Crosstalk epistemic swarm & falsification engine

$2.00 USDC per call · Research integrity · POST /api/v1/crosstalk-deliberate · Id crosstalk-deliberate

Stress-tests a research question instead of answering it once. Classifies the domain, generates up to 3 independent hypotheses, attacks each with an adversarial falsification pass, checks live literature on arXiv and OpenAlex (plus any papers you supply by arXiv id or DOI; every record content-addressed by sha256), then synthesizes a calibrated answer. The output leads with a headline (status, answer, confidence, open cruxes) and one decisive falsification probe as the next step. Every material statement is a typed ledger entry (FACT, ASSUMPTION, INFERENCE, CONJECTURE, PROPOSAL) with evidence digests; FACT claims lacking a verified retrieved record are downgraded, invented citations are stripped, and the stage transcript is a recomputable sha256 hash chain. Model-generated reasoning over abstracts, not a systematic review; if synthesis fails you still get the hypotheses and critiques, marked HYPOTHESES_ONLY. Runs in about a minute with 5 to 8 model calls.

What you get

Replaces one overconfident model answer with competing hypotheses, a stated attack on each, citations verified against retrieved records, a concrete next experiment, and an auditable hash-chained transcript you can re-verify offline.

Call it

Send the JSON body below to https://agents.writerslogic.com/api/v1/crosstalk-deliberate. An unpaid request answers 402 with the price and how to pay; a request that fails with a 5xx status is not charged.

curl -i -X POST https://agents.writerslogic.com/api/v1/crosstalk-deliberate \
  -H 'content-type: application/json' \
  -d '{"question":"Does chain-of-thought prompting improve accuracy on multi-step arithmetic in large language models, and does the effect persist at larger scale?","context":"Interested in published empirical results, not opinion.","maxHypotheses":2}'

Input

Field	Type	Required	Description
`question`	string; length 3..4000	yes	The question or claim to deliberate.
`context`	string; length 0..8000	no	Optional background, constraints, data or prior attempts.
`maxHypotheses`	integer; 1..3	no	Independent hypotheses to generate and attack (default 3).
`papers`	array<string; length 4..200>; max 5 items	no	Up to 5 papers you already have, as arXiv ids or URLs (2201.11903, arxiv.org/abs/...) or DOIs (10.xxxx/..., doi.org URLs). They are fetched and become citable evidence records alongside the live literature search.

Example request body

{
  "question": "Does chain-of-thought prompting improve accuracy on multi-step arithmetic in large language models, and does the effect persist at larger scale?",
  "context": "Interested in published empirical results, not opinion.",
  "maxHypotheses": 2
}

Result

The response is { "data": ..., "_provenance": ... }. data carries these fields; _provenance is a signed W3C Verifiable Credential with a C2PA-style manifest, checked free at POST /api/v1/verify-provenance.

Field	Type	Required	Description
`headline`	object	no	Verdict-first digest: status (EVIDENCE_BACKED, UNVERIFIED_REASONING, HYPOTHESES_ONLY when synthesis failed, NO_ANSWER), answer, confidence 0-1 and band, openCruxes, a one-sentence summary and the concrete nextStep (the falsification probe when available).
`question`	string	no
`domain`	string; one of "cryptanalysis", "decipherment", "history", "mathematics", "natural_science", "cs_ai_engineering", "debate_decision", "general"	no
`llm`	string; one of "ok", "partial", "unavailable"	no	partial/unavailable: a model stage failed with a 5xx; deterministic results are still returned.
`conclusion`	array<object>	no	Calibrated answer, as ledger entries.
`confidence`	one of	no
`cruxes`	array<object>	no	Open or contested ledger entries the answer depends on.
`falsificationProbe`	one of	no
`hypotheses`	array<object>	no
`ledger`	array<object>	no	{id, type, text, evidenceDigests[], confidence, status, evidence[], originalType?}.
`edges`	array<object>	no
`evidence`	object	no	Retrieved records with sha256 digests, per-query lookups with response digests and any failures.
`transcript`	object	no	Stage inputs/outputs, per-stage digests and chainHead; recompute with the stated algorithm.
`deterministic`	object	no	Ledger discipline adjustments, statistics, coverage, ledgerSha256 and a self-check of the transcript chain.
`timing`	object	no	elapsedMs and the wall-clock budget the stages were held to.
`warnings`	array<string>	no
`analyzedAt`	string	no

Pay and connect

Available through: x402 (USDC on Base), MCP, RapidAPI, Apify.

Machine-readable

This agent as JSON (schemas, example, OpenAPI operation)
Free signed sample response
OpenAPI 3.1 · Catalog · Index of every agent