probability and calibration workbench
Superforecasting command center
Forecasts are useful only when the event is framed, probabilities are separated, reality can settle the claim, and calibration can score the result. This page keeps that whole chain visible before a forecast is treated as evidence.
KELLY lane
independent probability
User lane
private comparison
Truth lane
settlement before learning
forecast contract
What must be visible?
forecast audit path
Probability is not proof until reality can score it.
The workbench forces the practical Superforecasting sequence: define the event, split KELLY and user probabilities, name the settlement authority, then score the closed outcome through calibration.
Forecast question
A real forecast starts with a future event, close date, and measurable outcome.
Two probability lanes
KELLY probability and user probability stay separate so scoring stays honest.
Settlement authority
Every scoreable forecast needs the source or condition that will resolve it later.
Calibration after close
Reality closes first; Brier-style scoring and memory eligibility come after settlement.
better forecast shape
Event + close date + probability + resolver.
Example: "Will this event happen by July 31, what is KELLY's probability, and which source settles it?"
no-public-write boundary
A public user forecast can guide comparison and attention, but it cannot become KELLY truth. KELLY only learns after approved reality, settlement, calibration, replay, and governed memory paths agree.
How KELLY forecasts with usersseparate lanes
A user's forecast can point KELLY toward a useful reality loop, but it never becomes KELLY's forecast by itself. The value is in comparing both lanes, settling against reality, and calibrating over time.
User forecast
The user can enter a probability. It is saved as the user's lane, not KELLY's truth.
KELLY forecast
KELLY forms its own probability from approved data, memory, source mode, and uncertainty.
Reality settlement
The event later resolves against reality, producing a settlement and calibration record.
Proof boundary
Receipts and replay show what was known, what was inferred, and what remained uncertain.
Live proof mapchecking
Follow KELLY's current governed learning lane from source to observation, receipt, settlement, calibration, replay, and governed memory. The headline metrics come from the live runtime; archived bootstrap lineage is kept only for traceability.
Sources
0
Observations
0
Hypotheses
0
Closures
0
Visible cells
0
Replayable
0
Total Neon cells
0
Visible replay receipts
0
Next candidates
0
Velocity
unknown
Source: unknown. Blocker: none. Public users cannot directly write ground truth; archived ledger backlog is monitored separately from current cell replay receipts.
none
Loop state: waiting
Waiting for public-safe proof-map evidence.
Last checked: not yet. Public user writes: blocked.
WOW feature
Dual-lane forecast scoreboard
Not another chat box: this lens has its own instrument, forecast posture, and artifact trail.
Users want
Calibration honesty: user probability and KELLY probability scored separately when reality closes.
Proof that matters
Every important output should open backward to source and forward to replay, settlement, calibration, or memory lineage.
Open proof targetprobability workbench
Forecast lane discipline
Can this question be scored by a named future outcome?
primary operator read
event + resolver
A forecast is useful only after the event, close date, probability lane, and settlement source are explicit.
Event frame
bounded
Outcome and horizon named before probability.
KELLY lane
separate
Independent system probability stays separate from user probability.
Settlement
required
No calibration without an approved resolving source.
Replay
after close
Closed outcomes flow to calibration and memory eligibility.
probability rail
forecast anatomy
Probability contract
The page keeps the scoring sequence visible before a forecast can look authoritative.
Frame
Question is converted into a measurable claim.
Forecast
Two lanes, never blended.
Settle
Reality source closes the loop.
calibration posture
Scoring queue
The workbench shows what can be scored now and what is still waiting.
Open
Unsettled forecasts stay out of memory.
Closed
Brier and reliability only after settlement.
Stale
Missing close data becomes a source task.
Sector loop
Frame
Ask the forecastable question.
Precommit
Record KELLY and user lanes separately.
Resolve
Wait for the named authority.
Score
Update skill after reality closes.
Conversation flow
Ask
You pose a bounded question in plain language.
Source plan
KELLY chooses what reality to consult before answering.
Evidence
Live and historical artifacts are compared under uncertainty.
Answer
A bounded reply separates known from uncertain.
Watch
Open questions get a private watch lane until reality closes.
High-value forecast
Probability with closure
Turn an uncertain future event into a forecast with a closing condition and settlement date.
Use presetAsk KELLY here
In-lab conversation
Ask for the event, base rate, settlement date, and the strongest update trigger.
Ask in this labProof artifact
Forecast split
KELLY probability, user probability, settlement condition, and calibration trail.
Open proofDedicated app
Open full lens
Use the dedicated sector app when you want more room for charts, timelines, proof cards, and live updates.
Open Superforecasting appLens artifact passport
Every serious output must trace to an artifact. The answer stays simple on the surface, the heartbeat visible in the top bar proves the engine clock is present, and these links open the source, health, guide, user controls, or Glass Lab proof when the user wants depth.
Dedicated app
Open Superforecasting app opens the standalone Superforecasting surface when a sector needs more room.
OpenLive forecast API
Machine-readable sector output for forecast cards, proof widgets, or status monitors.
Not exposed yetArtifact spine
Eight-step proof chain for this lens: source, observation, hypothesis, proof, replay, settlement, calibration, memory.
Not exposed yetHealth check
Fast check that the standalone lab is reachable and serving the current runtime.
Not exposed yetUser controls
Export, delete, location, privacy, and feedback controls stay visible from every lens.
OpenAsk KELLY inside this lab
The question stays in this sector lens. KELLY answers here with a bounded answer, visible uncertainty, proof path, and the next reality check when more evidence is needed. No page jump, no generic redirect.
KELLY response
This panel shows the answer inside the lab. It does not create public ground truth, memory, or GraphState writes. The proof link is the handoff when you want the moving parts under the magnifying glass.
How KELLY should answer in this lens
Use plain decision language. Name the target, time horizon, stakes, uncertainty, and what evidence would change the answer. KELLY should respond with a bounded artifact, not an engine dump.
Source families
Live signals
Proof windows
User workflow
Define a future event with a close date and measurable outcome.
Place your probability beside KELLY's independent read.
Inspect what evidence would move the forecast.
KELLY work behind the lens
Builds scenario lanes and keeps probability separate from confidence.
Tracks settlement so the forecast can be graded later.
Feeds closed outcomes into calibration instead of treating text as truth.
Artifact trace rail
Every lens output should be traceable backward to source and proof, then forward to replay, settlement, calibration, memory eligibility, or quarantine.
Answer
visible output
Source
reality path
Proof
why it said it
Replay
rebuild path
Score
reality grades it
Lab contract
mode
read-only lens
authority
KELLY Prime
proof target
#calibration
Public boundary
A user forecast can become a search/probe request. It is not treated as reality.
Why this is different
KELLY does not just answer from text. Each lens keeps the answer connected to source posture, proof, replay, settlement, calibration, and memory eligibility so the user can see how the engine knows, what it is watching, and what would change its mind.
next scoring checkpoint
Closed forecasts should flow to settlement, calibration, replay, and memory eligibility before any durable learning claim appears in Glass Lab.