Original assumption
The forecast relied on a particular event occurring before the deadline.
PREDICTION MARKETSNew evidence.
New evidence.
Lasting reasoning.
Keep sources, forecasts and outcomes together. Let research agents revisit earlier assumptions as new evidence arrives.
Observe→Recall→Record
MEMORY IN PRACTICE
New evidence should build on earlier reasoning.
Forecasting is a sequence of updates. Preserve the sources, assumptions and revisions behind each forecast so a research agent can explain what changed between sessions.
EXAMPLE WORKFLOWPrediction markets
USER REQUEST
Start with the question.
New evidence has arrived. What should we revisit in this forecast?
Start with the question.WHAT THE WORKFLOW CAN REMEMBER01 02 03
Source record
Keep the publication date and reference behind that assumption.
Previous revision
Record which evidence changed the last assessment.
Example inputs and responses for this workflow.
THE PROBLEMWhen the context
When the context
doesn’t carry forward.
Forecasts become isolated answers
A probability without its assumptions is hard to revisit when new evidence appears.
Sources get separated from claims
Research notes spread across sessions, making the origin and age of an observation difficult to follow.
Resolved outcomes go unused
Without a recorded history, teams cannot easily compare what they expected with what actually happened.
WHAT MEMORY MAKES POSSIBLEKeep what matters.
Keep what matters.
Build on what came before.
01EVIDENCE RECORD
Evidence with its context
Keep source references, publication dates and research notes together. Recall the evidence behind a forecast instead of relying on an unsupported summary.
Source reference
Observed at
Research notes
02ASSUMPTION MEMORY
Assumptions worth revisiting
Write the conditions that support a forecast as explicit records. A later session can retrieve them when a related event or observation arrives.
Original premise
New evidence
Revised reasoning
03FORECAST HISTORY
A history of revisions
Persist each assessment with its timestamp and rationale. Your application can compare revisions while retaining a reference to earlier content.
Initial assessment
Revision
Resolution
04OUTCOME REVIEW
Review after resolution
Link the outcome to earlier assumptions and evidence. Use your own scoring and review process to decide what the team should carry into future research.
Forecast
Actual outcome
Review notes
THE WORKFLOWFrom a new observation
From a new observation
to lasting context.
Write the evidence
Record source references, timestamps, market tags and the assumptions they inform.
Recall before revising
Retrieve related evidence and previous assessments when new information arrives.
Close the review loop
Record the resolved outcome and review it against the research history.
Lighthouse preserves the research record. Source accuracy, forecast scoring and market actions remain part of your application.
DESIGN YOUR INTEGRATIONA memory layer.
A memory layer.
Your application’s logic.
Connect the memory runtime to the moments your application needs to write, recall and review context.
Explore memory documentation ↗01
Choose an event or market identifier that stays consistent across sessions.
02
Store observations separately from the agent’s interpretation.
03
Timestamp each forecast and record the resolution criteria.
04
Implement forecast scoring and market actions in your own application.
COMMON QUESTIONS
Before you build.
Does Lighthouse predict market outcomes?
No. It provides a way to persist and recall the research context your forecasting application uses. The model and application produce the assessment.
Can it check whether a source is true?
A stored reference helps trace a claim back to its source. Evaluating source reliability and factual accuracy remains part of the research workflow.
How do we organise related markets?
Your application can use namespaces and tags to organise records by event, market or topic, then query the relevant context.
Can we review earlier forecasts?
Persist the forecast, timestamp, rationale and source references. Your application can retrieve these records alongside the resolved outcome for comparison.