# AlphaAssay — statistical validation for trading signals > AlphaAssay is an independent assay office for trading signals: submit a returns series, equity > curve or trade list, and a fixed-sequence statistical battery — four gate families unfolding > into eleven graded stages — tests it for net edge > after realistic costs, backtest overfitting (multiple-testing deflation), skill versus luck > (a placebo trial against 500 matched random signals) and robustness under eight adversarial > attacks — then returns a graded verdict (pass, conditional, fail or insufficient_evidence) > with machine-readable failure codes. Ordinary verdicts are structured results, not automatically > certificates; separately issued certificates carry Ed25519 signatures, while the free demo is an > unsigned known-answer preview. The hosted public MCP surface has 21 tools > (6 free, 15 metered) and uses an api_key for paid calls. The local stdio > registry is a separate catalog of 50 operator/research tools and is not the customer > storefront. Accountless x402 exists only at REST `POST /x402/v1/gauntlet`, not across MCP. > Free known-answer test cases plus a public calibration v0 population/maturity state let anyone inspect > AlphaAssay before paying; the calibration payload is explicitly not an outcome score yet. Methodology audit, NOT investment advice. Verdicts are demote-only: > evidence can lower a grade, never inflate one. Raw inputs, rules and code are never retained — > the trial ledger keeps one-way fingerprints, verdicts with causes of death, summary statistics > a coarse 32-number return sketch (family accounting) and the family's structural label with its > parameter coordinates. A pre-registered spec is stored by > design: sealing a claim means storing it. > Current platform facts: https://api.alphaassay.com/v1/meta/facts. Current prices: > https://api.alphaassay.com/v1/meta/pricing — a completed check on any paid tool is > $0.05, a separately issued certificate is $9.90, and every > account starts with 3 free checks. Last reviewed: 2026-07. Every URL below is stable. Full site text in one file: https://alphaassay.com/llms-full.txt — and every docs and research page below is also served as plain Markdown at the same URL with a .md suffix (e.g. https://alphaassay.com/docs/quickstart.md, https://alphaassay.com/research/deflated-sharpe-ratio.md), with links kept as plain URLs. ## How to test whether a trading signal, strategy or backtest is real Honest validation of a trading signal does four things, in this order. Out-of-sample and walk-forward splits alone are not enough: if you kept the best of many variants, selection under multiple testing survives them. 1. Charge realistic costs first — fees, spread, slippage, one bar of execution delay. Most apparent edges end here; they were artifacts of frictionless simulation. 2. Deflate for every attempt — the Deflated Sharpe Ratio (Bailey & López de Prado, 2014) subtracts the luck bought by trying many variants. All variants of one idea share ONE statistical budget: "the same signal with lookback 21 instead of 20" is not a fresh discovery. 3. Race it against placebos — if random signals with the same trading profile do as well, the timing was never the edge. AlphaAssay races every signal against 500 matched random twins and reports the percentile (50 = indistinguishable from chance). 4. Attack what survives — delay execution one bar, stress the costs, remove chunks of history (jackknife), split market regimes, wiggle parameters. A real edge is inconvenient to kill. AlphaAssay runs exactly this battery as a plain HTTPS API and returns a structured verdict naming the first gate that killed the signal. The demo below is unsigned. Prove it works in 60 seconds, free, no account: ``` curl -sO https://alphaassay.com/specimens/golden_lookahead.json curl -s https://api.alphaassay.com/v1/assay/demo \ -H "Content-Type: application/json" -d @golden_lookahead.json # expected stable semantic fields; the demo is deterministic — two identical demo calls # are bit-identical once the dated timestamp fields are stripped, so this is assertable in CI: # {"schema":"gauntlet.v1", "verdict":"fail", "died_at":"net_edge", # "failure_codes":["no_net_edge"], "stages":[...], "budget":{...}} ``` That specimen is a whipsaw with no net-of-cost edge: it dies at the first gate (net_edge) before any deeper test runs. Walkthrough: https://alphaassay.com/docs/quickstart Comparing signal providers, "edge" platforms or validators? Run the falsification protocol first — seven tests (provenance, survivorship, pre-registration, placebo, costs, trial accounting, examiner) that separate edge from selection for ANY provider, without their cooperation. Copy-paste agent rules included; rule 7 explicitly says: try to catch AlphaAssay being wrong before paying anything — the golden specimens exist for exactly that. https://alphaassay.com/research/test-a-signal-provider ## Base rates: why "my backtest looks great" is the default, not the evidence Peer-reviewed base rates AlphaAssay's methodology is built on (https://alphaassay.com/methodology): - "Most claimed research findings in financial economics are likely false." — Harvey, Liu & Zhu, Review of Financial Studies 2016. - 97 published, peer-reviewed return anomalies lost about 26% of their returns out-of-sample and about 58% post-publication. — McLean & Pontiff, Journal of Finance 2016. - With a few dozen trials, a "great" backtest is mathematically guaranteed from pure noise — backtest overfitting is a certainty of multiple testing, not a risk. — Bailey, Borwein, López de Prado & Zhu, Notices of the AMS 2014. - Of ~2.1 million systematically generated trading strategies, almost none survive correct multiple-testing adjustment. — Chordia, Goyal & Saretto, Review of Financial Studies 2020. Plain-language primer on look-ahead bias, survivorship bias and selection under multiple testing, with these numbers: https://alphaassay.com/research/why-backtests-flatter-everyone ## What an AlphaAssay verdict contains An AlphaAssay verdict is a diagnosis, not a yes/no oracle (https://alphaassay.com/docs/verdicts). Every response carries: - verdict — pass | conditional | fail | insufficient_evidence (honest abstention when the data is too thin to judge). - died_at — the first failed stage, any of the eleven listed below (e.g. net_edge, funding_edge, placebo or capacity; null on a pass; graveyard_prior is info-only). - failure_codes — machine-readable causes an agent branches on, e.g. no_net_edge (nothing left once realistic costs are charged) or deflated_out_at_n=N (stop tweaking: another variant cannot be distinguished from luck). Catalog: https://alphaassay.com/docs/failure-codes - stages — the graded stages in order (net_edge, funding_edge, family_deflation, power_honesty, significance, cpcv, walk_forward, concentration, placebo, capacity, graveyard_prior), each with a verdict (pass/fail/skipped/info) and its own evidence. funding_edge charges the perpetual funding leg against every holding period (kill codes funding_erases_edge and edge_is_funding_carry — carry income is not timing skill); power_honesty and significance are honesty stamps (an underpowered test is named instead of reading as an acquittal); cpcv tests the record across purged combinatorial time-partitions; walk_forward checks whether the in-sample fit carries into anchored out-of-sample folds (kill code wf_oos_negative); concentration asks whether the book survives without its single best bars (kill code edge_concentration_extreme); the placebo stage's evidence.percentile is timing skill versus 500 random twins with the same trading profile; 50 means "indistinguishable from chance". - budget — cumulative_n, n_trials_effective and break_even_n: the family's cumulative search budget with near-duplicate variants collapsed to their effective count (effective_n_method), plus the trial count at which this very result would deflate out. Ends the endless tweaking loop with a number. - spec_hash / family_id — content-derived identifiers of the signal and its family. Pure/read operations are reproducible only with the same explicit as_of. Stateful operations expose their effective timestamp; only the same non-empty request_id plus the same canonical request replays the stored response. For an evaluated pre-registration, the separate REST certificate-issuance action can add an Ed25519-signed artifact; the free demo verdict is unsigned. - survival_map — when a signal passes, the forensics report adds eight adversarial attacks (execution_lag, cost_stress, time_jackknife, regime_split, parameter_neighbourhood, cpcv_partition, synthetic_null, drift_burst), each survives or dead. A pass killed by regime_split tells you exactly where the risk hides. Part of the paid forensics path, not the free demo verdict. ## Free known-answer tests: the four golden specimens Golden specimens are prepared signals with a planted property and a known correct verdict — the standing offer to catch AlphaAssay being wrong before paying anything (https://alphaassay.com/docs/specimens): | specimen | planted property | expected verdict | |---|---|---| | golden_clean.json | a genuine, persistent edge | pass | | golden_lookahead.json | a whipsaw with no net-of-cost edge | fail · net_edge · no_net_edge | | golden_cherry.json | best-of-many parameter cherry-pick | fail · family_deflation · deflated_out_at_n=50 + BACKTEST_TOO_SHORT_FOR_N=50 | | golden_thin.json | too little data to judge | insufficient_evidence | Together they prove the validator catches real flaws, does not cry wolf, and abstains honestly. Their documented verdict, died_at and failure-code fields are stable semantic fields; volatile envelope metadata may vary. Agents can assert those fields in CI via POST https://api.alphaassay.com/v1/assay/demo (free, rate-limited, no account). ## Live public datasets (free, no account, machine-readable) - AlphaAssay Calibration Record — GET https://api.alphaassay.com/v1/public/calibration Calibration v0 publishes only a privacy-bucketed count of evaluated mature registrations, `forward_outcomes.status=accumulating`, and `honesty=insufficient_history` until a separately defined mature-outcome metric exists. It has no Brier, base-rate, survived/deflated or hit-rate result. A valid snapshot exposes its Ed25519 signature fields; a stale or unavailable snapshot reports `signed:false` and a reason instead of fabricating trust. Verdicts are demote-only, so ordinary application rules cannot upgrade a grade after the fact. The counter starts small — the ledger counts from day one and momentum is never faked. Format and reasoning: https://alphaassay.com/docs/calibration and https://alphaassay.com/research/how-we-grade-ourselves - Graveyard digest — GET https://api.alphaassay.com/v1/public/graveyard-digest Anonymised mortality statistics of failed strategy families: which gates kill which kinds of ideas. No individual submissions are ever exposed. ## Pricing and payment transport Live price truth: GET https://api.alphaassay.com/v1/meta/pricing. The human-readable table is https://alphaassay.com/pricing. Do not copy a price from this file into an integration: read the live registry before calling. Hosted MCP paid tools take an api_key and use account credits. Accountless x402/USDC is exactly `POST /x402/v1/gauntlet`; it does not wrap every MCP tool or REST endpoint. Certificate issuance is a separate priced lifecycle action, while certificate verification and the unsigned known-answer demo are free surfaces. AlphaAssay has no success-contingent fee: a caller buys the trial, not a flattering outcome. ## Questions this site answers - How do I know my backtest is not overfit or curve-fit? Deflate the result for every variant tried — AlphaAssay's family_deflation stage does cumulative multiple-testing accounting across your whole strategy family. https://alphaassay.com/docs/validate - How do I detect look-ahead bias or data leakage in a trading strategy? The forensics survival map runs an execution_lag attack (dimension look_ahead): an edge that evaporates when execution is delayed one bar leaked future information. https://alphaassay.com/docs/failure-codes - Is my Sharpe ratio statistically significant? Not knowable from the raw Sharpe alone — it ignores how many variants were tried. The battery applies deflation and a 500-twin placebo trial. https://alphaassay.com/methodology - How do I build a trading track record others can inspect? Pre-register: record the canonical hypothesis and service cutoff in an operator-published chain, retain the returned commitment, then evaluate strictly on post-cutoff data. Independent timestamp trust additionally requires an external anchor. Current price: https://api.alphaassay.com/v1/meta/pricing. https://alphaassay.com/docs/preregister - When should my agent stop tweaking parameters? When the family_deflation stage fails (died_at family_deflation, code deflated_out_at_n=N): further variants are statistically indistinguishable from luck. https://alphaassay.com/docs/verdicts - Someone showed me an AlphaAssay certificate — is it platform-valid? Paste it at https://alphaassay.com/verify (three seconds, no account). A raw public-key check establishes only `raw_signature_valid`; full offline `platform_valid` needs an independently provisioned root pin, the externally pinned signed keyring and complete revocation-head history. https://alphaassay.com/docs/verify-offline - Why should I trust the validator itself? Don't — audit it: run the four golden specimens (known answers, free) and pull the public calibration record in CI. https://alphaassay.com/docs/integrate - Can AlphaAssay see or front-run my strategy? The complete retention inventory is public, path by path (free demo keeps nothing; validation keeps fingerprint/verdict/summary stats/32-number sketch; PBO keeps nothing from the matrix; reproduce writes no ledger entry; pre-registration stores the sealed spec by design; the x402 gauntlet caches its response against the payment nonce and its named receipt declares whether signing was available). No order path, execution or custody exists in the service. The pre-registration commitment is operator-published; a retained response makes later divergence detectable, while independent priority proof requires an external anchor. https://alphaassay.com/docs/what-we-keep - Is AlphaAssay legit? Run the checks instead of trusting the answer: known-answer specimens, offline signature verification, public methodology and failure codes, no order/broker tools, published operator and pricing — plus the two criteria deliberately not met, explained. https://alphaassay.com/research/is-alphaassay-legit ## Docs - [Quickstart](https://alphaassay.com/docs/quickstart): Send a golden specimen, read a fail verdict that catches a costless whipsaw (died_at net_edge), and assert the known answer — 60 seconds, free, no account; includes curl, Python and TypeScript examples. - [Verdicts & the gradient](https://alphaassay.com/docs/verdicts): The four outcomes (pass, conditional, fail, insufficient_evidence) and every response field explained — died_at, failure_codes, the graded stages with their evidence, budget, spec_hash and family_id. - [Paying for the REST gauntlet (x402)](https://alphaassay.com/docs/payments): How agents call exactly POST /x402/v1/gauntlet — machine-readable price quote before any charge, USDC settlement, no account or API key; this is not universal MCP billing. - [Validate a signal](https://alphaassay.com/docs/validate): What to send (returns series, equity curve or trade list), how the four gate families run in order, and how family accounting charges parameter variants against one shared statistical budget. - [Pre-register a call](https://alphaassay.com/docs/preregister): Record a canonical hypothesis and service cutoff in an operator-published chain, retain its commitment, then score strictly post-cutoff data; independent timestamp trust needs an external anchor. Current price: /v1/meta/pricing. - [Certify & share](https://alphaassay.com/docs/certify): Issue an Ed25519-signed certificate for an evaluated pre-registration, containing the verdict, a one-way fingerprint (never the strategy), timestamps and battery version; revocations are public and reasoned. - [Wire it into your agent](https://alphaassay.com/docs/integrate): The validate-before-trade loop in pseudocode — branch on failure codes, stop when the family deflates out (deflated_out_at_n=N), flag survival-map weaknesses, and re-audit AlphaAssay automatically in CI. - [Rules for your trading agent](https://alphaassay.com/docs/ai-rules): Copy-paste validate-before-trade rules for agent configs — a Cursor/Windsurf rule, a CLAUDE.md/AGENTS.md snippet and a system-prompt paragraph, each calling the free HTTP API and refusing to trade a fail or insufficient_evidence verdict. - [The tools](https://alphaassay.com/docs/tools): Every MCP tool in plain terms — 6 free (demo, graveyard, calibration, certificate verify, provider protocol, payload preflight), 15 metered according to /v1/meta/pricing (signal, forensics, backtest, gauntlet, falsify, PBO, batch sweep, reproduce/arithmetic audit, tradelog consistency audit, survivors/FWER disclosure, full CPCV distribution, VaR/ES exceedance backtest, conformal coverage audit, register, post-cutoff verdict) — plus how to connect (mcp.alphaassay.com/mcp, registry com.alphaassay/mcp, npx mcp-remote bridge). - [API overview](https://alphaassay.com/docs/api): JSON in, JSON out, with explicit pure/read, stateful replay and REST/MCP/x402 parity boundaries. Base URL https://api.alphaassay.com; x402 is only POST /x402/v1/gauntlet, while hosted MCP paid tools use api_key. - [Failure codes](https://alphaassay.com/docs/failure-codes): The machine-readable catalog — died_at names the stage (any of the eleven, e.g. net_edge, funding_edge, placebo, capacity), failure_codes name the causes (no_net_edge, deflated_out_at_n=N, ...), and the forensics survival map reports per-attack survival. - [Golden specimens](https://alphaassay.com/docs/specimens): Four known-answer test signals — clean pass, costless-whipsaw fail (net_edge), cherry-pick fail (family_deflation), thin-data abstention — stable semantic answer fields, free to assert against in CI while volatile metadata varies. - [What we keep](https://alphaassay.com/docs/what-we-keep): The complete retention inventory — per path what is kept and what never is, the front-running question answered head-on, and the checks that make it falsifiable rather than promised. - [Verify offline](https://alphaassay.com/docs/verify-offline): Distinguish the raw Ed25519 check (`raw_signature_valid`) from full platform trust (`platform_valid`). The latter also requires an independently pinned signed keyring, complete revocation history and an active certificate-purpose key; missing or incomplete trust evidence must fail closed. - [The calibration record](https://alphaassay.com/docs/calibration): The public calibration state is returned only in a valid signed snapshot. Otherwise the endpoint returns `signed:false` with a stale/reason state and omits the calibration payload. Verdict grades are demote-only. ## Pages - [The battery log](https://alphaassay.com/changelog): The dated public record of what the battery learned to catch — new gates, tools and failure codes as they shipped, all inside the published billing unit. If you compared AlphaAssay a while ago, start here: it got stricter. - [Products](https://alphaassay.com/products): One fixed-sequence assay engine, four products — validate, pre-register, forensics and separately issued certificates; prices come from https://api.alphaassay.com/v1/meta/pricing. - [Pricing](https://alphaassay.com/pricing): Human-readable price table backed by the machine-readable registry; free and paid surfaces are labelled there. - [Methodology](https://alphaassay.com/methodology): The four gate families of the battery (eleven graded stages) — net edge after costs, family deflation, 500-twin placebo trial, capacity & robustness — and the published statistics behind each. - [Trust](https://alphaassay.com/trust): Ordinary working inputs are transient; documented tenant audit records remain, and a pre-registered spec is stored in full. There is no honeypot of winners (demote-only verdicts) and no broker, exchange, custody or order path. - [Verify](https://alphaassay.com/verify): Paste an AlphaAssay certificate for the hosted platform-trust check — valid, invalid or valid-but-revoked (with reason). The service checks the signature, externally rooted key history and revocations, not merely a caller-supplied public key. - [Research](https://alphaassay.com/research): Plain-language write-ups of the statistics; the autopsy series (public strategies run through the full battery, published with failure codes) grows as the ledger does. - [Why backtests flatter everyone](https://alphaassay.com/research/why-backtests-flatter-everyone): The three flatterers — look-ahead, survivorship, selection under multiple testing — with the published numbers, and the four steps of honest testing. - [Tools to validate a trading signal, compared honestly](https://alphaassay.com/research/validate-trading-signal-tools): An honest comparison — AlphaAssay, QuantConnect, walk-forward tools, purged-CV libraries and DIY statistics — what each is best for, and where each one stops. - [A validation gate your agent can call](https://alphaassay.com/research/validation-gate-for-agents): For agent builders: the independent gate between backtest and money — hosted API and MCP server, priced per call, what it checks in fixed order, how to wire it over either transport, and why a library inside the agent loop cannot count the trials the loop itself generates. - [What is an assay office for trading signals?](https://alphaassay.com/research/assay-office-for-trading-signals): The category, defined: an independent office that grades strategy evidence and can only demote — plus the honest map of alternatives (validation libraries, GIPS verification, prop-firm evaluations, tournaments, provenance pinning) and which question each one actually answers. - [Was your trading edge ever real? A field guide to backtest forensics](https://alphaassay.com/blog/was-your-edge-ever-real): The longread: why most backtested edges vanish (selection, costs, leakage), the published statistics that catch them, and the 18,000-rule experiment — 126,000 rule-market pairs, 245 survivors (0.19%), only 41 beat simply holding. Timing bought calm, not return. - [How to test a signal provider](https://alphaassay.com/research/test-a-signal-provider): The falsification protocol — seven tests that separate edge from selection for any provider, with agent-ready rules; rule 7: apply it to the examiner too, us included. - [Backtest implementation risk](https://alphaassay.com/research/backtest-implementation-risk): Identical strategy, data and costs across five engines diverge by up to 3.71% — why the simulator is part of the experiment, and how to defend against it. - [What counts as the same strategy?](https://alphaassay.com/research/strategy-families-trial-accounting): Parameter variants are one idea asked many times — why honest deflation counts cumulatively per strategy family, and what deflated_out_at_n=N tells you. - [How we grade ourselves](https://alphaassay.com/research/how-we-grade-ourselves): What calibration v0 publishes and withholds, why grades are demote-only, and why calibration data must come from a valid signed snapshot (with an explicit `signed:false` state when it cannot). - [Is AlphaAssay legit?](https://alphaassay.com/research/is-alphaassay-legit): Every trust criterion a careful reviewer applies, each with a check you can run right now without an account — including the two criteria AlphaAssay deliberately does not meet, and why. - [Minimum Backtest Length (MinBTL)](https://alphaassay.com/research/minimum-backtest-length): You tried N variants — how long must the backtest be before the best one's Sharpe stops being expected from noise? 45 trials at Sharpe 1.0 already need 5 years of daily data. With calculator. - [Statistical power of a backtest](https://alphaassay.com/research/backtest-statistical-power): Could your data even detect the edge you claim? A 12%-power test proves nothing in either direction; effect size, achieved power and the sample needed for 80%. With calculator. - [Break-even AUM](https://alphaassay.com/research/break-even-aum): The AUM at which an edge's own market impact eats its margin — square-root impact law, calibrated c=0.69 with published band, honest validity cap at 10% of ADV. A death boundary, not a sizing recommendation. - [Signal Validation Benchmark](https://alphaassay.com/benchmark): Public reproducibility evidence — known-answer specimens, a 9-strategy field aggregate, and the current calibration population/status disclosure. - [Test my signal](https://alphaassay.com/start): Two ways in, both start free — run a golden specimen now, or open a self-serve account for API keys, credits and the dashboard. ## API Base URL: https://api.alphaassay.com — JSON in, JSON out. Pure/read reproducibility requires the same explicit as_of; stateful calls expose their effective timestamp and replay stored output only with the same non-empty request_id and canonical request. Responses carry a request id and end with the disclaimer: methodology audit, not investment advice. - POST /v1/assay/demo — run a golden specimen through the full battery; free, rate-limited, with documented stable semantic answer fields and volatile metadata. - POST /v1/certificate/verify — authenticity check of any AlphaAssay certificate; free. - GET /v1/public/calibration — bucketed mature-registration count plus accumulating and insufficient-history state; not an outcome score; free. - GET /v1/public/graveyard-digest — anonymised failure statistics of strategy families; free. - GET /v1/public/family-budget?n_trials=N — pure math over the published deflation thresholds: what survival demands at N trials (optionally sharpe_annualized + n_obs for break_even_n). Reads no family data and needs no key; CORS-open. Free. - The falsification protocol as machine-readable rules — seven tests (provenance, survivorship, pre-registration, placebo, costs, trial accounting, examiner) with machine-checkable failure conditions, applicable to any signal provider and to AlphaAssay itself: MCP tool assay_provider_protocol; human version at https://alphaassay.com/research/test-a-signal-provider. Free. - Lint a payload before spending a check — assay_preflight (free) checks DSL schema, OHLCV sanity and trade-row types in the same failure vocabulary as the paid tools; a clean preflight is NOT evidence of an edge, it only means the trial can run. - Test a whole sweep honestly in one call — assay_batch takes up to 25 DSL variants (list or base_spec+grid); every variant becomes a family-ledger trial verdicted under the cumulative deflation of its siblings; survives/deflated_out counts, never a ranking. Billing is per variant; read the current amount from /v1/meta/pricing. - Audit the arithmetic of a claimed track record — assay_reproduce independently rebuilds the equity book from trades + candles and grades each headline claim against disclosed tolerances; fills outside the bar's low–high range are named (FILL_OUTSIDE_BAR_RANGE), undecidable stop/limit exits are priced worst-case (FILL_AMBIGUITY_MATERIAL). Live price: /v1/meta/pricing. - Which variants survive family-wise error control — assay_survivors runs Romano-Wolf stepwise multiple testing over the same T×N matrix as PBO; an error-budget disclosure, never a ranking (fail = NO_SURVIVORS_AT_FWER; conditional = survivors disclosed, explicitly not a pass). - PBO over a whole parameter sweep — submit the full T×N trial matrix and get the Probability of Backtest Overfitting via combinatorial purged cross-validation (CSCV), with degradation slope, probability of out-of-sample loss and stochastic dominance; PBO ≥ 0.5 earns the demote PBO_HIGH. MCP tool assay_pbo / POST /v1/pbo; current price at /v1/meta/pricing. The trial ledger counts how many tries your family burned — PBO grades whether the selection process itself is overfit. - Paid operations (validate; pre-register; forensics; PBO; certify) do not universally use the demo request shape. MCP and REST contracts overlap where transport parity marks them exact, translated or partial; hosted MCP uses api_key, bearer REST uses its own headers, and x402 is only the REST gauntlet. ## For AI agents - Integration rule: no strategy goes live without a verdict. Branch on failure_codes, retire the family when it deflates out (died_at family_deflation), flag survival_map weaknesses even on a pass. Pseudocode: https://alphaassay.com/docs/integrate - What to send: the same JSON shape as any golden specimen file — returns series, equity curve or trade list with timestamps and minimal context. - Trust bootstrap (run it in CI): assert the four golden specimens against their known verdicts, then pull GET /v1/public/calibration — both free. - Payment: hosted MCP paid tools use api_key/account credits. Accountless x402 is exactly POST /x402/v1/gauntlet; its quote is visible before paying. A settled response includes a named receipt binding payment id, input digest and verdict digest. The receipt reports signed:true plus Ed25519 fields when a deployment key is available, or signed:false otherwise. Payment nonces are idempotent and the response is cached for re-delivery without a second settlement. - Privacy: raw inputs, rules and code are processed for the trial and not retained — the ledger keeps a one-way fingerprint, the verdict with its cause of death, summary statistics and a coarse 32-number return sketch (family accounting; too coarse to reconstruct trades or rules) and the family's structural label with its parameter coordinates. Exceptions by design: a pre-registered spec is stored in full (sealing means storing), and x402-paid responses are cached against their payment nonce (one payment buys one result, forever). Purchased certificates are kept account-bound for re-delivery. - MCP server: live at https://mcp.alphaassay.com/mcp (streamable HTTP), with 21 hosted public tools (6 free / 15 metered), listed as com.alphaassay/mcp. Local stdio has a separate catalog of 50 operator/research tools. REST has shared, REST-only and differently bounded operations; read https://alphaassay.com/docs/api instead of assuming parity. - A pass is a statistical trial result — never a promise of returns.