The BayesBoard JSON format

A board is one JSON file. You can write it by hand, have an AI agent generate it, or export one from the editor and edit it. This page says what each field means, gives three boards to copy, and explains the errors the importer gives.

Which way the arrow goes

The arrow goes from what informs to what is judged. “Evidence E supports hypothesis H” is drawn E → H.

In the file that is an edge with "source": "E1" and "target": "H1". The edge kind sets the direction of the effect: supports and tests push the target up; weakens, falsifies and contradiction push it down. You always enter the strength as 1 or more; the kind decides up or down. An edge drawn H → E asks what the hypothesis says about the evidence, which is almost never what you mean.

How strong: the house bands

Strength is a likelihood ratio: how many times more likely you would be to see this evidence if the hypothesis were true than if it were false. A ratio of 1 means no information. Use these bands when you are unsure what number to write:

BandStrength (supports)TypicalSame strength against
weak1.2 to 1.5×1.3420.667 to 0.833
moderate1.6 to 2.5×20.4 to 0.625
strong2.6 to 5×3.6060.2 to 0.385
decisivemore than 5×10below 0.2

In the file, write the strength of an edge against a claim the same way, as 1 or more, and use a weakens, falsifies or contradiction edge: { "type": "weakens", "confidenceImpact": "×2" }. The bands are deliberately gentler than the classic 3, 10 and 100 cut-offs, because strengths from many pieces of evidence add up. The editor flags a single piece above 100 as implausibly strong.

Fields, one by one

Required means the schema requires it. The importer is more forgiving than the schema for older files (see Import errors), but a file that follows the schema always works. The full machine-readable version is the JSON Schema.

The document (document)

One JSON object. The five lists must be present, even when empty.

FieldTypeRequiredMeaning
versiontextyesThe schema version, "1.3" for new files. Any "1.x" is accepted on import, so "1.2" files keep working.
metaobjectyesAbout the board itself: name (required), created and updated (ISO 8601 timestamps, required), description, author and tags (optional).
configobjectyesBoard-level engine settings. Today only channelTrust (see below).
hypotheseslistyesThe claims you are judging. Can be empty.
experimentslistyesTests you run or have run. Can be empty.
evidencelistyesThings you observed. Can be empty.
outcomeslistyesWhat you care about in the end. Can be empty.
edgeslistyesThe arrows between cards, with their strength. This is where the numbers live.
presetslistnoNamed what-if scenarios that mute several cards or edges at once (new in 1.3).
abouttextnoThe board story, as Markdown, shown in the About tab. Never HTML.
aboutVisibility"public" or "group"noWho may read the About. Defaults to public.
docslistnoDocs kept on the board itself: id, title, position, body (Markdown), version, visibility, updatedAt.
updateslistnoThe engine audit log from the last evaluation (one record per card). Written by exports; safe to leave out, it is recomputed.
annotationslistnoCanvas drawings (boxes, text, arrows). Written by the editor; leave out when generating a board.

Hypothesis hypotheses[]

A claim that can be true or false. Specific and falsifiable beats vague: "At least 30% of trial users return within 7 days", not "the product is good".

FieldTypeRequiredMeaning
idtextyesUnique across the whole board. Edges point at it by this id.
type"hypothesis"yesAlways the word hypothesis.
domaintextyesA free label to group cards, such as "product" or "finance".
nametextyesThe short title on the card.
statementtextyesThe claim in one full sentence.
priornumber 0-100noYour belief in percent BEFORE any evidence. If missing, the confidence value is used as the prior.
confidencenumber 0-100yesWhere the board stands now, in percent. The engine rewrites it from the prior and the edges, so a rough value is fine when generating a file.
statusone of pending, testing, validated, rejected, revisedyesWhere the claim is in its life. A label only; it does not change the math.
criticalityone of LOW, MEDIUM, HIGH, CRITICALyesHow much the decision depends on this claim. A label only.
validationLevelwhole number 0-5yesHow well tested it is. A label only.
dependencieslist of idsyesOther cards this one leans on. Can be an empty list. The arrows that count are in edges.
experimentGateslist of idsyesExperiments that gate this claim. Can be an empty list.
timelinetextyesWhen it must be settled. Can be an empty string.
metadataobjectyescreated and updated (ISO 8601), author, tags (list of text). All four are required inside it.
positionobject {x, y}yesWhere the card sits on the canvas, in pixels. Any numbers work; boards are laid out around 0 to 1000.
kindone of hypothesis, evidence, outcomenoEngine kind. Leave out; it defaults to hypothesis.
notestextnoFree notes shown in the side panel.
alternativeStatestextnoWhat the world looks like if the claim is false.
assignedTotextnoWho owns checking it.
referencestextnoWhere to read more.
hiddentrue or falsenotrue mutes the card: it stops counting as a source of evidence for others. Its own value is still computed.

Experiment experiments[]

A test you run. Write the question, the hypothesis it tests, the method and a threshold in acceptanceCriteria. When it has run, put the result in result AND record it as an evidence card whose resultOf names this experiment: from then on the experiment's own edges stop counting, so the result is counted once (see the experiment example).

FieldTypeRequiredMeaning
idtextyesUnique across the whole board.
type"experiment"yesAlways the word experiment.
nametextyesThe short title.
descriptiontextyesThe question and the method.
hypothesesTestedlist of idsyesThe hypotheses this experiment tests. Can be an empty list.
statusone of pending, active, running, completed, complete, failed, deferredyesWhere the experiment is. A label only.
positionobject {x, y}yesWhere the card sits on the canvas, in pixels. Any numbers work; boards are laid out around 0 to 1000.
acceptanceCriteriatextnoThe threshold you decided on before running it, for example "at least 8 points higher".
resulttextnoWhat actually happened, in words and numbers.
priornumber 0-100noBelief in percent before evidence. Experiments are read like hypotheses when edges point into them.
confidencenumber 0-100noCurrent value in percent. Recomputed by the engine.
completionnumber 0-100noHow far along it is.
timelinetextnoHow long it takes.
priorityone of LOW, MEDIUM, HIGH, CRITICALnoHow urgent it is. A label only.
preregistereddate YYYY-MM-DDnoThe day you wrote down the threshold, before looking at results.

Evidence evidence[]

Something you observed. An evidence card has no probability of its own; its strength lives on its outgoing edges.

FieldTypeRequiredMeaning
idtextyesUnique across the whole board.
type"evidence"yesAlways the word evidence.
titletextyesWhat you saw, in one line.
channelone of experimental, expert, document, data, intuitionyesWhere it came from. The board trust for that channel scales how much it counts (see channelTrust).
strengthnumber 0-1yesA dial on this one card: 1 counts it fully, 0.5 counts half as much in log-odds. Not the likelihood ratio; that is on the edge.
positionobject {x, y}yesWhere the card sits on the canvas, in pixels. Any numbers work; boards are laid out around 0 to 1000.
descriptiontextnoThe details and the source.
sourceUrltextnoA link to the source.
provenanceobjectnoSource quality notes (new in 1.3): sourceType, sourceUrl, citation, date, sampleSize, studyDesign, riskOfBias, directness, populationMatch, interventionMatch, outcomeMatch, precision, independence, notes, excerpt. All optional.
hiddentrue or falsenotrue mutes the card: it is kept but does not count.
resultOfidnoThe experiment this card is the recorded result of (new in 1.3). That experiment then stops counting on its own, so the result is counted once. The editor's Record result sets it.

Outcome outcomes[]

What you care about in the end. It gets a probability from the cards pointing at it, like a hypothesis.

FieldTypeRequiredMeaning
idtextyesUnique across the whole board.
type"outcome"yesAlways the word outcome.
titletextyesThe short title.
statusopen or closedyesWhether it has happened yet.
priornumber 0-100yesBelief in percent before any evidence.
confidencenumber 0-100yesCurrent value in percent. Recomputed by the engine.
positionobject {x, y}yesWhere the card sits on the canvas, in pixels. Any numbers work; boards are laid out around 0 to 1000.
descriptiontextnoWhat the outcome means.
hiddentrue or falsenotrue mutes the card.

Edge (the arrow) edges[]

The arrow goes from what informs to what is judged. "Evidence E supports hypothesis H" is drawn E to H: source E, target H. The kind sets the direction of the effect, the strength is always entered as 1 or more.

FieldTypeRequiredMeaning
idtextyesUnique among edges.
sourceidyesThe card that informs: the evidence, experiment or supporting claim.
targetidyesThe card being judged.
dataobjectyesThe kind and the strength (next two rows).
data.typeone of supports, weakens, tests, falsifies, conditional-on, dependency, contradictionyesThe kind. See the table of kinds below.
data.confidenceImpacttextyesThe strength as a likelihood ratio, written "×2", "*2", "5:1" or "2". Use 1 or more: the kind decides up or down. "×1" means no effect. "+35%" applies a fixed shift instead.
data.conditiontextnoA note on when the edge applies. Required in practice for conditional-on.
data.transferobjectnoHow directly the evidence applies here (new in 1.3): directness, populationMatch, interventionMatch, outcomeMatch, transferMultiplier (0-1, overrides the rest), rationale. All optional.
impactobjectnoA numeric copy of the strength for tools: kind (likelihoodRatio, fixedDelta or manualOverride), value, explanation. Import reads data.confidenceImpact, not this.
sourceHandle / targetHandletextnoWhich side of each card the line attaches to. Leave out.

Channel trust config.channelTrust

How much you trust each kind of source, from 0 to 1. It scales every card of that channel. Gut feeling at 0.2 counts for a fifth of what its likelihood ratio says; 0 is the same as muting the whole channel.

FieldTypeRequiredMeaning
experimentalnumber 0-1yesResults of tests you ran.
expertnumber 0-1yesOpinions of people who know the field.
documentnumber 0-1yesReports, papers, contracts.
datanumber 0-1yesMeasurements and analytics.
intuitionnumber 0-1yesGut feeling.

Presets (muted what-ifs) presets[]

A named scenario that mutes several things together, to see how much a conclusion depends on them. To mute one card in the file itself, set hidden to true on it.

FieldTypeRequiredMeaning
nametextyesFor example "Without the founder gut".
mutedNodeIdslist of idsnoCards muted in this scenario.
mutedEdgeIdslist of idsnoSingle edges muted in this scenario.

Edge kinds

KindEffect on targetMeaning
supportsraisesThe source makes the target more likely. The most common kind.
testsraisesAn experiment tests the target. Counts like supports, until a result is recorded as evidence (resultOf): then the experiment stops counting and the result counts instead.
weakenslowersThe source makes the target less likely. A softer "against".
falsifieslowersThe source is a direct counter-example. A hard "against".
contradictionlowersThe source cannot hold together with the target.
dependencyraisesThe target only matters if the source is probably true. Conducts nothing while the source is below 50%.
conditional-onraisesLike dependency, with a condition written in data.condition.

Three examples to copy

Each one imports cleanly and evaluates; an automated test runs all three through the real importer on every change. The files are also at /examples/json/.

Minimal: one hypothesis, one evidence, one supports edge

The smallest board that evaluates. Evidence E1 supports hypothesis H1 at about ×3, so a 40% prior moves to about 67%.

{
  "version": "1.3",
  "meta": {
    "name": "Minimal board",
    "description": "One hypothesis, one piece of evidence, one supports edge.",
    "author": "[email protected]",
    "created": "2026-11-15T09:00:00.000Z",
    "updated": "2026-11-15T09:00:00.000Z",
    "tags": [
      "example"
    ]
  },
  "config": {
    "channelTrust": {
      "experimental": 1,
      "expert": 0.8,
      "document": 0.9,
      "data": 1,
      "intuition": 0.5
    }
  },
  "hypotheses": [
    {
      "id": "H1",
      "type": "hypothesis",
      "domain": "product",
      "name": "Trial users come back",
      "statement": "At least 30% of trial users return within 7 days.",
      "criticality": "HIGH",
      "status": "testing",
      "confidence": 67,
      "prior": 40,
      "validationLevel": 1,
      "position": {
        "x": 420,
        "y": 120
      },
      "dependencies": [],
      "experimentGates": [],
      "timeline": "",
      "notes": "",
      "metadata": {
        "created": "2026-11-15T09:00:00.000Z",
        "updated": "2026-11-15T09:00:00.000Z",
        "author": "[email protected]",
        "tags": []
      }
    }
  ],
  "experiments": [],
  "evidence": [
    {
      "id": "E1",
      "type": "evidence",
      "title": "Pilot cohort: 14 of 40 trial users returned within 7 days",
      "description": "Product analytics export, first pilot cohort (35% return rate).",
      "channel": "data",
      "strength": 1,
      "position": {
        "x": 60,
        "y": 120
      }
    }
  ],
  "outcomes": [],
  "edges": [
    {
      "id": "E1-H1",
      "source": "E1",
      "target": "H1",
      "data": {
        "type": "supports",
        "confidenceImpact": "×3"
      }
    }
  ]
}

Download minimal.json

Full: every card type, every edge kind, channel trust, muted items

Three hypotheses, an experiment, four evidence cards (one muted with hidden: true), an outcome, all seven edge kinds, a lowered intuition trust and two mute presets.

{
  "version": "1.3",
  "meta": {
    "name": "Full board",
    "description": "Every node type, every edge kind, channel trust and muted items.",
    "author": "[email protected]",
    "created": "2026-11-15T09:00:00.000Z",
    "updated": "2026-11-15T09:00:00.000Z",
    "tags": [
      "example"
    ]
  },
  "config": {
    "channelTrust": {
      "experimental": 1,
      "expert": 0.8,
      "document": 0.9,
      "data": 1,
      "intuition": 0.2
    }
  },
  "hypotheses": [
    {
      "id": "H1",
      "type": "hypothesis",
      "domain": "product",
      "name": "Trial users come back",
      "statement": "At least 30% of trial users return within 7 days.",
      "criticality": "CRITICAL",
      "status": "testing",
      "confidence": 63,
      "prior": 40,
      "validationLevel": 1,
      "position": {
        "x": 520,
        "y": 60
      },
      "dependencies": [],
      "experimentGates": [
        "EXP1"
      ],
      "timeline": "",
      "notes": "",
      "metadata": {
        "created": "2026-11-15T09:00:00.000Z",
        "updated": "2026-11-15T09:00:00.000Z",
        "author": "[email protected]",
        "tags": []
      }
    },
    {
      "id": "H2",
      "type": "hypothesis",
      "domain": "product",
      "name": "Onboarding drives the return",
      "statement": "The new onboarding checklist, not the discount, is what brings trial users back.",
      "criticality": "HIGH",
      "status": "testing",
      "confidence": 73,
      "prior": 50,
      "validationLevel": 1,
      "position": {
        "x": 520,
        "y": 300
      },
      "dependencies": [
        "H1"
      ],
      "experimentGates": [],
      "timeline": "",
      "notes": "",
      "metadata": {
        "created": "2026-11-15T09:00:00.000Z",
        "updated": "2026-11-15T09:00:00.000Z",
        "author": "[email protected]",
        "tags": []
      }
    },
    {
      "id": "H3",
      "type": "hypothesis",
      "domain": "product",
      "name": "Support load stays manageable",
      "statement": "Support tickets per trial user stay below 0.2 per week.",
      "criticality": "MEDIUM",
      "status": "testing",
      "confidence": 57,
      "prior": 50,
      "validationLevel": 1,
      "position": {
        "x": 520,
        "y": 540
      },
      "dependencies": [],
      "experimentGates": [],
      "timeline": "",
      "notes": "",
      "metadata": {
        "created": "2026-11-15T09:00:00.000Z",
        "updated": "2026-11-15T09:00:00.000Z",
        "author": "[email protected]",
        "tags": []
      }
    }
  ],
  "experiments": [
    {
      "id": "EXP1",
      "type": "experiment",
      "name": "Two-week onboarding A/B",
      "description": "Half of new trial users see the checklist, half do not.",
      "hypothesesTested": [
        "H1",
        "H2"
      ],
      "status": "running",
      "position": {
        "x": 160,
        "y": 300
      },
      "acceptanceCriteria": "Return rate in the checklist arm is at least 8 points higher.",
      "timeline": "2 weeks",
      "priority": "HIGH",
      "prior": 50,
      "confidence": 50,
      "completion": 40
    }
  ],
  "evidence": [
    {
      "id": "E1",
      "type": "evidence",
      "title": "Pilot cohort: 14 of 40 returned within 7 days",
      "description": "First pilot cohort, product analytics export.",
      "channel": "data",
      "strength": 1,
      "position": {
        "x": -200,
        "y": 40
      }
    },
    {
      "id": "E2",
      "type": "evidence",
      "title": "Churn-risk interview with a trial user",
      "description": "One user said the empty first screen made them leave.",
      "channel": "expert",
      "strength": 0.8,
      "position": {
        "x": -200,
        "y": 200
      }
    },
    {
      "id": "E3",
      "type": "evidence",
      "title": "Competitor teardown: checklists lift retention",
      "description": "A public write-up by a comparable product.",
      "channel": "document",
      "strength": 0.7,
      "position": {
        "x": -200,
        "y": 360
      },
      "sourceUrl": "https://example.com/teardown"
    },
    {
      "id": "E4",
      "type": "evidence",
      "title": "Founder gut: this will not move the needle",
      "description": "A hunch with no data behind it. Muted so it does not count.",
      "channel": "intuition",
      "strength": 1,
      "position": {
        "x": -200,
        "y": 520
      },
      "hidden": true
    }
  ],
  "outcomes": [
    {
      "id": "O1",
      "type": "outcome",
      "title": "Extend the trial program",
      "description": "Fund a second quarter of the trial program.",
      "status": "open",
      "confidence": 63,
      "prior": 50,
      "position": {
        "x": 900,
        "y": 300
      }
    }
  ],
  "edges": [
    {
      "id": "E1-H1",
      "source": "E1",
      "target": "H1",
      "data": {
        "type": "supports",
        "confidenceImpact": "×3"
      }
    },
    {
      "id": "E2-H2",
      "source": "E2",
      "target": "H2",
      "data": {
        "type": "supports",
        "confidenceImpact": "×1.8"
      }
    },
    {
      "id": "E3-H2",
      "source": "E3",
      "target": "H2",
      "data": {
        "type": "supports",
        "confidenceImpact": "×1.4"
      }
    },
    {
      "id": "E4-H1",
      "source": "E4",
      "target": "H1",
      "data": {
        "type": "weakens",
        "confidenceImpact": "×2"
      }
    },
    {
      "id": "EXP1-H2",
      "source": "EXP1",
      "target": "H2",
      "data": {
        "type": "tests",
        "confidenceImpact": "×1.5"
      }
    },
    {
      "id": "E2-H1",
      "source": "E2",
      "target": "H1",
      "data": {
        "type": "falsifies",
        "confidenceImpact": "×1.3"
      }
    },
    {
      "id": "H1-O1",
      "source": "H1",
      "target": "O1",
      "data": {
        "type": "supports",
        "confidenceImpact": "×2.5"
      }
    },
    {
      "id": "H3-O1",
      "source": "H3",
      "target": "O1",
      "data": {
        "type": "contradiction",
        "confidenceImpact": "×1.5"
      }
    },
    {
      "id": "H1-H2",
      "source": "H1",
      "target": "H2",
      "data": {
        "type": "dependency",
        "confidenceImpact": "×1"
      }
    },
    {
      "id": "H2-H3",
      "source": "H2",
      "target": "H3",
      "data": {
        "type": "conditional-on",
        "confidenceImpact": "×1.5",
        "condition": "Only if the checklist does not add support questions."
      }
    }
  ],
  "presets": [
    {
      "name": "Without the founder gut",
      "mutedNodeIds": [
        "E4"
      ]
    },
    {
      "name": "Data only",
      "mutedNodeIds": [
        "E2",
        "E3",
        "E4"
      ]
    }
  ]
}

Download full.json

Experiment with a result

The experiment records its result in words, and the result is also an evidence card (channel experimental, resultOf EXP1) that carries the strength. Because E1 names EXP1 in resultOf, the experiment's own edge no longer counts, so the result is counted once.

{
  "version": "1.3",
  "meta": {
    "name": "Experiment with a result",
    "description": "A finished experiment whose result is recorded as evidence and counted once.",
    "author": "[email protected]",
    "created": "2026-11-15T09:00:00.000Z",
    "updated": "2026-11-15T09:00:00.000Z",
    "tags": [
      "example"
    ]
  },
  "config": {
    "channelTrust": {
      "experimental": 1,
      "expert": 0.8,
      "document": 0.9,
      "data": 1,
      "intuition": 0.5
    }
  },
  "hypotheses": [
    {
      "id": "H1",
      "type": "hypothesis",
      "domain": "product",
      "name": "The checklist lifts the return rate",
      "statement": "The onboarding checklist raises 7-day return by at least 8 percentage points.",
      "criticality": "CRITICAL",
      "status": "validated",
      "confidence": 70,
      "prior": 40,
      "validationLevel": 1,
      "position": {
        "x": 520,
        "y": 140
      },
      "dependencies": [],
      "experimentGates": [
        "EXP1"
      ],
      "timeline": "",
      "notes": "",
      "metadata": {
        "created": "2026-11-15T09:00:00.000Z",
        "updated": "2026-11-15T09:00:00.000Z",
        "author": "[email protected]",
        "tags": []
      }
    }
  ],
  "experiments": [
    {
      "id": "EXP1",
      "type": "experiment",
      "name": "Two-week onboarding A/B",
      "description": "Question: does the checklist raise 7-day return? Method: random half of new trial users see it.",
      "hypothesesTested": [
        "H1"
      ],
      "status": "completed",
      "position": {
        "x": 160,
        "y": 320
      },
      "acceptanceCriteria": "Return rate in the checklist arm is at least 8 points higher than in the control arm.",
      "result": "Checklist arm 41% (82 of 200), control arm 29% (58 of 200): +12 points. Recorded as evidence E1 and counted there once.",
      "timeline": "2 weeks",
      "priority": "HIGH",
      "prior": 50,
      "confidence": 50,
      "completion": 100,
      "preregistered": "2026-11-01"
    }
  ],
  "evidence": [
    {
      "id": "E1",
      "type": "evidence",
      "title": "A/B result: 41% vs 29% return within 7 days",
      "description": "Result of EXP1. Threshold was +8 points; observed +12. Moderate sample (200 per arm).",
      "channel": "experimental",
      "strength": 1,
      "resultOf": "EXP1",
      "position": {
        "x": 60,
        "y": 120
      }
    }
  ],
  "outcomes": [],
  "edges": [
    {
      "id": "E1-H1",
      "source": "E1",
      "target": "H1",
      "data": {
        "type": "supports",
        "confidenceImpact": "×3.5"
      }
    },
    {
      "id": "EXP1-H1",
      "source": "EXP1",
      "target": "H1",
      "data": {
        "type": "tests",
        "confidenceImpact": "×1",
        "condition": "The result is recorded on E1 (resultOf), so this edge no longer counts."
      }
    },
    {
      "id": "EXP1-E1-result",
      "source": "EXP1",
      "target": "E1",
      "label": "Result",
      "data": {
        "type": "dependency",
        "confidenceImpact": "×1"
      }
    }
  ]
}

Download experiment-with-result.json

Import errors and what they mean

Every import error starts with the path of the field that failed, such as edges[2].data.type (the third edge, its data object, its type). Lists count from 0. Fix the field it names, then import again; the importer reports the first problem it finds.

Paste a BayesBoard JSON object.

(document)

Meaning: The text parsed as JSON but is a list, a string or null, not an object.

Fix: Wrap everything in one pair of curly braces, with version, hypotheses and the other lists inside.

Expected a BayesBoard JSON document (schema version 1.x).

version

Meaning: version is missing, is not text, or does not start with 1.

Fix: Set "version": "1.3" (text, in quotes).

Missing required array: <name>.

hypotheses, experiments, evidence, outcomes or edges

Meaning: One of the five lists is absent or is not a list.

Fix: Add it, as an empty list [] if you have nothing to put in it.

Missing the data object (edge "<id>").

edges[<n>].data

Meaning: An edge has source and target but no data object holding the kind and strength.

Fix: Add "data": {"type": "supports", "confidenceImpact": "×2"}.

Unknown edge type (edge "<id>"): got "<value>". Use one of ...

edges[<n>].data.type

Meaning: The kind is missing or spelled differently from the seven allowed ones.

Fix: Use supports, weakens, tests, falsifies, conditional-on, dependency or contradiction. Spelling and hyphens must match exactly.

Missing the strength string (edge "<id>"), for example "×2".

edges[<n>].data.confidenceImpact

Meaning: The strength is missing, empty, or a number instead of text.

Fix: Write it as text: "×2" or "2" in quotes. Use 1 or more; the kind sets the direction.

Invalid input at <path>: <reason>. (API only)

bundle.<field path>

Meaning: POST /api/boards/import checks the same document more strictly: ids must be non-empty text up to 128 characters, edge source and target are required, strength text is 1 to 50 characters. The error names the first failing field; the details list has all of them.

Fix: Fix the field named in the message, then send again. The path starts at the request body, so the board is under bundle.

Unsupported bundle version "<value>" (API only)

bundle.version

Meaning: The API accepts versions 1.0 to 1.x only.

Fix: Set "version": "1.3".

The numbers in a board are your own judgments made consistent, not predictions. See BayesBoard for agents for the API and the MCP server.