Solstice
Strategic Intelligence
Seven adversarial engines attack your strategy from every dimension. Intervention modeling shows which levers move the needle. Board-ready intelligence in minutes, not months.
PII redaction • Audit logging • No model lock-in • Export MD, JSON, HTML, PDF, PPTX
Each engine attacks the problem from a different analytical dimension. When multiple engines converge on the same conclusion, confidence compounds. Findings carry engine provenance — you know what produced every insight.
Survivability and advantage trajectories with confidence envelope, threshold bands, inflection markers, and adversary pressure attribution.
Branch futures from any finding. Counterfactual ghost lines show which interventions materially shift survivability — and which don't.
Margin compression, EBITDA impact, partner churn, compliance cost — structural risk translated into the language of budget authority.
100K+ perspective synthesis. When diverse viewpoints agree, confidence rises.
Multi-round adversarial war-gaming with attack/defense scoring and judge verdicts.
Surfaces hidden premises the strategy depends on. If these break, so does the plan.
Maps what each actor actually wants. Follow the money, follow the power.
Second-order effects. What breaks downstream when you pull a lever.
Rare but plausible tail events that reshape the entire landscape.
Pattern-matches against structural analogues from prior market cycles.
Risk x likelihood grid of hidden premises. Stress-test each one.
Stakeholder power/interest mapping with alignment scores across actor coalitions.
Effect chains with feedback loops and critical path markers.
Tail risk scenarios with cascade propagation and structural consequences.
Pattern-matched precedents with structural similarity scoring.
Multi-perspective convergence matrix across diverse analytical frames.
Outlier detection across engines. The signals everyone else missed.
This is not a chatbot. It's a simulation workspace that produces decision artifacts. Each run is traceable: engines engaged, assumptions tested, risks surfaced, and why the verdict moved.
The Aggregate Exposure Surface translates structural pressure into the financial language executives use for budget authority. Every simulation produces a capital risk profile.
Margin Compression
0%
Projected range
EBITDA Impact
High
Earnings pressure
Partner Churn
Moderate
Coalition stability
Compliance Cost
Accelerating
Regulatory burden
Built for high-stakes decisions where second-order effects and adversarial response matter. Compresses weeks of analysis into a run you can interrogate.
Model retaliation, narrative capture, channel fragility, and coalition formation before launches or pricing moves.
Stress-test transparency constraints, compliance shocks, and policy shifts that structurally compress margins.
Run cascading failure scenarios and identify irreducible vulnerabilities vs mitigations across adversarial phases.
The product produces structured intelligence briefs in the same formats used by top-tier strategy firms and defense planners. The difference: minutes, not months.
Every completed run generates a full investor brief — exportable as interactive HTML, print-ready PDF, branded PowerPoint deck, raw Markdown, or structured JSON. Same intelligence, five formats, zero reformatting.
MD, JSON, HTML, PDF. Board-ready verdict + risk surface + exposure.
Full report, one-pager, or slide deck. Situation → Complication → Question → Answer.
Executive verdict, full analysis, or round-by-round battle log with judge reasoning.
10-slide branded deck. Verdict, risks, exposure, interventions — boardroom ready.
Auto-discovers every convergence/crucible experiment pair, then runs five strategic lenses across every domain — and across all domains simultaneously. Adversarial challengers stress-test the output before you ever see it. 300,000 underlying perspectives. ~28 LLM calls. One unified brief.
Auto-scan experiment directory for convergence/crucible JSON pairs. Parse into structured digests with worldview summaries.
Per-domain strategic extraction — validated findings, contested claims, destroyed hypotheses, funding implications.
5 lenses × N domains + 5 cross-domain syntheses. Each lens produces per-domain and cross-domain intelligence.
3 adversarial challengers tear the strategy apart. Neutral arbiter judges what survives.
Unified strategic brief — executive summary, meta-pattern, research priorities, investment implications, policy recommendations, risk register.
Concentration gaps, institutional capture, heterogeneity blindness, and overlapping overlooked angles across all experiments.
JSON + human-readable results. Portfolio allocation, risk-adjusted domain ranking, systemic reform priorities.
Every convergence finding gets tested against published research from PubMed, Semantic Scholar, ClinicalTrials.gov, and Europe PMC. Phases 1–2 run without any LLM — they work even when all API quotas are exhausted. Phase 4 actively tries to break the convergence answer.
Regex-based extraction of testable claims from convergence and crucible results — gene/protein names, pathway references, species mentions, and therapeutic keywords. Classifies each claim as mechanism, therapeutic, species, or epidemiological. Builds 3 PubMed search queries per claim: broad, high-quality evidence, and negation/falsification seed. Zero LLM calls.
Searches 4 free biomedical APIs in parallel with rate limiting and deduplication. PubMed (XML fetch with MeSH terms and publication type classification), Semantic Scholar (citation counts and external IDs), ClinicalTrials.gov (phase classification, completion status), and Europe PMC (full-text, preprints). Cross-API deduplication by DOI and PMID with data merging.
LLM classifies each paper as SUPPORTS, CONTRADICTS, or NEUTRAL relative to the claim. Falls back to keyword heuristics if LLM is unavailable. Computes an evidence pyramid score weighted by study type (systematic review = 1.0, RCT = 0.85, cohort = 0.55, preprint = 0.25), citation count boost, and recency decay. Flags overclaimed, underclaimed, contradicted, and no-evidence claims.
Actively tries to break the convergence answer using two strategies. Negation search: constructs queries designed to find papers that disprove the convergent mechanism. Clinical trial reality check: compares therapeutic claims against actual trial outcomes (completed vs terminated). Rates contradiction strength: DEVASTATING, MODERATE, WEAK, or NONE. Only claims that survive falsification are trusted.
Generates full HTML evidence reports with claim-level confidence deltas, evidence pyramid scores, supporting and contradicting paper citations, falsification results, and overall grounded confidence. The output shows exactly how AI-generated convergence compares to published literature — where it was right, where it overclaimed, and where real evidence is stronger than expected.
MeSH-classified full-text search via NCBI E-utilities. Publication type mapping to evidence pyramid levels.
Citation-count enrichment and cross-reference resolution. Academic impact weighting.
Phase classification, completion status, and therapeutic claim reality checks against actual trial outcomes.
Full-text access including preprints. Broader European literature coverage and cross-API deduplication.
Systematic reviews and meta-analyses weighted highest (1.0). RCTs (0.85), cohort studies (0.55), case reports (0.30), preprints (0.25). Citation count and recency modifiers.
Gemini → OpenAI → Anthropic fallback chain ensures analysis completes even when individual providers hit quota limits. Phases 1–2 need no LLM at all.
Compares AI convergence confidence to literature-grounded confidence. Flags when AI overclaims (Δ < −20%) or underclaims (Δ > +20%) relative to published evidence.
We run one scenario end-to-end: survivability trajectory, irreducible vulnerabilities, intervention modeling, financial exposure, and board-ready brief export. If you're evaluating Solstice, this is the fastest way to understand the moat.