Skip to OverLift overview Skip to interactive globe Skip to current OverLift workspace

OverLift is an interactive, private and security-bounded synthetic logistics proof. Use the overview, the complete twelve-step OverLift Method, Failure Lab, guided proof, simulation, or engineering workspace. Browser-local semantics and minimum disclosure reduce data exposure; workers, WebAssembly, and explicit capabilities reduce execution power; AI proposes options while deterministic authority decides what may change. The Failure Lab demonstrates insufficient evidence, denied permission, duplicate suppression, exact tool approval, replay, audit, and evaluation gates.

OverLift
OverLift turns governed evidence into exact operational decisions. Beyond the model. Let AI explore. Keep truth exact.
Online
OverLift overview active.
The globe is starting slow automatic rotation.
OverLift shell ready. Heavy evidence and simulation packs load only after you choose an action.
The OverLift Method · governed agentic AI

Let AI explore.
Keep truth exact.

OverLift is Sean Findley’s browser-native method for turning ambiguous operational questions into grounded, governed, replayable decisions—while keeping sensitive evidence inside the chosen application boundary and giving agents explicit capabilities instead of ambient host authority.

  • Explore broadlyCompare bounded paths and trade-offs.
  • Ground every claimUse current, authorized evidence only.
  • Keep authority exactAI proposes. Deterministic rules decide.
  • Protect private contextLocal-first semantics and minimum disclosure by design.
CHOOSE YOUR LENS

Privacy by architecture Browser-local MiniLM and vector evidence can stay inside the application; provider use is optional and minimum-disclosure.

Security by bounded power Workers and WebAssembly contain risky computation; a capability broker exposes only approved data, tools, budgets, and deadlines.

Security architecture
Browser-local MiniLM · local vector evidence · minimum disclosure · deterministic Truth Core · replayable receipts

The globe begins rotating slowly. Drag to take control. Use the mouse wheel or plus and minus keys to zoom, arrow keys to rotate, A to pause or resume rotation, R to reset, and E to switch between Earth and network presentation.

Selected route GX-2047 · Dubai recovery Recommended global recovery · risk 37
OriginSingapore DisruptionSuez closed DestinationUtrecht
Layers
Physical Earth: day, night lights, water, relief, atmosphere. Atmospheric raster excluded by design.
Starting the live WGS84 globe nowSource-backed preview · live runtime is starting immediately
deferred
Starting globe
L0 preview1200×900 immediate preview Immediate first paint
Regional clipmap idleZoom closer for regional tiles 0.0 / 48 MiB cacheB2F3 source pack not installed
— FPSPresented — msFrame p50 — msFrame p95 — msFrame p99 —Geo vertices Global viewView center
NASA Blue Marble · WGS84 C17/Wasm placement · synthetic route corridors
The OverLift Method Four authority boundaries around a twelve-step operating cycle.

These four words define who may do what. The actual operating method continues through twelve explicit steps—from framing the question and admitting evidence to exact execution, verification, and replay.

  1. ExploreFrame the decision and generate bounded options without changing official state.
  2. GroundAdmit current, visible evidence and preserve uncertainty instead of guessing.
  3. GovernApply identity, policy, budget, approval, and idempotency limits to every path.
  4. ProveApprove one exact proposal, execute, verify, seal the receipt, and replay it.
Built by Sean FindleyProduct thinking and deep engineering in one system.

OverLift demonstrates senior-level ownership across operational AI, browser-native systems, deterministic authority, and product communication.

  • Product designTurns complex architecture into a usable decision journey.
  • Agentic systemsTyped planning, retrieval, tools, and bounded proposals.
  • Evidence integrityFreshness, permissions, provenance, and explicit unknowns.
  • Performance and proofWasm, GPU rendering, replay, receipts, and validation.
What this demo provesEvidence-backed. Not hype.
Implementedsemantic compilation · governed evidence · deterministic state · replay · receipts
Demonstrated withsealed synthetic logistics data
Active researchOverLift Pathfinder
Not claimedquantum hardware or certainty about future events
Cross-device parityqualification pending
Failure behavior · six deterministic synthetic tests A trustworthy agent must prove how it stops.

Run a synthetic failure path. OverLift exposes the gate that stopped the request, the action that never ran, the canonical state that stayed protected, and the receipt or evaluation evidence left behind.

0 of 6 tests run
Test 1 of 6 · Evidence boundary What happens when the evidence is insufficient?
Ready to test
Test request

“Guarantee that OD-2566 arrives on time even though the current carrier-capacity record is missing.”

  1. Identity and scopeWaiting
  2. Evidence admissionWaiting
  3. Support thresholdWaiting
  4. Tool dispatchWaiting
  5. Canonical stateWaiting
OverLift response Abstain and request the missing evidence.

The system may return supported facts and name the missing capacity record, but it cannot manufacture a guarantee, rank a route as safe, or dispatch a tool.

Decision
Pending test
Tool call
Not evaluated
Canonical state
Protected
Quality signal
Abstention correctness
Proof left behind Failure receipt, explicit unknown, and unchanged-state witness

The receipt records which evidence was admitted, which required fact was absent, why no proposal was eligible, and that no tool invocation occurred.

OVL-FAIL-EVID-PENDING

Failure Lab ready.

Retrieval correctness Recall@k · nDCG@k · evidence-admission precision/recall · citation precision Also checks stale or unauthorized selection, supported-claim rate, and correct abstention when evidence is missing.
Execution correctness Typed-schema validity · policy decision · tool-call exactness · idempotency Postconditions, exact state transition, receipt integrity, and replay-hash parity must all agree.
Current public claim boundary Sealed evaluation design; blind retrieval benchmark not yet claimed. The shipped Quality Lab publishes the corpus, split, holdout, and receipt boundaries without inventing a score that has not been run.

OverLift shows what it understood, which evidence it used, and where authority stops before anything consequential can happen.

Try asking
Not sure where to start? Choose a starter or type your own question.
Loads when the proof starts C17/Wasm authority starts with governed execution Sealed world ready on demand Browser agents ready on demand · deterministic fallback · local MiniLM on supported queries
plan, retrieve, ground, and govern locally. still owns any consequential state.
Scene 1 of 7 · Human question Ask one question. Watch the machine expose every step.
Story mode Meaning Vault initializing
What happens if the Amsterdam facility goes offline?

Press Run question to watch the words become typed meaning, evidence, uncertainty, governed possibilities, exact execution, and proof.

Operationwaiting Entitywaiting Measureswaiting Moderead-only interpretation
Verified explanation

No result yet. The screen will not fill missing facts with plausible prose.

Questionwaiting Evidencewaiting Policywaiting
: waiting : waiting Search receipt: pending
Ask a question, then step through the real calculations.
Keep exploring
Method map
Retrieval Evidence Meaning Density Uncertainty Governance Commit Proof
0 of 8 concepts explored
Architecture & proofSee evidence become exact, governed state.
Evidence → meaning → authority → execution → receipt
Watch the OverLift process

See a question become a verified outcome.

Initializing deterministic visual state
Stage 1 of 5

Evidence admitted

Six scoped records enter the evidence packet; two unrelated records are rejected by tenant and role filters.

6 admitted 2 rejected 0 unknown owners