{
  "code": "ATS",
  "site": "AI Trading Signals",
  "domain": "https://bestaitradingsignals.pages.dev",
  "planned_domain": "https://bestaitradingsignals.com",
  "builder": "_ats_build.py",
  "lede": "Research AI signal claims through model versions, data availability, untouched tests and forward execution, with machine learning kept distinct from ordinary automation.",
  "directory_title": "AI research platforms and adjacent alert tools",
  "scope": "An algorithmic subscription is not automatically an AI or machine-learning product. No supplied evidence establishes Vector Ridge's ML architecture, training procedure or untouched validation, so this site does not name it a verified AI winner.",
  "research_scope": "Primary-document research dated 7 October 2026. We have not trained these vendors' models, operated their paid accounts, certified live profitability or collected a genuine user-rating sample.",
  "proof_note": "The stated commitment format is VR2|salt|family|ticker|direction|grade|entry|signal_time. It excludes model weights, preprocessing, stop, target and actual fills. A verified completed attestation bounds the covered bytes; a declared signal_time or pending submission does not independently establish their earlier existence. No actual receipt or private model run was independently reproduced for this edition.",
  "snapshot_note": "The older supplied model summary reports 690 signals, 70% wins and +1,227% raw for 2026, without the original cutoff, complete ledger, cost/size basis or audit scope needed here. The four supplied component outcomes total 1,226%, leaving a one-point reconciliation outstanding. This is not a newly measured October result, an AI holdout or subscriber-account performance.",
  "person": "The founder context has separate sources. The organiser lists Darren O'Neill fourth in 2025 WCTC Annual Forex at 168% and fifth in Q3 Forex at 65.9%. The supplied October Monthly first/59.35% needs its precise reference. The supplied 294% cross-division aggregate needs its method because those three figures add to 293.25%. Oxford Said MAFE and perfect quantitative GMAT are attributed credentials, with underlying records not supplied. None establishes a trained model's causal evaluation.",
  "channel_note": "Judge a named model or author, not an entire delivery channel. Chat feeds, scanners and marketplaces need their own retained forward records; no blanket fraud or performance verdict follows from the format.",
  "risk_note": "Model errors, data leakage, changing market conditions and execution costs can defeat an attractive backtest. This research is educational, not an automated investment recommendation.",
  "nav_label": "AI research directory",
  "library_paths": [
    "primers/index.html",
    "benchmarks/index.html"
  ],
  "old_images": [
    "images/og-image.png"
  ],
  "reviews": [
    {
      "slug": "quantconnect",
      "name": "QuantConnect",
      "category": "Algorithm research and model training",
      "fit": "A research environment for building your own experiments, not a vendor-certified profitable signal.",
      "core": true,
      "sections": [
        [
          "Visible expanded edition",
          "QuantConnect assessment Algorithm research and model training A research environment for building your own experiments, not a vendor-certified profitable signal. Earlier product descriptions were inspected 7 October 2026. This expanded edition adds public-source research and original educational cases on 13 October 2026. We have not trained these vendors' models, operated their paid accounts, certified live profitability or collected a genuine user-rating sample. Training capability QuantConnect documents model-training workflows. That is useful infrastructure for a researcher who must define features, labels and validation. It does not mean every algorithm run on the platform uses machine learning or has passed an untouched test. Freeze the experiment Retain data availability, model version, training end, label horizon and the first permitted test observation. Optimisation over a period makes that period part of development, not an untouched result. Compare simulation assumptions with the intended live route separately. No account verdict We have not trained or deployed a model on this platform for the assessment. Product functionality is not a measured edge. A user still needs leakage checks, declared costs, a forward record and operational monitoring before making a defensible performance claim. Sources QuantConnect model training documentation QuantConnect optimization parameters Compare the product categories Research infrastructure and selected experiment QuantConnect's training and optimisation references describe capabilities for user-authored research. They do not establish that every algorithm is a trained predictor or that an optimisation winner has an independent edge. Retain the selected code, data, transform, objective and run identity rather than reviewing the platform name as though it were one signal strategy. This page inspected documentation, not a paid workspace, trained market model, prospective vendor stream or actual account. Its role is a programmable research-environment comparison. A completed optimisation-history packet A fictional user compares twelve parameter settings on period P and exports the highest selected result. The packet also retains all twelve trials and P's viewed status. A later claim that the chosen setting passed an untouched test on P would conflict with that selection history. The correct review can describe a reproducible development comparison while requiring another declared evaluation object for the stronger claim. No real QuantConnect run, result or charge is supplied by this scenario. Training is not output ownership A training call, completed artefact and version used by a prediction are distinct transitions. Preserve their dates and bindings alongside the original output. A research LIVE label or repository merge does not prove deployed SHA, subscriber route or production certification. The source may provide a narrower manifest when weights are private, but a platform API reference cannot fill missing runtime evidence. Account fills and monetary outcomes remain another boundary beyond model training. The useful buyer comparison Compare input control, experiment exports, model/version ownership and documented operating responsibilities with the task you intend to perform. Confirm current data and compute terms directly. FreqAI and TrendSpider offer other documented research workflows, not measured substitutes in a common profit table. A buyer seeking only explanations needs a different packet from one seeking audited forecasts. This publication creates no paid capacity or trading route, and unknown causal evaluation remains unknown rather than a zero or positive provider score. Direct sources and reviewed scope Public documents, not private vendor experiments. The worked cases are invented; source dates do not establish a model-validation engagement. QuantConnect model training documentation QuantConnect optimization parameters scikit-learn probability calibration scikit-learn preprocessing and leakage Core consumer protocols Search all AI evidence pages"
        ]
      ],
      "sources": [
        [
          "QuantConnect model training documentation",
          "https://www.quantconnect.com/docs/v2/writing-algorithms/machine-learning/training-models"
        ],
        [
          "QuantConnect optimization parameters",
          "https://www.quantconnect.com/docs/v2/writing-algorithms/optimization/parameters"
        ]
      ]
    },
    {
      "slug": "freqai",
      "name": "FreqAI",
      "category": "Open-source machine-learning extension",
      "fit": "A documented ML workflow within Freqtrade; an implementation is not proof of an investable edge.",
      "core": true,
      "sections": [
        [
          "Visible expanded edition",
          "FreqAI assessment Open-source machine-learning extension A documented ML workflow within Freqtrade; an implementation is not proof of an investable edge. Earlier product descriptions were inspected 7 October 2026. This expanded edition adds public-source research and original educational cases on 13 October 2026. We have not trained these vendors' models, operated their paid accounts, certified live profitability or collected a genuine user-rating sample. What is documented FreqAI describes itself as a machine-learning extension for Freqtrade. Its framework can support user research, while the researcher remains responsible for the chosen features, labels, model and experiment definition. Versioned records Keep configuration and data identity with every training and evaluation run. Distinguish historical simulation, dry-run observation and actual orders. A new model identifier or altered feature set should not silently inherit an older experiment's validation claim. Safety boundary This page does not install a bot, supply trading credentials or deploy automated execution. No live profit or reliability test was conducted. Open-source availability enables inspection; it does not guarantee causal inputs, sensible position rules or favourable outcomes. Sources FreqAI introduction Compare the product categories Framework capability without a strategy badge FreqAI's originating documentation identifies an ML extension within Freqtrade. That supports a framework description, not proof that a selected configuration produces profitable or independent out-of-sample results. Retain the configured feature definitions, target, model artefacts, training windows and software version. Open-source inspectability can help a researcher ask precise questions while still requiring actual run records. This review installs no bot, shares no credentials and operates no live or dry-run trading account. A retraining identity example In a fictional review, configuration A generates an earlier series while configuration B changes a feature and training schedule. B must not silently inherit A's original evaluation label. The packet keeps both configurations, artefact IDs and producing-output bindings. A later reconstruction with B can be informative development research, but it cannot establish that the earlier prospective outputs came from B. Missing source data or runtime records limit the conclusion; they do not prove the framework or every user strategy failed. Data and execution layers remain separate Preserve actual feature availability and the complete prediction population, including missing or unresolved labels. A recorded prediction, a simulated order and a filled venue transaction answer different questions. A dry-run state cannot be upgraded to funded execution by a chart caption. A favourable historical research-fill window retains its own hypothetical status. Costs, collateral and real venue permissions need their actual definitions before a monetary comparison can be made. Costs and operating scope to confirm Free source code does not prove that data, computation, hosting or account implementation are free. Confirm the selected setup and what owns its normal operation before relying on its outputs. This publication does not create a personal-computer production dependency, managed bot or paid service. Compare FreqAI for the documented ML research task, with experiments and records at the required boundary. Its existence is neither an automatic edge guarantee nor a reason to invent an unsupported negative rating. Direct sources and reviewed scope Public documents, not private vendor experiments. The worked cases are invented; source dates do not establish a model-validation engagement. FreqAI introduction scikit-learn probability calibration scikit-learn preprocessing and leakage Core consumer protocols Search all AI evidence pages"
        ]
      ],
      "sources": [
        [
          "FreqAI introduction",
          "https://www.freqtrade.io/en/stable/freqai/"
        ]
      ]
    },
    {
      "slug": "trendspider",
      "name": "TrendSpider",
      "category": "Machine-learning and strategy research tools",
      "fit": "An explicitly described ML product to investigate, with results still dependent on the user's experiment.",
      "core": true,
      "sections": [
        [
          "Visible expanded edition",
          "TrendSpider assessment Machine-learning and strategy research tools An explicitly described ML product to investigate, with results still dependent on the user's experiment. Earlier product descriptions were inspected 7 October 2026. This expanded edition adds public-source research and original educational cases on 13 October 2026. We have not trained these vendors' models, operated their paid accounts, certified live profitability or collected a genuine user-rating sample. AI-specific product evidence TrendSpider's ML product description gives a concrete basis for discussing machine-learning tooling rather than inferring AI from a scanner label. Treat its advertised functionality as a product claim, not an independent return audit. Testing questions Request or retain the chosen model, feature definitions, training window and test history. Once a result has influenced a new rule, that result is development evidence. Keep later untouched or forward evaluation separate. Workflow boundary A strategy test, chart condition and delivered alert are separate stages. This desk has not subscribed, trained a model or compared live fills. Confirm current access and execution dependencies before assuming an advertised workflow applies to the selected plan. Sources TrendSpider machine-learning product description TrendSpider strategy tester documentation Compare the product categories Use the current product identity TrendSpider's current ML Quant Lab page and strategy support reference describe a model-building workflow. The older AI Strategy Lab wording is retained as historical context, not a separate independently tested vendor. Public documentation supports the presence of ML tools; it does not validate a user's chosen strategy. Save the selected feature, target, model and evaluation configuration rather than assuming a familiar interface label identifies one immutable predictor. A reviewed period becomes development evidence A fictional user examines a model's evaluation display, changes its feature set and studies the same period again. The revised choice was influenced by that period, regardless of whether the interface still names it test. The reviewer preserves the sequence and requests later independently declared evidence for a stronger claim. No actual TrendSpider model was trained or backtested here. A product's documented default process cannot prove that an arbitrary user's iterative choices preserved untouched evaluation. Output meaning before a confidence claim Keep the actual target and horizon beside any model score. A training-quality measure, a strategy outcome and a delivered instruction need separate identities. Do not translate a user-selected risk/reward target into a calibrated subscriber success probability without the corresponding records. A platform's ability to build a strategy is meaningful product evidence, while realised account fills, cost treatment and market permissions require a separate review. Workflow fit, not a profit league table Compare this model-building interface with the programmable and framework-based workflows documented by QuantConnect and FreqAI. Current package access, data rights and execution responsibilities belong in a direct supplier enquiry. No paid account or strategy trial was operated for this edition. The alternative profiles each have their own task and evidence boundary; none receives an invented common return rank. The free synthetic illustration explains metric distinctions without assigning its toy values to this product or any other vendor. Direct sources and reviewed scope Public documents, not private vendor experiments. The worked cases are invented; source dates do not establish a model-validation engagement. TrendSpider ML Quant Lab product description TrendSpider trained-strategy definition scikit-learn probability calibration scikit-learn preprocessing and leakage Core consumer protocols Search all AI evidence pages"
        ]
      ],
      "sources": [
        [
          "TrendSpider ML Quant Lab product description",
          "https://trendspider.com/product/machine-learning-quant-trading-strategy-lab/"
        ],
        [
          "TrendSpider trained-strategy definition",
          "https://help.trendspider.com/kb/ai-strategies/what-is-an-ai-strategy"
        ]
      ]
    },
    {
      "slug": "trade-ideas",
      "name": "Trade Ideas",
      "category": "Scanning and algorithmic tooling",
      "fit": "Assess the specific advertised feature; scanning and automation alone do not certify ML validation.",
      "core": true,
      "sections": [
        [
          "Visible expanded edition",
          "Trade Ideas assessment Scanning and algorithmic tooling Assess the specific advertised feature; scanning and automation alone do not certify ML validation. Earlier product descriptions were inspected 7 October 2026. This expanded edition adds public-source research and original educational cases on 13 October 2026. We have not trained these vendors' models, operated their paid accounts, certified live profitability or collected a genuine user-rating sample. Name the component The Trade Ideas product guide is the source for separating its tools. An AI-labelled component needs its own evidence; a normal scanner condition should not inherit that label merely because both appear under the same brand. Evidence request Ask which model or rule produced the observed alert, whether it changed during the sample and how the record treats missed entries. This review does not certify training data, out-of-sample results or subscriber returns. Keep platform capabilities apart from claims about a particular strategy. Sources Trade Ideas product guide Compare the product categories A scanner, a strategy claim and a model definition The originating Trade Ideas guide is an information and scanning reference. A marketed AI function or a selected automated strategy still needs its own definition, producing version and evaluation population. Do not infer trained architecture from a scanner condition or treat every output as the same model. This assessment did not operate a paid account, inspect proprietary weights or reproduce a profitable strategy. Unknown components are listed as evidence requests rather than guessed from brand language. A shortlisted universe example A fictional scan produces ten original candidates, but a later report displays only three that had favourable price changes. The review retains the original ten and asks how the missing seven were handled. It does not describe the selected display as a complete model hit rate. If the original universe was not retained, that population remains unavailable. The case neither asserts a real Trade Ideas omission nor computes a vendor score; it illustrates the record a stronger selected-winner claim would require. Timing and transformation questions Preserve scan rule, data version, feature meaning and the time the observation was available. A retrospective chart marker is not necessarily an original delivered instruction. If the output is a rank, keep its rank definition rather than labelling it an event probability. A confidence or learning claim needs the originating model evidence. An equity observation also does not supply the exact contract, quantity or cost basis of an unrelated subscriber route. An adjacent task can still be useful Compare documented discovery functions for the task they support. A consumer requesting market explanations or screening may not need a private trained-model audit. A consumer requesting independently validated forecasts does need more than the general guide. Confirm current access and account terms at the origin. This page keeps those purchases separate and does not manufacture a low score for an adjacent tool to make another service appear the universal AI winner. Direct sources and reviewed scope Public documents, not private vendor experiments. The worked cases are invented; source dates do not establish a model-validation engagement. Trade Ideas product guide Core consumer protocols Search all AI evidence pages"
        ]
      ],
      "sources": [
        [
          "Trade Ideas product guide",
          "https://www.trade-ideas.com/guide/"
        ]
      ]
    },
    {
      "slug": "tradingview-signals",
      "name": "TradingView alerts",
      "category": "Scripts and chart notifications",
      "fit": "A rule-delivery layer; a script is not necessarily trained machine learning.",
      "core": true,
      "sections": [
        [
          "Visible expanded edition",
          "TradingView alerts assessment Scripts and chart notifications A rule-delivery layer; a script is not necessarily trained machine learning. Earlier product descriptions were inspected 7 October 2026. This expanded edition adds public-source research and original educational cases on 13 October 2026. We have not trained these vendors' models, operated their paid accounts, certified live profitability or collected a genuine user-rating sample. Explicit rule versus trained model TradingView documents chart alerts and script behaviour. A fixed indicator threshold can be algorithmic without being AI. Ask the author what is actually trained, if anything, rather than taking a script name as model documentation. Live reconstruction Retain version, inputs and original notifications. Repainting or revised historical plots can change the final display without reproducing the buyer's earlier information. The existence of a charting account does not validate the author's method; no author leaderboard or live profit test is claimed here. Sources TradingView alert documentation TradingView repainting documentation Compare the product categories A condition is not necessarily machine learning TradingView's alert documentation describes price and user-defined conditions with notification routes. A condition can be systematic without being a trained predictor. Retain the named rule author, saved settings, chart input and version. Public platform capability does not establish the architecture or profitability of every community script. This review has not bought a plan, trained a vendor model, used a private alert history or connected a broker account. A saved-settings packet A fictional alert is created under setting A, while the visible chart is later changed to setting B. The originating documentation explains that changing indicator parameters does not automatically replace the saved alert settings. The reviewer therefore keeps A's record separate from B's later screenshot. No live reliability figure or strategy return is assigned. If the actual saved configuration is unavailable, output identity remains a missing field, not a reason to infer that the newest chart defined the original decisions. Notification and forecasting evidence A notification or webhook response supports its own transport layer, not necessarily probability quality or actual account execution. Preserve original idea IDs and duplicate delivery handling before counting samples. Actual routing requires contract mapping, accepted order and fill records. A script's displayed confidence percentage also needs an event definition and suitable calibration evidence; an appealing label is not a substitute for those quantities and populations. A bounded comparison decision Use the documented alert tool as one component of a specific workflow. If the actual script is trained, ask for its model and experiment record; if it is a fixed rule, review that rule honestly without treating the absence of ML as inferior. Current plan, data access and compatibility need source confirmation. The comparison keeps information, model, delivery and execution functions distinct rather than awarding one platform a fabricated all-purpose AI rank. Evidence required at the handoff At the saved-alert handoff, record which parameter values were captured when the alert was created and which later edits affected only the chart. A fictional field showing the newest script title does not identify the original alert version. Preserve its actual configuration record. The inquiry can establish a deterministic condition honestly while leaving trained-model architecture, probability interpretation and external account outcomes unconfirmed. Direct sources and reviewed scope Public documents, not private vendor experiments. The worked cases are invented; source dates do not establish a model-validation engagement. TradingView alert documentation TradingView repainting documentation Core consumer protocols Search all AI evidence pages"
        ]
      ],
      "sources": [
        [
          "TradingView alert documentation",
          "https://www.tradingview.com/support/solutions/43000520149-introduction-to-tradingview-alerts/"
        ],
        [
          "TradingView repainting documentation",
          "https://www.tradingview.com/pine-script-docs/concepts/repainting/"
        ]
      ]
    },
    {
      "slug": "cryptohopper",
      "name": "Cryptohopper",
      "category": "Crypto automation marketplace",
      "fit": "A routing and marketplace comparison; confirm the selected algorithm's claims separately.",
      "core": true,
      "sections": [
        [
          "Visible expanded edition",
          "Cryptohopper assessment Crypto automation marketplace A routing and marketplace comparison; confirm the selected algorithm's claims separately. Earlier product descriptions were inspected 7 October 2026. This expanded edition adds public-source research and original educational cases on 13 October 2026. We have not trained these vendors' models, operated their paid accounts, certified live profitability or collected a genuine user-rating sample. Automation is not validation The signal marketplace connects different authors and configurations. A bot following an external message does not prove the message was generated by a trained model. Identify the original decision source and its version. Trace the whole chain Keep source signal time, receipt time, bot action, exchange acknowledgement and fill as distinct records. Request complete cancellations and rejected actions alongside gains. This desk has not run an account or established machine-learning performance for marketplace authors. Sources Cryptohopper signal marketplace Compare the product categories Platform function versus marketplace author Cryptohopper's marketplace reference identifies signal offers in a cryptocurrency context. A selected seller's original output, version and evidence need to be retained separately from the hosting platform. The presence of automation does not by itself establish a trained predictive model. This profile has not subscribed to a seller, inspected proprietary training artefacts or connected an exchange. It describes the consumer questions required to identify the actual purchased capability. One originating event, several repeats An invented seller event is copied into three later notifications with the same original ID. The record has one originating idea and three transmissions, not three independent model predictions. If the author's model architecture and full outcome population are absent, those claims stay unestablished. A selected successful chart cannot resolve them. The case is synthetic and makes no allegation about a real marketplace seller or the platform's current transport behaviour. Product and model scope Ask whether the selected offer supplies market context, fixed conditions, a learned score or account-routing inputs. Preserve venue, quotation units and any derivative specification independently. A token-pair label does not establish a dated future, collateral convention or payout. A probability claim requires its own target and observed outcomes; a bot or marketplace badge cannot supply that statistical record. Actual account results then require their own quantities, costs and cash reconciliation. Access and evidence handoff Confirm current seller terms, software access, data and venue permissions at their origins. Free marketplace browsing is not an indefinitely free strategy or account service. Retain what can be exported and any source-data restrictions. Unknown AI architecture is not proof the seller never uses ML, and unknown external evidence is not a measured bad result. A stronger claim must identify the chosen offer rather than inheriting general platform marketing or another publication's unverified preferred-provider verdict. Evidence required at the handoff A seller's offering should identify the component that turns market inputs into outputs. If one component is trained and another merely transports messages, keep both roles explicit. A fictional packet with only transport logs can demonstrate delivery observations but cannot establish model calibration. Confirm the selected seller's export and data rights separately from platform access, without guessing a proprietary feature pipeline or private venue compatibility. Direct sources and reviewed scope Public documents, not private vendor experiments. The worked cases are invented; source dates do not establish a model-validation engagement. Cryptohopper signal marketplace Core consumer protocols Search all AI evidence pages"
        ]
      ],
      "sources": [
        [
          "Cryptohopper signal marketplace",
          "https://www.cryptohopper.com/marketplace/signals"
        ]
      ]
    },
    {
      "slug": "vector-ridge",
      "name": "Vector Ridge",
      "category": "Systematic subscription comparison",
      "fit": "Useful for comparing evidence claims, but its machine-learning status is unestablished.",
      "core": false,
      "sections": [
        [
          "Visible expanded edition",
          "Vector Ridge assessment Systematic subscription comparison Useful for comparing evidence claims, but its machine-learning status is unestablished. Earlier product descriptions were inspected 7 October 2026. This expanded edition adds public-source research and original educational cases on 13 October 2026. We have not trained these vendors' models, operated their paid accounts, certified live profitability or collected a genuine user-rating sample. Credentials are a separate axis The founder context has separate sources. The organiser lists Darren O'Neill fourth in 2025 WCTC Annual Forex at 168% and fifth in Q3 Forex at 65.9%. The supplied October Monthly first/59.35% needs its precise reference. The supplied 294% cross-division aggregate needs its method because those three figures add to 293.25%. Oxford Said MAFE and perfect quantitative GMAT are attributed credentials, with underlying records not supplied. None establishes a trained model's causal evaluation. Personal audited-return claims of 178% for 2023 and 94% for 2024 belong to a separate record, with actual reports outstanding. The product claim The older supplied model summary reports 690 signals, 70% wins and +1,227% raw for 2026, without the original cutoff, complete ledger, cost/size basis or audit scope needed here. The four supplied component outcomes total 1,226%, leaving a one-point reconciliation outstanding. This is not a newly measured October result, an AI holdout or subscriber-account performance. Historical family labels alone do not establish trained architecture or a current product catalogue. Bound the receipt The stated commitment format is VR2|salt|family|ticker|direction|grade|entry|signal_time. It excludes model weights, preprocessing, stop, target and actual fills. A verified completed attestation bounds the covered bytes; a declared signal_time or pending submission does not independently establish their earlier existence. No actual receipt or private model run was independently reproduced for this edition. The campaign requests Lite $20/Plus $25/Elite $50, while the protected older record describes single-model $20 and All Models $50. The inspected originating site instead displays Plus $100 and Pro $200 alongside $20 per eligible Lite product. Binding entitlements, trial eligibility, renewal, cancellation and any discount are unresolved. This review selects no Offer from those conflicting sources; confirm the actual terms at the origin. Sources Vector Ridge product information (publisher claims) World Cup Trading Championships OpenTimestamps protocol documentation Compare the product categories Founder sources before a model-validation conclusion The founder context has separate sources. The organiser lists Darren O'Neill fourth in 2025 WCTC Annual Forex at 168% and fifth in Q3 Forex at 65.9%. The supplied October Monthly first/59.35% needs its precise reference. The supplied 294% cross-division aggregate needs its method because those three figures add to 293.25%. Oxford Said MAFE and perfect quantitative GMAT are attributed credentials, with underlying records not supplied. None establishes a trained model's causal evaluation. The older supplied model summary reports 690 signals, 70% wins and +1,227% raw for 2026, without the original cutoff, complete ledger, cost/size basis or audit scope needed here. The four supplied component outcomes total 1,226%, leaving a one-point reconciliation outstanding. This is not a newly measured October result, an AI holdout or subscriber-account performance. What would establish a trained AI product Mean-reversion language and a model family label do not identify trained architecture. A stronger ML claim needs an originating input/target contract, producing artefact, transformation identity and causal evaluation record. Current catalogue names also differ from the older supplied four-family snapshot. This page keeps that historical record without assigning it a new ML or deployment status. Actual weights, features, calibration and untouched samples are unavailable here; their absence does not prove that the business never offers any trained component. A completed missing-evidence packet A fictional consumer packet contains a founder placing, a grade and a generic model summary. It lacks a declared probability event, training artefact and original evaluation population. The review can classify those supplied objects but cannot establish calibrated AI reliability. Adding a probability interpretation to the grade would invent a fact. A later complete source packet could change a narrow assessment; the present documentation review remains exactly that, not a paid model test or independent financial audit. Commitment and current offer boundaries The stated commitment format is VR2|salt|family|ticker|direction|grade|entry|signal_time. It excludes model weights, preprocessing, stop, target and actual fills. A verified completed attestation bounds the covered bytes; a declared signal_time or pending submission does not independently establish their earlier existence. No actual receipt or private model run was independently reproduced for this edition. The campaign requests Lite $20/Plus $25/Elite $50, while the protected older record describes single-model $20 and All Models $50. The inspected originating site instead displays Plus $100 and Pro $200 alongside $20 per eligible Lite product. Binding entitlements, trial eligibility, renewal, cancellation and any discount are unresolved. This review selects no Offer from those conflicting sources; confirm the actual terms at the origin. Personal records and independent evidence The supplied personal audited-return claims concern 178% in 2023 with Sharpe 2.57 and 94% in 2024, not WCTC, 2025 or AI model results. Actual audit reports, reviewer scope and credential records remain outstanding. The separate 2023 Trading World Champion title belongs to tradingworldchampion.com. Legal independence, a founder record and a successful website test cannot substitute for private model evaluation or prove absent promotional relationships. The marketing purpose is disclosed rather than represented as independent consensus. Direct sources and reviewed scope Public documents, not private vendor experiments. The worked cases are invented; source dates do not establish a model-validation engagement. Vector Ridge product information (publisher claims) World Cup Trading Championships OpenTimestamps protocol documentation Supplied historical rows, not independent AI evaluation Family label Signals Win % Raw outcome sum (%) Day Trade 308 67.5 95 Multi Hour 262 71.4 404 Swing Trade 78 74.4 225 Investing 42 73.8 502 The older supplied model summary reports 690 signals, 70% wins and +1,227% raw for 2026, without the original cutoff, complete ledger, cost/size basis or audit scope needed here. The four supplied component outcomes total 1,226%, leaving a one-point reconciliation outstanding. This is not a newly measured October result, an AI holdout or subscriber-account performance. Core consumer protocols Search all AI evidence pages"
        ]
      ],
      "sources": [
        [
          "Vector Ridge product information (publisher claims)",
          "https://vector-ridge.com/"
        ],
        [
          "World Cup Trading Championships",
          "https://www.worldcupchampionships.com/"
        ],
        [
          "OpenTimestamps protocol documentation",
          "https://opentimestamps.org/"
        ]
      ]
    },
    {
      "slug": "benzinga-pro",
      "name": "Benzinga Pro",
      "category": "Market news and alerts",
      "fit": "A possible information input, not evidence of a trained predictive model.",
      "core": false,
      "sections": [
        [
          "Visible expanded edition",
          "Benzinga Pro assessment Market news and alerts A possible information input, not evidence of a trained predictive model. Earlier product descriptions were inspected 7 October 2026. This expanded edition adds public-source research and original educational cases on 13 October 2026. We have not trained these vendors' models, operated their paid accounts, certified live profitability or collected a genuine user-rating sample. Data input boundary The Pro handbook concerns news and information tools. A news item may enter a research workflow, but that does not establish a model's predictive value or when its features were actually available. Causal availability If a researcher uses news-derived features, preserve the original publication, receipt and processing times. A later revised article cannot automatically be backfilled into a historical prediction. This review neither certifies an ML product nor measures a news-based strategy. Sources Benzinga Pro handbook Compare the product categories An information product has its own evidence Benzinga Pro's handbook describes an information workflow, not the trained identity of every decision a reader later makes. A news notification, generated explanation and predictive signal are different objects. Retain original publication, observation and any later article revision. This desk did not operate a paid feed, measure latency or inspect a model that produced the vendor's information. The profile is not an independent profitability or forecasting certificate. A later summary does not prove an earlier input A fictional bulletin is observed at 10:00, adds a detail at 10:20 and is summarised at 10:30. The later summary may be useful while its new detail cannot establish the earlier reader's information set. The packet preserves each version and clock. No measured Benzinga delivery interval is claimed. A source link that still resolves does not prove it contains the original text, and a later price move does not by itself establish the news item's causal effect. A predictive extension needs specific records If a buyer uses news as model input, retain the availability join, feature extraction, target and producing artefact. A persuasive explanation cannot stand in for those fields. Current model or AI feature assertions require their exact originating description and selected entitlement; this page does not guess a proprietary architecture. The language-summary chapter adds a concrete citation/timing workflow rather than translating a general information service into an unobserved trading model. Compare the capability actually purchased A reader seeking timely information may need source retention and access more urgently than a trained-model manifest. A reader seeking validated forecasts needs another evidence packet. Confirm current costs and responsibilities directly and preserve the task distinction in the comparison. This profile gives the information category a legitimate role without a fabricated accuracy score, hidden hands-on trial or assumed subscriber return. Market context can be useful while its separate effect on an account remains unknown. Evidence required at the handoff For a summary-to-forecast handoff, retain the source text actually available before the prediction. A final corrected article can support today's explanation without authenticating yesterday's model inputs. A fictional record containing only the newest URL leaves that earlier content unknown. Keep the generated paragraph and original decision under separate IDs, and request a dated retrieval record before claiming that the cited news explains the trained model's actual reasoning. Direct sources and reviewed scope Public documents, not private vendor experiments. The worked cases are invented; source dates do not establish a model-validation engagement. Benzinga Pro handbook Core consumer protocols Search all AI evidence pages"
        ]
      ],
      "sources": [
        [
          "Benzinga Pro handbook",
          "https://www.benzinga.com/pro/handbook-3"
        ]
      ]
    },
    {
      "slug": "mql5-signals",
      "name": "MQL5 Signals",
      "category": "Account-copying marketplace",
      "fit": "Copying trades does not reveal whether the source uses ML, fixed rules or discretionary decisions.",
      "core": false,
      "sections": [
        [
          "Visible expanded edition",
          "MQL5 Signals assessment Account-copying marketplace Copying trades does not reveal whether the source uses ML, fixed rules or discretionary decisions. Earlier product descriptions were inspected 7 October 2026. This expanded edition adds public-source research and original educational cases on 13 October 2026. We have not trained these vendors' models, operated their paid accounts, certified live profitability or collected a genuine user-rating sample. Decision source unknown The MetaTrader subscriber guide describes copying operations. Unless the selected provider supplies separate model documentation, the existence of copied trades cannot establish an AI architecture. Evidence layers Account history may support questions about that account's trades while leaving feature causality and validation unanswered. Subscriber execution is another layer again. We do not generalise one account's record to all marketplace providers or assign an invented AI score. Sources MetaTrader signal subscriber documentation Compare the product categories Copying and trained forecasting are separate claims The MetaTrader subscription reference describes a copying workflow. It does not establish that every source author uses ML or that a follower received a calibrated probability. Preserve the selected author's product, model claims, original record and account route. This review has not subscribed, audited a source account or reproduced follower returns. General subscription documentation can describe a function while leaving the selected strategy's architecture and causal evaluation unavailable. A source/follower comparison packet A fictional source records one decision while a follower's allocation and fill evidence are absent. The source percentage cannot reconstruct the follower account by applying a guessed quantity ratio. Its capital, costs and clocks are different missing records. If no training artefacts were supplied, the decision's ML status also remains unconfirmed. The review preserves both gaps without alleging fraud or assigning an accuracy rating to the copying platform. Which record would answer the AI claim? Request the originating model or fixed-rule definition, features, target, version history and evaluation boundary. A source account's profitable period may be relevant financial evidence under its own method but does not certify model probabilities. Likewise, a matching message ID does not establish identical follower execution. Keep source predictions, source account and follower account as linked but separate populations, with modifications and unresolved quantities visible. Purchase terms and task fit Confirm the selected author, renewal terms, permissions and actual broker product definitions directly. Similar symbols can have different unit conventions; a shared name cannot supply mapping. A buyer interested only in copying may ask different questions from one seeking independently validated AI. This publication distinguishes those purposes, preserves unknown architecture and refuses to rank an adjacent workflow as either an automatically poor or automatically superior prediction service without comparable evidence. Evidence required at the handoff The follower's broker symbol and quantity definitions also need their own source. A copied label cannot establish identical contract units or account exposure. A fictional source lot of 0.1 is not converted to a standard exchange contract by assumption. Retain each actual specification and documented mapping. Neither that mapping nor a subscription badge supplies the model's target, training history or complete independent outcome population for a separate AI claim. Direct sources and reviewed scope Public documents, not private vendor experiments. The worked cases are invented; source dates do not establish a model-validation engagement. MetaTrader signal subscriber documentation Core consumer protocols Search all AI evidence pages"
        ]
      ],
      "sources": [
        [
          "MetaTrader signal subscriber documentation",
          "https://www.metatrader5.com/en/terminal/help/signals/signal_subscriber"
        ]
      ]
    },
    {
      "slug": "learn2trade",
      "name": "Learn 2 Trade",
      "category": "Forex instructions and education",
      "fit": "Signal instructions are not a machine-learning methodology record.",
      "core": false,
      "sections": [
        [
          "Visible expanded edition",
          "Learn 2 Trade assessment Forex instructions and education Signal instructions are not a machine-learning methodology record. Earlier product descriptions were inspected 7 October 2026. This expanded edition adds public-source research and original educational cases on 13 October 2026. We have not trained these vendors' models, operated their paid accounts, certified live profitability or collected a genuine user-rating sample. Keep the advertised offer bounded The cited forex guide describes how users interpret instructions. It does not provide this desk with a training dataset, model version or untouched validation sample. What would change the assessment A separately documented AI offer would need its own model and data evidence, not just a new headline. No paid-account trial or model inspection was performed here, so this profile remains an adjacent service reference rather than an AI shortlist recommendation. Sources Learn 2 Trade signal instructions Compare the product categories Signal use does not identify the decision engine The retained Learn 2 Trade reference describes a Forex signal-use workflow. It is not a supplied training protocol or independently reproduced ML evaluation. A fixed rule, discretionary instruction and learned forecast can all produce a message, so architecture requires originating evidence. No paid feed, private model, account route or genuine user-rating sample was operated for this assessment. Unknown AI status is a scope boundary, not a claim the vendor never uses another method. A number with no declared output meaning An invented message labels confidence 80 but supplies no event, horizon, scale or producing-model definition. The review records the literal label without converting it to an 80% future success chance. A later favourable move does not fill that missing definition. If the seller supplies a calibrated probability contract and full evaluation record, the claim can be reconsidered on those sources. The example is not an actual vendor message or evidence of a poor-performing model. Separate original instruction and account outcome Retain validity, amendments, delivery and the actual product. A Forex direction does not supply a selected futures month, tick value or account quantity. A source's hypothetical outcome can remain useful under its stated convention while subscriber execution requires its own accepted orders and fills. Preserve the original research entry window rather than replacing it with a hindsight price to make the account narrative look complete. A useful adjacent-category enquiry Ask what capability the selected offer promises and which records support it. Current subscription, trial and renewal conditions need direct confirmation. Documentation can support signal-use questions without proving a trained edge. This profile does not invent private tests or a universal profit ranking. The expanded taxonomy lets a buyer distinguish explanations, instructions and execution routes, and it identifies the additional evidence required before a stronger AI claim could be responsibly repeated. Evidence required at the handoff For a probability handoff, ask whether the numeric field is a score, a category or a probability of a named event. Keep the answer with the original output version. A fictional clarification saying it is an internal rank does not invalidate the rank; it prevents the publication from calling it a future success percentage. Compare the documented function instead of awarding a stronger statistical meaning because that would make a promotional headline more attractive. Direct sources and reviewed scope Public documents, not private vendor experiments. The worked cases are invented; source dates do not establish a model-validation engagement. Learn 2 Trade signal instructions Core consumer protocols Search all AI evidence pages"
        ]
      ],
      "sources": [
        [
          "Learn 2 Trade signal instructions",
          "https://learn2.trade/using-our-forex-signals"
        ]
      ]
    },
    {
      "slug": "zulutrade",
      "name": "ZuluTrade",
      "category": "Social account-copying platform",
      "fit": "A route for following selected accounts, not one certified AI strategy.",
      "core": false,
      "sections": [
        [
          "Visible expanded edition",
          "ZuluTrade assessment Social account-copying platform A route for following selected accounts, not one certified AI strategy. Earlier product descriptions were inspected 7 October 2026. This expanded edition adds public-source research and original educational cases on 13 October 2026. We have not trained these vendors' models, operated their paid accounts, certified live profitability or collected a genuine user-rating sample. Platform and trader differ A copied account can use any of several decision processes. The public platform description alone does not establish the selected trader's model design or training history. No inferred model audit Retain the provider identity and account record, but do not label it out-of-sample AI performance without the model's experimental provenance. This desk has not audited those records, trained a selected strategy or tested follower execution. Sources ZuluTrade product site Compare the product categories A social/copied record is not an ML manifest ZuluTrade's originating information supports a social/copying category review. It does not identify the decision engine of every listed leader. Save the selected source, product, architecture claim and observation period. A leader history, a follower allocation and actual account outcome need separate evidence. No selected leader was followed and no private account or model-training record was inspected by this desk for the edition. Two leaders do not automatically mean independent models A fictional account follows two leaders whose decisions concern the same underlying event. Their published percentages do not establish an independently diversified AI portfolio. The review needs actual allocation, quantities, overlapping positions, costs and source relationship. If architecture is unestablished, it cannot declare the leaders different trained models. The example preserves the overlap question without a measured result, a causality claim or an allegation about the real service. An AI claim needs its own original evidence Request the specific source's target, features, producing version, development history and untouched evaluation if it claims trained forecasting. A profitable copied-account observation can remain a different evidence object without those records. A model's reported win fraction also cannot supply a follower's fills or capital. Preserve missing and open states rather than presenting selected outcomes as a complete universal sample. How to compare the chosen task Compare documented source selection and copying responsibilities when that is the actual purchase. Compare model records only for identified forecasting claims. Current eligibility, allocation and costs come from originating terms. Unknown architecture is neither proof of a poor source nor evidence of a proven edge. This publication does not create a follower account, promise execution identity or substitute coordinated promotional repetition for an independent validation packet. Evidence required at the handoff An aggregate portfolio claim needs the selected leaders, weights, changes, source relationships and actual follower cash path. Multiple leader names do not prove independent models or errors. In a fictional packet with two overlapping decision streams and no account ledger, the monetary result stays unknown. Preserve the raw observations as supplied and request the missing allocation rather than adding source percentages or inventing a diversification benefit. Direct sources and reviewed scope Public documents, not private vendor experiments. The worked cases are invented; source dates do not establish a model-validation engagement. ZuluTrade product site Core consumer protocols Search all AI evidence pages"
        ]
      ],
      "sources": [
        [
          "ZuluTrade product site",
          "https://www.zulutrade.com/"
        ]
      ]
    }
  ],
  "guides": [
    {
      "path": "primers/ai-data-leakage-and-holdout-audit.html",
      "title": "AI data leakage and holdout audit",
      "description": "Audit feature availability, outcome overlap and test exposure with a frozen experiment record and a worked calendar-boundary example.",
      "sources": [
        [
          "scikit-learn: data leakage and preprocessing",
          "https://scikit-learn.org/stable/common_pitfalls.html"
        ],
        [
          "scikit-learn: probability calibration",
          "https://scikit-learn.org/stable/modules/calibration.html"
        ],
        [
          "CFTC: AI trading-bot consumer advisory",
          "https://www.cftc.gov/LearnAndProtect/AdvisoriesAndArticles/AITradingBots.html"
        ],
        [
          "ALFRED: vintage-data help",
          "https://alfred.stlouisfed.org/help"
        ]
      ],
      "sections": [
        [
          "Begin with one decision and its information set",
          "Choose an identifiable instruction rather than a polished aggregate chart. Record the decision clock, model artefact, feature specification and data identity used at that moment. Ask the seller which component actually produces the instruction: a fixed rule, trained predictor, explanatory language model or routing service. Unknown architecture should stay unknown, even when an application supports machine learning. A product capability and a particular deployed model are different objects.\n\nBuild a small availability ledger around that decision. For each input, retain what happened, when it was published, when the system received it and which revision was available. These clocks can disagree without anybody falsifying a record. A monthly observation label is not a release timestamp. A later corrected price file cannot silently become the historical file. This first audit step defines the question; it does not establish model accuracy, profitability or independent verification."
        ],
        [
          "Inspect transformations before inspecting scores",
          "A train/test separation can fail before the estimator is fitted. Ask how missing values, scaling, feature selection and category mappings were learned. The relevant distinction is between learning a transformation from development observations and applying that already learned transformation to evaluation observations. A retained pipeline specification helps identify the distinction, but its existence alone does not prove that a vendor followed it. Request run artefacts showing the actual data membership.\n\nConsider an invented review where a seller selected indicators using a full history and only then divided the rows into training and evaluation periods. The chronological labels on the resulting chart do not repair the earlier selection exposure. Preserve that experiment as development evidence and state the exact flaw. Do not claim that every later model is therefore fraudulent or useless. A replacement experiment requires its own data boundary, identity and untouched evaluation plan, not a retrospective change of headings."
        ],
        [
          "Work through an overlapping-label boundary",
          "Here is a fictional calendar check, not a tested strategy. A last training row dated 31 December 2025 has a target horizon of two calendar days. Its declared target reaches 2 January 2026. If evaluation starts on 2 January, the target and evaluation touch under the worksheet convention requiring the label end to be strictly earlier than the test start. Moving that start to 3 January clears only this simple comparison. It says nothing about exchange sessions or other overlapping rows.\n\nWrite the actual label construction down before choosing a purge or gap. A target can involve a variable exit, a maximum observation window or another instrument clock; a fixed number of calendar days may describe none of them. The date checker deliberately exports an untested plan. It does not inspect samples, prove feature causality or certify adequate separation. Keep timezone, label endpoint and boundary equality visible so a reviewer can reproduce precisely the limited arithmetic being offered."
        ],
        [
          "Preserve what each experiment exposed",
          "Create an append-only experiment index containing the candidate identity, data periods viewed, questions asked and decisions influenced. A test has become development evidence when its outcomes guide another threshold, feature or model choice. A different filename or retraining date does not make that same period unseen again. Record unsuccessful experiments as well as the candidate later shown to consumers. Otherwise a final performance figure conceals the search process that produced it.\n\nFreeze the entry objective and candidate identity before applying the predeclared exit evaluation. Exit-adjusted profit, holding drift and realised win rate can inform a later validation or rejection decision; they should not quietly redefine which entry candidates won the original search. Preserve the authorised hypothetical research-fill convention separately from broker execution. Changing the fill window after seeing outcomes creates a different experiment. A hash can identify retained bytes, but cannot prove that no unlogged experiment happened or that the declared access policy was obeyed."
        ],
        [
          "Turn a missing artefact into a specific question",
          "Use four distinct statuses: requested, supplied, inspected and unresolved. A link to an API manual is supplied documentation, not an inspected model run. A diagram describing training is not the training artefact. If the model weights are commercially sensitive, ask for a narrower audit object such as dated version identifiers, a feature-timing specification or a redacted experiment ledger. Explain what each object would answer without claiming that one object completes the entire investigation.\n\nPreserve the seller response with its date and exact product scope. An honest inability to share a file is still missing evidence for the claim requiring that file. Avoid converting silence into an invented score or assuming that a founder credential supplies the missing experiment history. ALFRED can help investigate vintage differences, but it does not provide the intraday receipt log of an arbitrary system. Consumer review language should describe that limitation plainly enough that a reader can ask the next useful question."
        ],
        [
          "Finish with a bounded audit conclusion",
          "A useful conclusion names the strongest layer actually reviewed. For example: documentation describes a training boundary, a fictional date exercise reproduces the stated comparison, and real feature receipts remain unavailable. That is more actionable than a universal clean-data badge. Separate software correctness, historical simulation, prospective observation and executed account evidence in the final record. Passing a worksheet test demonstrates the worksheet behaviour, not an independent strategy-validation sample.\n\nBefore paying, compare the unresolved items with the capability you need. A consumer seeking a reproducible experiment may reasonably require different evidence from somebody purchasing market explanations only. Do not turn the checklist into an allocation recommendation. Save the blank CSV with private identifiers removed before sharing it, and date later corrections rather than rewriting the original assessment. This publication supplies a research framework and linked documentation; it has not inspected a private model dataset, established a new holdout or verified the results described by the seller in an advertisement."
        ]
      ]
    },
    {
      "path": "primers/backtest-forward-and-live-evidence.html",
      "title": "Backtest, forward and live evidence",
      "description": "Reconcile simulated decisions, prospective instructions and account fills without promoting one evidence layer into another.",
      "sources": [
        [
          "scikit-learn: data leakage and preprocessing",
          "https://scikit-learn.org/stable/common_pitfalls.html"
        ],
        [
          "scikit-learn: probability calibration",
          "https://scikit-learn.org/stable/modules/calibration.html"
        ],
        [
          "CFTC: AI trading-bot consumer advisory",
          "https://www.cftc.gov/LearnAndProtect/AdvisoriesAndArticles/AITradingBots.html"
        ],
        [
          "ALFRED: vintage-data help",
          "https://alfred.stlouisfed.org/help"
        ]
      ],
      "sections": [
        [
          "Identify the claim before collecting files",
          "Write the advertised claim in a single sentence with its period, account or model identity and result convention. Historical model performance, prospective signal publication and subscriber profit require different records. If a seller says live, ask whether that means a running software process, an observation stream or actual funded execution. Do not choose the most favourable interpretation on behalf of the seller. Retain the original wording beside the review question.\n\nStart separate folders or ledger fields for each evidence layer. A backtest report belongs with its dataset and execution convention. A prospective record belongs with its publication history. An account statement belongs with deposits, withdrawals, orders and fills. Related records can be reconciled through stable identifiers without being merged into one ambiguous percentage. This separation makes a missing fill or amendment visible. It also prevents a genuine historical simulation from being dismissed simply because it is not something it never claimed to be."
        ],
        [
          "Reproduce the simulation convention, not the sales chart",
          "Ask which instrument histories, universe membership, timezone and corporate-action conventions were used. Retain the model version, entry objective, chosen exit and treatment of absent observations. Costs need units and timing, not just a word such as realistic. A favourable research fill under a declared observation window remains a hypothetical reference; it cannot be described as the price every subscriber obtained. Preserve the frozen convention when comparing releases.\n\nAn invented example contains twenty closed hypothetical decisions, fourteen labelled positive and six negative, plus five still open. The closed-only positive fraction is fourteen divided by twenty, or seventy percent. Twenty-five total decisions answer a different coverage question. The example supplies no actual trading result and says nothing about net return, position weights or the final outcomes of the open items. Keep the unresolved rows instead of forcing them into wins or dropping them from the instruction count. A chart alone cannot reveal these denominators."
        ],
        [
          "Count prospective instructions before counting profits",
          "A forward observation should retain decisions when they appear, including those later cancelled, expired, rejected or left unresolved. Record a model deployment change at the time it happens rather than assigning an earlier prediction to a better later artefact. Distinguish source publication from subscriber receipt: a timestamp attached to one does not prove the other. Corrections should link to originals rather than overwrite them invisibly.\n\nSuppose a fictional stream contains twelve messages: four commentaries, six actionable originals and two amendments to those originals. There are six original instructions, not twelve independent trades. If three of those six reached an account and three did not, execution coverage in this toy set is one half. That is not a profitability estimate. The amendment relationship and each unexecuted reason matter. A lost permission, stale instruction or missing acknowledgement belongs to an operating record, even if no profit can yet be calculated for it."
        ],
        [
          "Reconcile execution states without assuming a fill",
          "For every actionable instruction, map the requested order to acknowledgements, partial fills, cancellations and remaining quantity. A timeout is an unknown state until reconciled, not proof that an order never existed. An accepted order is not necessarily an executed order. Keep event ordering and venue identifiers where available, with private account information redacted in a public review. Do not place real orders to demonstrate this guide.\n\nA hypothetical request for one unit can fill 0.4 units and cancel the remaining 0.6. The account still owns 0.4 units; cancellation is not an exit. A later confirmed sale of 0.25 leaves 0.15. These arithmetic states illustrate why a source signal and an account trade can diverge. Fees, financing, currency conversion and mark-to-market conventions must then be applied to the actual quantities and clocks. No generic platform description proves that every provider or subscriber records those states correctly."
        ],
        [
          "Keep cash flow and model identity in the result",
          "An account value can rise after a deposit even when the trading result is flat or negative. Retain externally contributed and withdrawn cash with its timing. Name the return convention used and avoid presenting a raw balance ratio as a cash-flow-adjusted result. An account-level measure and an unweighted sum of model percentages are not interchangeable. If the aggregation method is unavailable, report the published number as an attributed claim rather than independently reproduced performance.\n\nSimilarly, a family label can cover several model versions. Preserve the sequence of versions, retirement dates and evaluation rules. A later corrected dataset should not silently alter the old release receipt. If a source changes its record, keep both the old assessment and the dated correction. Do not infer a complete population from one selected successful account. The purpose of reconciliation is to expose what the records support and where they stop, not to manufacture a universally comparable league table."
        ],
        [
          "Write an evidence-layer decision for the reader",
          "End the assessment with the reviewed period, supplied artefacts, missing states and the strongest supported statement. Documentation-only research remains documentation-only even if the interface looks convincing. A preserved prospective stream can support publication chronology without proving funded execution. A reconciled account can establish a specific account experience without forecasting the next subscriber outcome. Say which boundary applies instead of using an unqualified verified label.\n\nUse the checkpoint table to prioritise the next request. A consumer may need an actual cost ledger more urgently than another screenshot of gross profit. Someone comparing research platforms may instead need stable experiment identities and export formats. Preserve private information locally and share only redacted references. These pages have not run a paid model, subscribed a funded account or verified a live return. Their value is a reproducible way to distinguish evidence objects and prevent one valid object from being used to imply several unsupported conclusions."
        ]
      ]
    },
    {
      "path": "primers/algorithmic-versus-ai-signal-claims.html",
      "title": "Algorithmic versus AI signal claims",
      "description": "Map fixed rules, trained predictions, AI explanations and order routing to the specific consumer evidence each needs.",
      "sources": [
        [
          "scikit-learn: data leakage and preprocessing",
          "https://scikit-learn.org/stable/common_pitfalls.html"
        ],
        [
          "scikit-learn: probability calibration",
          "https://scikit-learn.org/stable/modules/calibration.html"
        ],
        [
          "CFTC: AI trading-bot consumer advisory",
          "https://www.cftc.gov/LearnAndProtect/AdvisoriesAndArticles/AITradingBots.html"
        ],
        [
          "ALFRED: vintage-data help",
          "https://alfred.stlouisfed.org/help"
        ]
      ],
      "sections": [
        [
          "Buy a named capability, not an elastic label",
          "Before comparing products, complete the sentence: I am paying for this component to perform this task. A market summary, a predictive score, an actionable instruction and an account-routing service are four different purchases. A vendor can offer several together, but each needs its own evidence and limitations. The word AI does not establish which task is trained, which is deterministic or which is added after the decision. Record unknown components instead of guessing their architecture.\n\nAsk for a dated representative output and the product terms covering that output. A platform able to train models may also host ordinary threshold alerts; its general feature page does not validate a particular strategy. A consumer directory should compare the specific offers, not rank unrelated capabilities as though they are substitutes. This guide deliberately does not create an AI quality percentage. The distinction starts with product identity and ends with what the available records actually establish."
        ],
        [
          "Describe a fixed rule without treating it as inferior",
          "A deterministic rule can be systematic, exportable and useful without learning statistical parameters. Request its input definitions, decision clock, thresholds, expiry and amendment policy. Ask how the same rule behaves when data are missing or revised. Reproducibility depends on those details as well as the expression printed on a product page. A fixed rule can still be overfit through human experimentation, so absence of machine learning is not proof of independence from past outcomes.\n\nFor a fictional example, an alert fires when a defined observed value exceeds a threshold selected before the observation period. Its routing system sends the alert to a phone. Neither the threshold nor delivery requires a trained predictor in this example. Calling the combination automated is accurate; calling it calibrated forecasting would require additional evidence. Preserve the rule version if the threshold later changes. Compare the rule with its intended function rather than awarding or withholding credibility merely because its marketing omits fashionable terminology."
        ],
        [
          "Inspect a trained model at the prediction boundary",
          "A trained predictor requires an identified artefact, target definition and account of how inputs become outputs. Consumers do not need public access to proprietary weights to ask when training stopped, what the output represents and how the evaluation was separated from development. Keep feature and model versions together; a stable commercial brand can conceal changing models. Distinguish retraining cadence from the actual deployment time of each artefact.\n\nImagine a seller output labelled 0.8. It might be a probability, a transformed score or an arbitrary ranking value. Until the definition is supplied, none of those meanings is established. Even a probability claim requires an outcome definition and suitable evaluation evidence. A hypothetical bin containing ten predictions and six observed successes illustrates a frequency of sixty percent in that tiny invented bin, not proof of calibration or a future success chance. Small, dependent or selectively retained observations cannot support a strong universal accuracy statement."
        ],
        [
          "Treat explanations as explanations",
          "An explanatory language model may describe an event or translate another engine output into ordinary language. Ask whether the explanation was produced before the instruction, alongside it or afterwards. A plausible narrative generated after a profitable move does not prove that the original model anticipated that move. Preserve the raw decision and explanation as separate dated objects, especially when a system can edit summaries after source data change.\n\nConsider a fictional alert with a numeric decision and an accompanying paragraph naming three reasons. If the paragraph was generated after the decision, those reasons are not automatically the variables the decision engine used. Request an actual feature mapping before describing them as model attribution. A readable explanation can still be a useful consumer feature without validating the forecast. This publication does not inspect hidden reasoning or certify that vendor narratives faithfully represent proprietary models. It asks for the relationship between the objects a reader can actually examine."
        ],
        [
          "Audit routing independently of forecasting",
          "A webhook, bot or copy connection can transport an instruction regardless of how that instruction was generated. Trace original publication, delivery, acknowledgement and account action separately. Check duplicate identifiers, stale messages, permission loss and reconciliation after an interrupted connection. Reliable delivery and reliable prediction are different properties; one cannot stand in for the other in a review headline. Keep operating records separate from the statistical evaluation.\n\nA fictional message sent twice with the same instruction identifier should be investigated as one instruction with duplicate delivery, not two independent trading opportunities. If the first account submission times out, the operating state remains unknown until checked; retrying blindly can create duplicate exposure. This guide is not a deployment runbook and does not connect a broker. It establishes the questions a consumer should ask before assuming that an appealing prediction interface includes an equally well evidenced execution service or a complete history of its failures."
        ],
        [
          "Use a scope-aware comparison verdict",
          "Build the comparison around task, evidence, access, operating responsibilities and total contractual cost. Record any limits on export, supported accounts and data access from the exact offer rather than transferring a claim from another subscription. A lower price is not evidence of a worse predictor, and a high founder return is not evidence of better routing. Avoid a single synthetic ranking when the products answer different consumer needs. Unknown capability is a useful finding, not a blank to fill with favourable prose.\n\nThe CFTC advisory is a relevant caution about extravagant AI profit claims; it does not establish that a particular listed seller is fraudulent. Retain the original claim and the missing evidence without unsupported allegations. A reader can choose to request clarification, use a free alternative or decline the purchase. None of those options requires an invented consensus score. This taxonomy supplies a decision vocabulary and blank record, not a personal trading recommendation, new model test or promise that AI will improve investment outcomes."
        ]
      ]
    }
  ],
  "rebuild": [
    [
      "methodology.html",
      "How AI signal claims are assessed",
      "An evidence taxonomy, not a leaderboard produced from untested model claims.",
      [
        [
          "Architecture",
          "Identify whether the decision comes from trained ML, a fixed rule, a language-model explanation or a human. A product's ability to train models does not establish that every user strategy employs one. Keep unknown architecture unknown."
        ],
        [
          "Causal validation",
          "Ask for version identity, feature availability, training boundaries, label horizons and a declared evaluation policy. A viewed test period cannot remain an untouched holdout after it influences development. Our date checker creates a planning record, not an experimental result."
        ],
        [
          "Execution evidence",
          "Separate predictions, hypothetical strategy outcomes, prospective observation and broker fills. Costs and missed instructions can change the executable result. A receipt for a model output cannot substitute for account evidence or a complete consecutive record."
        ],
        [
          "Editorial limits",
          "These profiles are primary-document research, not model inspection or funded-account trials. No genuine user-rating dataset was supplied, so no aggregate stars or fabricated review totals are published. A service is not named the AI winner merely because its founder has credentials."
        ]
      ]
    ],
    [
      "faq.html",
      "AI signal evidence questions",
      "Answers about automation, holdouts, backtests and cryptographic records.",
      [
        [
          "Is every trading algorithm AI?",
          "No. A fixed rule can be systematic without being trained ML. Identify the actual decision component and its documentation. A bot routing somebody else's message does not prove the message came from a learned model."
        ],
        [
          "Does a clean date split validate a model?",
          "No. The worksheet checks one calendar condition and the user's declared test status. It does not inspect data, detect all leakage, measure predictive skill or estimate trading results. Every export remains untested."
        ],
        [
          "Can a backtest be out of sample?",
          "Its evidential status depends on the frozen experiment and what influenced development. If results shaped later choices, that period is development evidence. Historical simulation also remains distinct from messages delivered or positions executed in real time."
        ],
        [
          "Does a receipt prove AI accuracy?",
          "A verified commitment concerns its covered bytes and attestation. It does not identify model weights, training data or omitted predictions unless those are separately documented. Performance and completeness require their own evidence."
        ]
      ]
    ],
    [
      "about.html",
      "About the AI research desk",
      "A consumer guide to what an AI trading claim can and cannot establish.",
      [
        [
          "Research focus",
          "This site concentrates on causal data, model identity, validation history and the handoff from prediction to execution. Its value is a clear set of questions and source-linked product distinctions, not stronger marketing adjectives."
        ],
        [
          "What was not tested",
          "No paid account, vendor model training run or live signal performance test was conducted for these profiles. Ownership and compensation have not been independently audited by this desk. We do not invent client rosters, testimonials, expert-team biographies or user counts."
        ],
        [
          "Corrections and new evidence",
          "A model-specific primary record can resolve an architecture or validation gap. Include the version, period and scope when proposing a correction. The editorial update date should never be used to imply current model performance or a completed independent audit."
        ]
      ]
    ],
    [
      "data-sources.html",
      "AI research source hierarchy",
      "Documentation describes functions; experiment and account artefacts establish different kinds of evidence.",
      [
        [
          "Product documentation",
          "Official guides support descriptions of training, charting and automation capabilities. Commercial product claims are attributed rather than treated as independent performance certification. Current plan access remains a supplier check."
        ],
        [
          "Data identity",
          "Economic vintages, article revisions and historical chart behaviour can affect what was causally available. Preserve the input version and availability timestamps in a real experiment. A present-day download is not automatically the historical information set."
        ],
        [
          "Experimental records",
          "An adequate validation packet needs frozen model and feature identity, split rules, run artefacts and a history of what results influenced development. Such a packet was not supplied for every listed product. Software correctness tests of our worksheet are not strategy-validation samples."
        ],
        [
          "Account records",
          "Live claims require actual order, fill and cost evidence under a declared account convention. A hypothetical aggregate or a single receipt does not supply it. The public category JSON reproduces this desk's current text and sources, not a stronger audit claim."
        ]
      ]
    ],
    [
      "field-brief.html",
      "An AI evaluation manifest",
      "The artefacts to preserve before a model's designated results are inspected.",
      [
        [
          "Frozen identity",
          "Record model and feature versions, data identity, training boundaries, label definition and decision rule. Note when retraining produces a new artefact. Do not let later improvements inherit an earlier model's supposed validation."
        ],
        [
          "Test provenance",
          "Save which data were used for development, which results were viewed and what remains genuinely untouched. Calendar separation is one narrow check, not a full leakage certification."
        ],
        [
          "Result classification",
          "Label simulation, forward observation and account execution separately. Preserve failed runs and missing outputs. A defensible report states exactly what was measured rather than calling an untested plan a validated trading system."
        ]
      ]
    ],
    [
      "benchmarks/cryptographic-receipts.html",
      "Cryptographic receipts and model claims",
      "Integrity of a committed payload is not proof of a model's training or returns.",
      [
        [
          "What was committed?",
          "Inspect the exact message layout, encoding and proof. A visible output may contain more fields than its hash input. Changing a field outside that payload need not change the commitment; do not promise otherwise."
        ],
        [
          "When was it confirmed?",
          "Initial calendar submission is not confirmed Bitcoin attestation. Verify the actual proof and separate its anchoring time from the provider's declared time, delivery and account execution. A timestamp written inside a payload is not independently certified as the moment of prediction."
        ],
        [
          "What remains outside the proof?",
          "A matching digest does not establish architecture, feature causality, complete prediction counts or profit. Model weights, data and experiment manifests need their own identities and checks. One verified instruction must remain one example, not an audit of all outputs."
        ]
      ]
    ],
    [
      "benchmarks/re-runnable-track-record.html",
      "Reconstructing an AI signal record",
      "Versioned outputs and explicit evaluation conventions before performance claims.",
      [
        [
          "Prediction sequence",
          "Request consecutive outputs with timestamps, model versions and the definition of the target or decision. Include missing or invalid outputs. A selected list of winners cannot supply the denominator."
        ],
        [
          "Evaluation layers",
          "Recompute the stated prediction metric separately from a simulated trade return. Preserve the chosen research-fill convention and costs; do not substitute a more favourable one. Actual subscriber executions require their own evidence."
        ],
        [
          "Aggregation",
          "State the period, exposure and formula for any combined figure. Summing model percentages does not automatically create portfolio growth. If a rounded summary cannot be reconciled from records, report the gap instead of inventing a compounding explanation."
        ]
      ]
    ],
    [
      "benchmarks/transparent-vs-black-box.html",
      "Transparency without invented certainty",
      "Know what evidence is visible and what commercial secrecy leaves unanswered.",
      [
        [
          "Useful disclosure",
          "A seller can explain decision scope, feature timing, retraining and validation without publishing every commercial parameter. Request those boundaries rather than treating either secrecy or an open code repository as a complete quality verdict."
        ],
        [
          "Inspectable is not validated",
          "Visible code can still leak future information or use unrealistic execution assumptions. Conversely, an opaque system can have account evidence relevant to a particular return claim. Match each conclusion to what was actually checked."
        ],
        [
          "Classification",
          "Describe an unknown architecture as unknown. Describe untested capability as documented capability. No platform, model family or distribution channel receives a universal profitability verdict from this desk."
        ]
      ]
    ],
    [
      "primers/verify-an-ai-signal-record.html",
      "Verify an AI signal record",
      "An evidence sequence that does not confuse a receipt with causal model validation.",
      [
        [
          "Ask for the experiment",
          "Identify the trained component, version, data availability and evaluation policy. A model family label is not enough. Preserve which results influenced development and which tests remain untouched."
        ],
        [
          "Audit the output record",
          "Choose a consecutive period and include every output status. Check original payload and confirmed attestation for any receipt, within its actual field coverage. Do not assume omitted stops, model weights or later instructions are committed."
        ],
        [
          "Classify the result",
          "Reconstruct the declared statistical or trading measure, then inspect real account executions separately if supplied. Report missing records and exact test scope. A successful calendar check or one receipt is not a completed AI performance audit."
        ]
      ]
    ],
    [
      "standards.html",
      "Editorial evidence standard",
      "AI claims need identified model architecture and causal evaluation evidence. Product documentation and planning worksheets do not certify an investable edge.",
      [
        [
          "Claims with sources",
          "Primary-document research dated 7 October 2026. We have not trained these vendors' models, operated their paid accounts, certified live profitability or collected a genuine user-rating sample. Product features are linked to primary documentation. A commercial claim is not presented as independent certification, and an editorial update date is not a current vendor-performance cutoff."
        ],
        [
          "No fabricated authority",
          "The assessments do not invent paid-account tests, experts, customers, testimonials or review counts. A source gap stays visible. A full ownership, compensation and legal audit has not been performed by this desk; stronger independence or conflict claims require records."
        ],
        [
          "Corrections and limits",
          "AI claims need identified model architecture and causal evaluation evidence. Product documentation and planning worksheets do not certify an investable edge. Corrections should identify the disputed page, dated source and exact claim it supports. We do not promise indexation, search rank or future returns from publishing deeper material."
        ]
      ]
    ],
    [
      "benchmarks/index.html",
      "AI evidence benchmarks",
      "Architecture, causality, experiment identity and execution are separate checks.",
      [
        [
          "Identify the decision process",
          "Determine what is trained and what is ordinary routing or fixed-rule logic. Product documentation can establish a capability without validating the selected strategy."
        ],
        [
          "Causal and untouched evidence",
          "Inspect feature availability, target horizon, split rules and the history of results seen during development. An outcome-guided change affects the status of the test period."
        ],
        [
          "Bound each proof",
          "A confirmed receipt concerns covered bytes. A backtest concerns its simulation convention. Account claims require real fills and costs. No blanket pass/fail matrix for whole product categories is published."
        ]
      ]
    ],
    [
      "primers/ai-trading-signal-red-flags.html",
      "AI claim gaps to investigate",
      "Ask for model-specific evidence without treating secrecy or a distribution format as automatic fraud.",
      [
        [
          "Unnamed trained component",
          "If the seller cannot explain what is trained, leave the architecture unknown. A bot, scanner or impressive explanation alone does not establish ML."
        ],
        [
          "Test used for development",
          "A period repeatedly used to choose rules is not untouched. Request the frozen manifest and exposure history. Renaming a window does not erase that history."
        ],
        [
          "Simulation called live",
          "Historical annotations or reconstructed fills are not evidence of delivered instructions and executed orders. Ask for the appropriate prospective or account records instead."
        ],
        [
          "Proof stretched beyond its scope",
          "A receipt cannot certify omitted fields, complete predictions or training quality. An open repository cannot guarantee causal data. These gaps limit a claim but do not justify a universal accusation against every opaque model or chat service."
        ]
      ]
    ],
    [
      "primers/are-ai-trading-signals-worth-it.html",
      "Are AI trading tools worth paying for?",
      "Capability, research convenience and validated trading performance are different purchase questions.",
      [
        [
          "Name the deliverable",
          "A subscription may provide model-development software, scanning, explanations or actual instructions. Compare that task with what the buyer needs. Do not assign a common return score to different products."
        ],
        [
          "Development value",
          "Tools can make experiments easier to organise while leaving research quality with the user. Confirm training and data access, version retention and execution assumptions. Convenience is not an independent edge certification."
        ],
        [
          "Performance evidence",
          "Ask for a causal validation protocol and the correct result layer: simulation, prospective observation or account execution. A selected backtest, credential or timestamp cannot supply missing model evidence."
        ],
        [
          "Keep the conclusion honest",
          "This desk has not run paid account trials or vendor training experiments. Its profiles support a documentation shortlist and precise follow-up questions. They do not promise that AI-labelled services outperform simpler rules after costs."
        ]
      ]
    ],
    [
      "primers/index.html",
      "AI research primers",
      "A path from product labels to a causal evaluation plan.",
      [
        [
          "Begin with the claim",
          "The taxonomy guide separates trained models, fixed rules, explanations and routing. Unknown model architecture remains unknown."
        ],
        [
          "Preserve the experiment",
          "The leakage guide and holdout checker help organise a declared plan without labelling it validated. Software tests of the checker are not performance samples."
        ],
        [
          "Read the correct evidence layer",
          "The backtest, forward and live guide keeps simulations and executed accounts apart. Product assessments describe documented capabilities, not fabricated AI returns."
        ]
      ]
    ],
    [
      "glossary.html",
      "AI evidence glossary",
      "Definitions used by this research desk, with product and evidence boundaries made explicit.",
      [
        [
          "Algorithmic and machine learning",
          "Algorithmic describes a procedure. ML describes a trained component. Automated routing and a language-model explanation need not be the decision model."
        ],
        [
          "Feature availability",
          "A feature's observation date, release time, receipt time and revision history can differ. Causal evaluation asks which value was actually available before a decision."
        ],
        [
          "Holdout and development",
          "An untouched holdout has not influenced experimental choices. Once its results shape a rule, its evidential role changes; renaming it does not restore independence."
        ],
        [
          "Version and retraining",
          "A version identifies a model artefact and its applicable pipeline. Retraining needs a recorded boundary and deployment time rather than retroactively assigning earlier predictions to an improved model."
        ],
        [
          "Backtest and forward record",
          "A backtest is historical simulation under assumptions. A prospective record retains outputs as they occur. Actual account results need executions and costs. These layers cannot substitute for one another."
        ],
        [
          "Receipt and confidence label",
          "A receipt concerns covered bytes and verified attestation, not trained-model quality. A confidence label is not automatically a calibrated probability, winning count or allocation rule."
        ]
      ]
    ],
    [
      "privacy.html",
      "Privacy and the holdout checker",
      "What the current static publication and client-side worksheet do, without unverified promises about host retention.",
      [
        [
          "Experiment-plan inputs",
          "The holdout checker evaluates dates and a version label in the user's browser. It contains no account connector or input-upload call; exporting downloads the entered plan and its untested classification. Use an experiment label, not a password or personal identifier."
        ],
        [
          "Host and font requests",
          "These pages are hosted on Cloudflare Pages. The page template also requests Google-hosted web fonts; those requests reach the respective services. The desk has not independently audited their logging, cookie configuration or retention. A static research interface is not a guarantee of zero third-party request processing."
        ],
        [
          "Other services",
          "Source links lead to independent product or documentation destinations with their own practices. This site has no registration interface and the checker does not submit a model to a vendor. The notice describes the implementation, not a completed legal compliance review or an assurance about a linked service."
        ]
      ]
    ]
  ],
  "legacy_corrections": [
    [
      "not an arithmetic|different bases|rounded.*(?:win|counts)|483|roughly.*wins|headline.*(?:sum|aggregate)",
      "The supplied 2026 Vector Ridge summary contains 690 signals, 70% wins and +1,227% raw aggregate. No complete dated ledger, current October cutoff or independently checked aggregation basis accompanies this desk assessment; these figures cannot establish out-of-sample AI performance."
    ],
    [
      "glass.box|best AI|top pick|#1|only.*(?:pick|provider|service)|one desk|clears all|all five|recommended (?:desk|service)|rule.*disclosed|disclosed.*rule",
      "An algorithmic subscription is not automatically an AI or machine-learning product. No supplied evidence establishes Vector Ridge's ML architecture, training procedure or untouched validation, so this site does not name it a verified AI winner."
    ],
    [
      "most.*(?:fail|don't|cannot)|by.*(?:architecture|built)|category.*(?:fails|failure)|no verifiable denominator|never left up|trips several",
      "Judge a named model or author, not an entire delivery channel. Chat feeds, scanners and marketplaces need their own retained forward records; no blanket fraud or performance verdict follows from the format."
    ]
  ],
  "tool": {
    "kind": "ai",
    "path": "tools/holdout-plan-checker.html",
    "title": "Holdout plan date checker",
    "description": "Check a simple calendar-day split and preserve a declared model version. This produces an untested plan, not a validation certificate.",
    "fields": [
      {
        "name": "version",
        "label": "Model version",
        "type": "text",
        "value": "demo-v1"
      },
      {
        "name": "trainEnd",
        "label": "Last training-row date",
        "type": "date",
        "value": "2025-12-31"
      },
      {
        "name": "horizon",
        "label": "Training outcome horizon (calendar days)",
        "value": 1,
        "min": 0,
        "step": 1
      },
      {
        "name": "testStart",
        "label": "Test period start",
        "type": "date",
        "value": "2026-01-02"
      },
      {
        "name": "testEnd",
        "label": "Test period end",
        "type": "date",
        "value": "2026-03-31"
      },
      {
        "name": "seen",
        "label": "Test outcomes have already influenced development",
        "type": "checkbox"
      }
    ],
    "sources": [
      [
        "QuantConnect optimization parameters",
        "https://www.quantconnect.com/docs/v2/writing-algorithms/optimization/parameters"
      ],
      [
        "ALFRED real-time and vintage data",
        "https://fred.stlouisfed.org/docs/api/fred/alfred.html"
      ]
    ],
    "sections": [
      [
        "The calendar condition",
        "The last training-row date plus the entered calendar-day outcome horizon must be earlier than the test start. Equal dates are not separated under this convention. The default example clears that limited chronology check, but it is only a planning illustration. Calendar days are not trading sessions, and the worksheet does not inspect a dataset."
      ],
      [
        "What 'unseen' means here",
        "The checkbox records the user's declaration about whether test outcomes influenced development. It cannot verify that declaration or recover a holdout after its results were used to improve a rule. Viewed evidence remains development evidence; a new model version does not erase exposure to it."
      ],
      [
        "Untested is a deliberate result",
        "Every valid export marks resultStatus as untested. Neither separated dates nor an unseen declaration certifies absence of leakage, statistical quality or tradability. Feature availability, overlapping labels, universe changes, costs and fills require separate investigation. The tool does not label a model PASS or publish an invented return."
      ],
      [
        "Keep a manifest",
        "Save the export with the feature specification, frozen data identity and predeclared evaluation method. A later change should create a distinct experiment record. This small calendar check can catch an obvious planning conflict, while the retained manifest helps a reviewer understand what was intended before outcomes were inspected."
      ]
    ]
  },
  "editorial_note": "AI claims need identified model architecture and causal evaluation evidence. Product documentation and planning worksheets do not certify an investable edge.",
  "general_proof": "A verified hash concerns its committed bytes, not model causality, complete prediction counts or profit. Calendar submission and confirmed attestation differ; declared output time and account fills require their own evidence.",
  "critical_sources": [
    [
      "QuantConnect optimization parameters",
      "https://www.quantconnect.com/docs/v2/writing-algorithms/optimization/parameters"
    ],
    [
      "ALFRED real-time and vintage data",
      "https://fred.stlouisfed.org/docs/api/fred/alfred.html"
    ],
    [
      "OpenTimestamps protocol documentation",
      "https://opentimestamps.org/"
    ]
  ],
  "commercial_note": "This AI research is educational. The desk has not performed an independent legal or commercial-relationship audit.",
  "expanded_protocols": {
    "ai-data-leakage-and-holdout-audit": {
      "description": "Audit feature availability, outcome overlap and test exposure with a frozen experiment record and a worked calendar-boundary example.",
      "sources": [
        [
          "scikit-learn: data leakage and preprocessing",
          "https://scikit-learn.org/stable/common_pitfalls.html"
        ],
        [
          "scikit-learn: probability calibration",
          "https://scikit-learn.org/stable/modules/calibration.html"
        ],
        [
          "CFTC: AI trading-bot consumer advisory",
          "https://www.cftc.gov/LearnAndProtect/AdvisoriesAndArticles/AITradingBots.html"
        ],
        [
          "ALFRED: vintage-data help",
          "https://alfred.stlouisfed.org/help"
        ]
      ],
      "columns": [
        "experiment_id",
        "model_version",
        "feature_version",
        "data_identity",
        "event_time",
        "publication_time",
        "receipt_time",
        "decision_time",
        "label_end",
        "train_end",
        "test_start",
        "test_end",
        "test_seen",
        "split_rule",
        "unresolved_question"
      ],
      "table": [
        [
          "Feature row",
          "Event, release, receipt and revision clocks",
          "A row date proves availability"
        ],
        [
          "Training target",
          "Last label observation and declared split",
          "Earlier row means separated target"
        ],
        [
          "Test exposure",
          "Experiment log and authorised readers",
          "Renaming restores unseen evidence"
        ]
      ],
      "sections": [
        [
          "Begin with one decision and its information set",
          "Choose an identifiable instruction rather than a polished aggregate chart. Record the decision clock, model artefact, feature specification and data identity used at that moment. Ask the seller which component actually produces the instruction: a fixed rule, trained predictor, explanatory language model or routing service. Unknown architecture should stay unknown, even when an application supports machine learning. A product capability and a particular deployed model are different objects.\n\nBuild a small availability ledger around that decision. For each input, retain what happened, when it was published, when the system received it and which revision was available. These clocks can disagree without anybody falsifying a record. A monthly observation label is not a release timestamp. A later corrected price file cannot silently become the historical file. This first audit step defines the question; it does not establish model accuracy, profitability or independent verification."
        ],
        [
          "Inspect transformations before inspecting scores",
          "A train/test separation can fail before the estimator is fitted. Ask how missing values, scaling, feature selection and category mappings were learned. The relevant distinction is between learning a transformation from development observations and applying that already learned transformation to evaluation observations. A retained pipeline specification helps identify the distinction, but its existence alone does not prove that a vendor followed it. Request run artefacts showing the actual data membership.\n\nConsider an invented review where a seller selected indicators using a full history and only then divided the rows into training and evaluation periods. The chronological labels on the resulting chart do not repair the earlier selection exposure. Preserve that experiment as development evidence and state the exact flaw. Do not claim that every later model is therefore fraudulent or useless. A replacement experiment requires its own data boundary, identity and untouched evaluation plan, not a retrospective change of headings."
        ],
        [
          "Work through an overlapping-label boundary",
          "Here is a fictional calendar check, not a tested strategy. A last training row dated 31 December 2025 has a target horizon of two calendar days. Its declared target reaches 2 January 2026. If evaluation starts on 2 January, the target and evaluation touch under the worksheet convention requiring the label end to be strictly earlier than the test start. Moving that start to 3 January clears only this simple comparison. It says nothing about exchange sessions or other overlapping rows.\n\nWrite the actual label construction down before choosing a purge or gap. A target can involve a variable exit, a maximum observation window or another instrument clock; a fixed number of calendar days may describe none of them. The date checker deliberately exports an untested plan. It does not inspect samples, prove feature causality or certify adequate separation. Keep timezone, label endpoint and boundary equality visible so a reviewer can reproduce precisely the limited arithmetic being offered."
        ],
        [
          "Preserve what each experiment exposed",
          "Create an append-only experiment index containing the candidate identity, data periods viewed, questions asked and decisions influenced. A test has become development evidence when its outcomes guide another threshold, feature or model choice. A different filename or retraining date does not make that same period unseen again. Record unsuccessful experiments as well as the candidate later shown to consumers. Otherwise a final performance figure conceals the search process that produced it.\n\nFreeze the entry objective and candidate identity before applying the predeclared exit evaluation. Exit-adjusted profit, holding drift and realised win rate can inform a later validation or rejection decision; they should not quietly redefine which entry candidates won the original search. Preserve the authorised hypothetical research-fill convention separately from broker execution. Changing the fill window after seeing outcomes creates a different experiment. A hash can identify retained bytes, but cannot prove that no unlogged experiment happened or that the declared access policy was obeyed."
        ],
        [
          "Turn a missing artefact into a specific question",
          "Use four distinct statuses: requested, supplied, inspected and unresolved. A link to an API manual is supplied documentation, not an inspected model run. A diagram describing training is not the training artefact. If the model weights are commercially sensitive, ask for a narrower audit object such as dated version identifiers, a feature-timing specification or a redacted experiment ledger. Explain what each object would answer without claiming that one object completes the entire investigation.\n\nPreserve the seller response with its date and exact product scope. An honest inability to share a file is still missing evidence for the claim requiring that file. Avoid converting silence into an invented score or assuming that a founder credential supplies the missing experiment history. ALFRED can help investigate vintage differences, but it does not provide the intraday receipt log of an arbitrary system. Consumer review language should describe that limitation plainly enough that a reader can ask the next useful question."
        ],
        [
          "Finish with a bounded audit conclusion",
          "A useful conclusion names the strongest layer actually reviewed. For example: documentation describes a training boundary, a fictional date exercise reproduces the stated comparison, and real feature receipts remain unavailable. That is more actionable than a universal clean-data badge. Separate software correctness, historical simulation, prospective observation and executed account evidence in the final record. Passing a worksheet test demonstrates the worksheet behaviour, not an independent strategy-validation sample.\n\nBefore paying, compare the unresolved items with the capability you need. A consumer seeking a reproducible experiment may reasonably require different evidence from somebody purchasing market explanations only. Do not turn the checklist into an allocation recommendation. Save the blank CSV with private identifiers removed before sharing it, and date later corrections rather than rewriting the original assessment. This publication supplies a research framework and linked documentation; it has not inspected a private model dataset, established a new holdout or verified the results described by the seller in an advertisement."
        ]
      ]
    },
    "backtest-forward-and-live-evidence": {
      "description": "Reconcile simulated decisions, prospective instructions and account fills without promoting one evidence layer into another.",
      "sources": [
        [
          "scikit-learn: data leakage and preprocessing",
          "https://scikit-learn.org/stable/common_pitfalls.html"
        ],
        [
          "scikit-learn: probability calibration",
          "https://scikit-learn.org/stable/modules/calibration.html"
        ],
        [
          "CFTC: AI trading-bot consumer advisory",
          "https://www.cftc.gov/LearnAndProtect/AdvisoriesAndArticles/AITradingBots.html"
        ],
        [
          "ALFRED: vintage-data help",
          "https://alfred.stlouisfed.org/help"
        ]
      ],
      "columns": [
        "record_id",
        "evidence_layer",
        "model_version",
        "decision_id",
        "decision_time",
        "delivery_time",
        "order_id",
        "order_time",
        "fill_time",
        "requested_quantity",
        "filled_quantity",
        "status",
        "cost_basis",
        "cash_flow",
        "missing_record"
      ],
      "table": [
        [
          "Simulation",
          "Data, fill window, costs and model identity",
          "A subscriber received that price"
        ],
        [
          "Prospective record",
          "Original, amended and unresolved decisions",
          "Every message became an order"
        ],
        [
          "Account outcome",
          "Orders, fills, cash flows and equity basis",
          "A screenshot is a complete ledger"
        ]
      ],
      "sections": [
        [
          "Identify the claim before collecting files",
          "Write the advertised claim in a single sentence with its period, account or model identity and result convention. Historical model performance, prospective signal publication and subscriber profit require different records. If a seller says live, ask whether that means a running software process, an observation stream or actual funded execution. Do not choose the most favourable interpretation on behalf of the seller. Retain the original wording beside the review question.\n\nStart separate folders or ledger fields for each evidence layer. A backtest report belongs with its dataset and execution convention. A prospective record belongs with its publication history. An account statement belongs with deposits, withdrawals, orders and fills. Related records can be reconciled through stable identifiers without being merged into one ambiguous percentage. This separation makes a missing fill or amendment visible. It also prevents a genuine historical simulation from being dismissed simply because it is not something it never claimed to be."
        ],
        [
          "Reproduce the simulation convention, not the sales chart",
          "Ask which instrument histories, universe membership, timezone and corporate-action conventions were used. Retain the model version, entry objective, chosen exit and treatment of absent observations. Costs need units and timing, not just a word such as realistic. A favourable research fill under a declared observation window remains a hypothetical reference; it cannot be described as the price every subscriber obtained. Preserve the frozen convention when comparing releases.\n\nAn invented example contains twenty closed hypothetical decisions, fourteen labelled positive and six negative, plus five still open. The closed-only positive fraction is fourteen divided by twenty, or seventy percent. Twenty-five total decisions answer a different coverage question. The example supplies no actual trading result and says nothing about net return, position weights or the final outcomes of the open items. Keep the unresolved rows instead of forcing them into wins or dropping them from the instruction count. A chart alone cannot reveal these denominators."
        ],
        [
          "Count prospective instructions before counting profits",
          "A forward observation should retain decisions when they appear, including those later cancelled, expired, rejected or left unresolved. Record a model deployment change at the time it happens rather than assigning an earlier prediction to a better later artefact. Distinguish source publication from subscriber receipt: a timestamp attached to one does not prove the other. Corrections should link to originals rather than overwrite them invisibly.\n\nSuppose a fictional stream contains twelve messages: four commentaries, six actionable originals and two amendments to those originals. There are six original instructions, not twelve independent trades. If three of those six reached an account and three did not, execution coverage in this toy set is one half. That is not a profitability estimate. The amendment relationship and each unexecuted reason matter. A lost permission, stale instruction or missing acknowledgement belongs to an operating record, even if no profit can yet be calculated for it."
        ],
        [
          "Reconcile execution states without assuming a fill",
          "For every actionable instruction, map the requested order to acknowledgements, partial fills, cancellations and remaining quantity. A timeout is an unknown state until reconciled, not proof that an order never existed. An accepted order is not necessarily an executed order. Keep event ordering and venue identifiers where available, with private account information redacted in a public review. Do not place real orders to demonstrate this guide.\n\nA hypothetical request for one unit can fill 0.4 units and cancel the remaining 0.6. The account still owns 0.4 units; cancellation is not an exit. A later confirmed sale of 0.25 leaves 0.15. These arithmetic states illustrate why a source signal and an account trade can diverge. Fees, financing, currency conversion and mark-to-market conventions must then be applied to the actual quantities and clocks. No generic platform description proves that every provider or subscriber records those states correctly."
        ],
        [
          "Keep cash flow and model identity in the result",
          "An account value can rise after a deposit even when the trading result is flat or negative. Retain externally contributed and withdrawn cash with its timing. Name the return convention used and avoid presenting a raw balance ratio as a cash-flow-adjusted result. An account-level measure and an unweighted sum of model percentages are not interchangeable. If the aggregation method is unavailable, report the published number as an attributed claim rather than independently reproduced performance.\n\nSimilarly, a family label can cover several model versions. Preserve the sequence of versions, retirement dates and evaluation rules. A later corrected dataset should not silently alter the old release receipt. If a source changes its record, keep both the old assessment and the dated correction. Do not infer a complete population from one selected successful account. The purpose of reconciliation is to expose what the records support and where they stop, not to manufacture a universally comparable league table."
        ],
        [
          "Write an evidence-layer decision for the reader",
          "End the assessment with the reviewed period, supplied artefacts, missing states and the strongest supported statement. Documentation-only research remains documentation-only even if the interface looks convincing. A preserved prospective stream can support publication chronology without proving funded execution. A reconciled account can establish a specific account experience without forecasting the next subscriber outcome. Say which boundary applies instead of using an unqualified verified label.\n\nUse the checkpoint table to prioritise the next request. A consumer may need an actual cost ledger more urgently than another screenshot of gross profit. Someone comparing research platforms may instead need stable experiment identities and export formats. Preserve private information locally and share only redacted references. These pages have not run a paid model, subscribed a funded account or verified a live return. Their value is a reproducible way to distinguish evidence objects and prevent one valid object from being used to imply several unsupported conclusions."
        ]
      ]
    },
    "algorithmic-versus-ai-signal-claims": {
      "description": "Map fixed rules, trained predictions, AI explanations and order routing to the specific consumer evidence each needs.",
      "sources": [
        [
          "scikit-learn: data leakage and preprocessing",
          "https://scikit-learn.org/stable/common_pitfalls.html"
        ],
        [
          "scikit-learn: probability calibration",
          "https://scikit-learn.org/stable/modules/calibration.html"
        ],
        [
          "CFTC: AI trading-bot consumer advisory",
          "https://www.cftc.gov/LearnAndProtect/AdvisoriesAndArticles/AITradingBots.html"
        ],
        [
          "ALFRED: vintage-data help",
          "https://alfred.stlouisfed.org/help"
        ]
      ],
      "columns": [
        "product_name",
        "offer_date",
        "component_id",
        "component_role",
        "input_type",
        "output_type",
        "training_claim",
        "model_identity",
        "probability_claim",
        "validation_record",
        "routing_record",
        "contract_scope",
        "unknown_capability"
      ],
      "table": [
        [
          "Decision engine",
          "Rule or trained artefact and input timing",
          "Automation establishes machine learning"
        ],
        [
          "Probability claim",
          "Definition, evaluation period and calibration record",
          "A confidence label predicts success"
        ],
        [
          "Explanation and routing",
          "Original output, revision and order mapping",
          "Persuasive prose proves execution"
        ]
      ],
      "sections": [
        [
          "Buy a named capability, not an elastic label",
          "Before comparing products, complete the sentence: I am paying for this component to perform this task. A market summary, a predictive score, an actionable instruction and an account-routing service are four different purchases. A vendor can offer several together, but each needs its own evidence and limitations. The word AI does not establish which task is trained, which is deterministic or which is added after the decision. Record unknown components instead of guessing their architecture.\n\nAsk for a dated representative output and the product terms covering that output. A platform able to train models may also host ordinary threshold alerts; its general feature page does not validate a particular strategy. A consumer directory should compare the specific offers, not rank unrelated capabilities as though they are substitutes. This guide deliberately does not create an AI quality percentage. The distinction starts with product identity and ends with what the available records actually establish."
        ],
        [
          "Describe a fixed rule without treating it as inferior",
          "A deterministic rule can be systematic, exportable and useful without learning statistical parameters. Request its input definitions, decision clock, thresholds, expiry and amendment policy. Ask how the same rule behaves when data are missing or revised. Reproducibility depends on those details as well as the expression printed on a product page. A fixed rule can still be overfit through human experimentation, so absence of machine learning is not proof of independence from past outcomes.\n\nFor a fictional example, an alert fires when a defined observed value exceeds a threshold selected before the observation period. Its routing system sends the alert to a phone. Neither the threshold nor delivery requires a trained predictor in this example. Calling the combination automated is accurate; calling it calibrated forecasting would require additional evidence. Preserve the rule version if the threshold later changes. Compare the rule with its intended function rather than awarding or withholding credibility merely because its marketing omits fashionable terminology."
        ],
        [
          "Inspect a trained model at the prediction boundary",
          "A trained predictor requires an identified artefact, target definition and account of how inputs become outputs. Consumers do not need public access to proprietary weights to ask when training stopped, what the output represents and how the evaluation was separated from development. Keep feature and model versions together; a stable commercial brand can conceal changing models. Distinguish retraining cadence from the actual deployment time of each artefact.\n\nImagine a seller output labelled 0.8. It might be a probability, a transformed score or an arbitrary ranking value. Until the definition is supplied, none of those meanings is established. Even a probability claim requires an outcome definition and suitable evaluation evidence. A hypothetical bin containing ten predictions and six observed successes illustrates a frequency of sixty percent in that tiny invented bin, not proof of calibration or a future success chance. Small, dependent or selectively retained observations cannot support a strong universal accuracy statement."
        ],
        [
          "Treat explanations as explanations",
          "An explanatory language model may describe an event or translate another engine output into ordinary language. Ask whether the explanation was produced before the instruction, alongside it or afterwards. A plausible narrative generated after a profitable move does not prove that the original model anticipated that move. Preserve the raw decision and explanation as separate dated objects, especially when a system can edit summaries after source data change.\n\nConsider a fictional alert with a numeric decision and an accompanying paragraph naming three reasons. If the paragraph was generated after the decision, those reasons are not automatically the variables the decision engine used. Request an actual feature mapping before describing them as model attribution. A readable explanation can still be a useful consumer feature without validating the forecast. This publication does not inspect hidden reasoning or certify that vendor narratives faithfully represent proprietary models. It asks for the relationship between the objects a reader can actually examine."
        ],
        [
          "Audit routing independently of forecasting",
          "A webhook, bot or copy connection can transport an instruction regardless of how that instruction was generated. Trace original publication, delivery, acknowledgement and account action separately. Check duplicate identifiers, stale messages, permission loss and reconciliation after an interrupted connection. Reliable delivery and reliable prediction are different properties; one cannot stand in for the other in a review headline. Keep operating records separate from the statistical evaluation.\n\nA fictional message sent twice with the same instruction identifier should be investigated as one instruction with duplicate delivery, not two independent trading opportunities. If the first account submission times out, the operating state remains unknown until checked; retrying blindly can create duplicate exposure. This guide is not a deployment runbook and does not connect a broker. It establishes the questions a consumer should ask before assuming that an appealing prediction interface includes an equally well evidenced execution service or a complete history of its failures."
        ],
        [
          "Use a scope-aware comparison verdict",
          "Build the comparison around task, evidence, access, operating responsibilities and total contractual cost. Record any limits on export, supported accounts and data access from the exact offer rather than transferring a claim from another subscription. A lower price is not evidence of a worse predictor, and a high founder return is not evidence of better routing. Avoid a single synthetic ranking when the products answer different consumer needs. Unknown capability is a useful finding, not a blank to fill with favourable prose.\n\nThe CFTC advisory is a relevant caution about extravagant AI profit claims; it does not establish that a particular listed seller is fraudulent. Retain the original claim and the missing evidence without unsupported allegations. A reader can choose to request clarification, use a free alternative or decline the purchase. None of those options requires an invented consensus score. This taxonomy supplies a decision vocabulary and blank record, not a personal trading recommendation, new model test or promise that AI will improve investment outcomes."
        ]
      ]
    }
  },
  "edition": "2026-10-13",
  "disclosure": "The owner describes these publications as legally independently owned and asks that their useful research support awareness of Vector Ridge and Darren O'Neill. That promotional purpose is disclosed here. It does not establish independent editorial direction, no sponsorship or no referral relationship. Those arrangements need an accountable register, and no preferred product receives an unsupported AI-validation verdict."
}
