VENTURE · ADAPTIVE TRAVEL DECISION SYSTEM
ReturnVector
KEEP GOOD MOVES ALIVE.
An itinerary tells you what you planned. ReturnVector tells you what still makes sense now.
Travel changes after the plan is made. Seats disappear. Prices move. Weather changes. Transit closes. Connections weaken. The traveler gets tired. ReturnVector continually recomputes whether to go, wait, commit, pivot, or return from the reality that exists now—not the itinerary that existed this morning.
The deeper question is not only “What is the best move now?” It is “Which move preserves enough good options that I am not trapped if the next thing goes wrong?”
Evidence state: extensively operator-tested as a working decision system; predictive scoring remains Stage 0 until repeated prospective calibration.
UNDER THE HOOD
Constrained inventory · disruption risk · trip utility · separate confidence · time-dependent recovery paths · correlated-failure discounting · decision timers · traveler state · frozen predictions · later calibration.
Its HomeGraph models whether a backup route actually remains feasible through time; two apparent alternatives count less when they share the same fragile hub, carrier, weather system, final segment, or other failure point.
THE PROBLEM
An itinerary is a snapshot. Travel is a moving system.
A conventional itinerary represents a sequence of intended events: departure, arrival, reservation, attraction, return. Reality is less obedient.
Seats disappear. Weather moves. Prices change. Transit stops running. A cheap route becomes expensive after ancillary cost. The body becomes tired. An attraction that made sense six hours ago stops being worth its price. A final flight home becomes too fragile to support one more stop.
A decision can become wrong without having been wrong when it was made.
ReturnVector recomputes from present reality instead of defending the original plan.
THE DECISION PROBLEM
What should I do now—and what will remain possible afterward?
ReturnVector evaluates feasible actions against four modeled dimensions and one separate epistemic question.
SEAT AVAILABILITY LIKELIHOOD
How strong is the current evidence that constrained inventory will be obtainable at the relevant decision window?
DISRUPTION RISK
How exposed is this plan to operational degradation?
RETURN INTEGRITY
If the intended path fails, how much credible recoverability remains?
TRIP UTILITY
Given cost, time, experience value, route geometry and body state, is this trip actually worth doing for this traveler?
CONFIDENCE
How much evidence supports the estimate? Confidence remains separate from the score. A high estimate supported by weak evidence is still weak evidence.
A score is not a fact. Confidence is not probability. Precision is not calibration.
HOMEGRAPH
A backup is not a backup until it still exists.
ReturnVector represents recovery through a time-dependent graph of flights, airports, rail, coach, ground transport and other feasible movement capable of satisfying the traveler’s declared return constraint.
An alternative contributes to recoverability only while it remains sufficiently verified, chronologically feasible and physically plausible. Two written itineraries are not necessarily two independent backups. Routes that share a critical hub, final segment, aircraft or crew rotation, carrier failure domain, weather system or other common vulnerability are correlated and must be discounted accordingly.
Raw route count is not recovery depth.
The governing question is simpler: If the primary plan breaks now, what can I still actually do?
DECISION TIMERS
Waiting has an expiration time.
Travel decisions are often framed as choices between actions. ReturnVector also models the cost of not acting.
A better option may remain available for thirty minutes, two hours or one day. After that, a flight departs, inventory closes, a transfer ceases to be possible, weather worsens, daylight disappears or the traveler no longer has sufficient capacity. A decision timer marks the point at which continued waiting becomes less attractive than changing course.
It is not a reminder. It is a boundary in the decision state.
THE TRAVELER IS PART OF THE SYSTEM
A technically feasible trip can become humanly infeasible.
Routing systems usually treat the traveler as cargo. ReturnVector does not.
Sleep, heat, fatigue, pain, mobility, hunger, companion state, tolerance for friction, experience preferences and the value of returning cleanly can alter the utility of an otherwise identical itinerary. The system therefore distinguishes movement that remains technically possible from movement that remains rational for the person who must execute it.
Personalization belongs in the traveler model, not in universal product assumptions.
LEARNING
The system keeps the prediction it made before reality answered.
ReturnVector separates prediction from outcome.
01 · TARGET
Declare what is being predicted before the outcome exists.
02 · FREEZE
Preserve the prediction and model version before reality answers.
03 · OBSERVE
Record the outcome independently rather than repairing the forecast.
04 · EVALUATE
Measure error, calibration and subgroup behavior against simpler baselines.
Material misses remain part of the record. Changes to features, targets or weights require a new model version rather than retrospective repair.
The objective is not to produce numbers that look scientific. It is to make the system increasingly difficult to fool—including by itself.
PRODUCT STATUS
The system works in repeated operator use. Its predictive models remain Stage 0 until prospective calibration.
ReturnVector has been exercised extensively across repeated operator travel decisions and replanning. That operating use supports the utility of the decision architecture; it does not make its current heuristic scores empirically calibrated probabilities.
Its targets, prospective observation protocol, evaluation governance, prediction/outcome separation, HomeGraph evidence requirements, and model-version discipline are defined. The remaining evidence question is prospective calibration of the predictive layer, not whether the decision system can be used.
Current calibration gate · one reproducible state-change evaluation
- Freeze the original prediction. Preserve the model version, evidence timestamp, current trip state, recommendation, and the estimate made before reality changes.
- Name the recheck trigger. State which change in inventory, price, weather, time, infrastructure, or traveler state should force recomputation.
- Observe the changed state. Record the new evidence independently rather than rewriting the earlier snapshot.
- Recompute the decision. Determine whether the recommendation appropriately stays, pivots, waits, commits, or returns under the changed state.
- Preserve recoverability. Record the viable HomeGraph route or the evidence that recovery depth has materially deteriorated.
- Record the outcome separately. Do not repair the original prediction after seeing what happened.
- Evaluate later. Compare the prediction, adaptive response, and outcome against error measures and simpler baselines before changing model rules.
Claim boundary. Extensive operator use establishes functional use of ReturnVector as a decision system. Passing the prospective gate would demonstrate one auditable adaptive-decision path, not calibrated predictive accuracy. Calibration requires repeated prospective cases and independent outcome review.
System maturity can be high while model calibration remains early.
PRODUCT BOUNDARY
Frontier is a proving ground, not the product boundary.
ReturnVector began in unusually constrained travel conditions because constrained travel makes the decision problem visible: uncertain inventory, narrow booking regimes, policy overlays, ancillary economics, fragile recovery paths and changing traveler state.
The architecture separates carrier and fare rules, traveler configuration, benefits, current evidence, HomeGraph, model outputs and optimization. The architectural aim is that a new carrier, fare regime or transportation mode can be introduced without redefining the core decision problem.
RESEARCH · PUBLISHED
The Geometry of Getting Home
A multidisciplinary inquiry into recoverability, agency, temporal networks, decision theory, embodied rationality, infrastructure, resilience and the intelligence of remaining able to choose.
The best plan is not the one that predicts the future correctly. It is the one that leaves you with good moves when the future refuses to cooperate.