The 20-year term policy lapses at member age 78 while the spouse will be 81 — if the member dies after 78, the spouse loses both the pension income and the insurance backstop. NYSTRS survivor elections are irrevocable, so this tail risk cannot be re-hedged later.
AI decision infrastructure · IAA §206 · SEC Marketing Rule
One model is a guess.
Five models is a decision.
A signed, timestamped, role-attributed rationale record for every AI-assisted recommendation. IAR-attributable. Built to file into a books-and-records workflow.
The SEC's 2026 Examination Priorities(released 2025-11-17) named adviser use of AI and the documentation of recommendations as focus areas. The Council produces a documented basis an adviser can use to support fiduciary review, books-and-records workflows, and supervisory review. Compliance remains the firm's responsibility.
144 Council deliberations runlive
Watch a sample advisory decision get rendered.
A live cycle on an irreversible pension-election question that hits an advisor's desk every retirement season. Four roles reason in parallel. The Chairman synthesizes. One timestamped, signed record results.
Anonymized synthetic scenario. For demonstration only. Not investment, tax, legal, or financial advice.
Input · advisor scenario
A retiring NYSTRS member (age 58, 32 years of service) is choosing the pension election: take the Maximum single-life benefit ($6,400/mo) plus a $750K 20-year term policy to protect the spouse — the pension-maximization play — versus the 100% joint-and-survivor benefit ($5,600/mo). The spouse is 61. The election is irrevocable. Asks what they're missing before locking it in.
Confidence85%
Agreement4 / 4
Dissent1 / 4
Preserved dissent · attributed to Contrarian
The Maximum-plus-term structure only outperforms if the member dies inside the 20-year window; if both spouses live past member age 78 — the statistically likely path — the household has surrendered $800/mo of guaranteed lifetime survivor income.
Hover to pause · 18-second cycle · loops
Internal evaluation · sample of 20 · April 2026
How the Council compares to a single-model baseline.
On 20 founder-judged anonymized advisory scenarios, the Council produced at least one material risk the single-model baseline did not, in each scenario reviewed. Single rater. Sample designed by the founder; baseline = GPT-5.4-pro single-shot on the same scenario text and response schema.
Internal evaluation only · not third-party validated · not predictive of any other scenario, advisor, or client · not a representative sample of all advisory situations · methodology at /methodology; raw responses available under NDA.
For VCs and design-partner advisors evaluating Teranode for pilot. Not investment advice. Not a marketing communication of any registered investment adviser. Direct: founders@teranode.ai
Try it
What scenario do you want to run through the Council?
Just describe your decision in plain words. Everything else on this page is optional — add it only when you want an even sharper record.
The Carusos · NYSTRS pension election
NYSTRS Tier 4. Member retiring at age 58 with 32 years of service. Maximum single-life benefit $6,400/mo; 100% joint-and-survivor benefit $5,600/mo. Spouse age 61, good health. Pension-maximization plan: take the Maximum and buy a $750,000 20-year term policy (~$3,200/yr) to protect the spouse — the policy lapses at member age 78. The election is IRREVOCABLE after the filing deadline.
The question · We're leaning toward the Maximum single-life benefit and buying a $750K 20-year term policy to protect my spouse — the pension-maximization play. The election is irrevocable. What are we missing before we lock it in?
Fictional New York composites for illustration of the Council's reasoning — not real clients, not investment advice, and not recommendations. Edit any figure to fit your own situation. Teranode does not provide investment advice.
See it work
New here? Tap any example below to load it instantly — or write your own in the box.
Hypothetical examples for illustration of the Council's reasoning process. Not investment advice, not recommendations, and not based on actual client communications. Teranode does not provide investment advice. Past results of the Council's analysis do not predict future outcomes.
2 · Run it
↑ Type your decision above — or tap an example — to begin.
Takes ≈ 45s. No signup.
For demonstration only. Not investment, tax, legal, or financial advice.
01 · what it is
AI decision infrastructure for regulated financial advisors.
Teranode is the rationale-record layer for advisory workflows that have to be defensible. Every Council run produces one signed, timestamped, role-attributed decision record an advisor can put on file — the documented basis IAA §206 and books-and-records retention rules (204-2) already require.
Built by five independent reasoning roles — Risk, Builder, Strategist, Contrarian, and Chairman. The Chairman synthesizes the record; the Contrarian's dissent is preserved on it; the IAR signs it.
02 · example decision record
A sample advisory decision — anonymized.
An anonymized synthetic example: a client choosing between money markets, a guaranteed-interest annuity, and a term annuity structure. Here is the record the Council produces.
input
Client E is deciding whether to remain in money markets, purchase a guaranteed-interest annuity, or use a term annuity structure.
Riskrisks · failure modes
Primary risk is locking up liquidity for incremental yield without confirming emergency cash and near-term spending needs.
Builderstructured plans
Create a side-by-side comparison: money market liquidity, GIA guarantee, term annuity rate/term, surrender schedule, tax treatment, and worst-case exit scenario.
Strategistsecond-order
If the client's objective is principal preservation and predictable income, a partial allocation is more defensible than an all-or-nothing move.
Contrarian DISSENTpressure-test
Remaining in money markets is rational when liquidity, simplicity, and optionality matter more than locking in a higher stated rate.
Final Recommendation (example output)
Do not reallocate out of money markets until a documented comparison confirms alignment across liquidity, time horizon, and tax factors.
confidence
74%
agreement
3 / 4
dissent
1 / 4
preserved dissent
Contrarian retained on record: money markets remain viable when liquidity and optionality dominate yield.
status
Awaiting advisor sign-off.
fp · sha256:7e2c…91b2026-04-24 · 14:02 UTC
Hypothetical example for licensed-adviser review. Not investment advice and not a recommendation to any client or prospective client. The Council produces a decision record an adviser reviews and signs off on — it is not direct advice.
live · reliability graph
Step inside the data.
The Council's eval history, rendered as a live system you can explore: every role, every model, every run.
Backed by the live eval_runs table and model-policy.json.
liveStep inside the data03 · why it matters
The window to build this is now. Not in two years.
01
Single-model answers are not auditable.
No attribution, no preserved alternatives, no record of the reasoning that produced the output. You can't defend what you can't show.
02
Regulators are moving on a fixed timeline.
Firms that don't have documented AI supervision processes by the time rules land will be building them under enforcement pressure — not ahead of it.
- Apr 2025FINRA AI guidance issued
- Aug 2026EU AI Act high-risk classification enforced
- pendingSEC predictive-data-analytics rule
03
The Council produces one defensible record.
Timestamped, signed, with confidence and preserved dissent — the artifact compliance files, advisors review, and clients see.
Single-model AI gives you an output. Five-model adversarial reasoning with preserved dissent gives you an artifact built specifically for a documented supervised review.
04 · decision record schema
Every record is structured.
The same JSON shape is emitted by every Council run — built to populate compliance systems, client portals, and downstream tooling without custom parsing.
{
"client_label": "Client E",
"scenario": "Cash vs guaranteed income decision",
"roles": {
"risk": "Primary risk is locking up liquidity…",
"builder": "Create a side-by-side comparison…",
"strategist": "Partial allocation is more defensible…",
"contrarian": "Remaining in money markets is rational…"
},
"recommendation": "Do not reallocate out of money markets until a documented comparison confirms alignment across liquidity, time horizon, and tax factors.",
"confidence": 0.74,
"agreement": "3 / 4",
"dissent": "1 / 4",
"preserved_dissent": "Contrarian retained on record: money markets remain viable…",
"fingerprint": "sha256:7e2c…91b",
"timestamp": "2026-04-24T14:02:00Z"
}05 · objections we expect
Three questions a sharp reader will ask.
Why not just fine-tune a single model?
Because a fine-tuned model produces an answer; the Council produces a record. An answer is a feature. A timestamped, signed, attributable record with preserved dissent is the artifact compliance can file. You can't defend a black box.
What if all five roles agree — isn't that overconfidence?
It can be, and the record makes that visible. Confidence is reported against structural agreement across the four reasoners (4/4 vs 3/4); the Chairman synthesizes, it does not vote. A unanimous record is logged as such, not flagged as more correct. The advisor still owns the decision; the Council's job is to make the inputs to that decision legible, not to manufacture consensus.
What data leaves the browser?
Your decision text is sent to a stateless Council function, processed in memory, and returned. By default, nothing is persisted server-side— only when you explicitly opt into “Help improve the Council” does the run land in our anonymized learning corpus. Shared records are encoded directly into the URL — the link is the data — so a record can move between two parties without an account, login, or shared backend. Full disclosure on /privacy.
pre-seed · raising now
Send your hardest decision. We'll send back the record.
Email a meeting summary, an investment question, or any high-stakes tradeoff on your desk this week. You'll get a structured Council record back inside 24 hours.
Free during pre-seedNo sales callNo follow-up unless you want one