem the method was built for. The Propagation Trace — a conditional operation — is named for it: ANCHOR → INHERIT → COMPOUND → ARRIVE traces a propagated decision back to its origin and exposes the compounding steps the chain never audited.
This is a direction under active build. Prosoche v2 is promoted and wired into the engine, and a first set of operations runs today; much of the canon is named as method and not yet operable. What is described here is the frame and the discipline.
One boundary, stated plainly: Prosoche is a methodology, not a validated psychometric instrument. It can structure evidence, expose contradictions, state its confidence, and abstain when the grounding will not carry a verdict; it should not be read as clinical measurement or automated truth certification.