Retrieval physics
Every note is one complex number, Γ, seen from the query at the center. Re-ask the query 16 ways and each note traces a locus — and that single trace carries the whole Smith model at once: how well it matches, what kind of mismatch, how robust it is to phrasing, and whether it resonates. Synthetic demo set.
Tick 2+ notes in the table below, then find the single shared context that pulls the whole cluster toward center at once.
| ∈ | Note | Fit |Γ| | VSWR | Robustness | Sensitivity (Q) | Reads as |
|---|
Angle answers what kind of mismatch, radius answers how much — the two things a single cosine number can't separate.
Switch the query to watch every note move: relevance is measured relative to Z₀, not absolutely. Select a note and hit Tune to match to walk its L-network path to the center — the Smith chart's original job, now reading as a recipe of memory moves. Or tick several notes and Find common match: the one shared context (a broadband matching network) that pulls a whole topic inward at once, with a no-free-lunch audit. Switch Sweep axis to Aging to scrub an as-of date instead: notes drift toward the rim as they age, each at its own claim-aware rate (a concept barely moves; a market note rots fast), and Revalidate snaps them back in. The exact bridge underneath: log(VSWR) = 2·artanh|Γ| = hyperbolic distance from center. Age ≠ truth — a stale note can still be right. Select a note past the rim (contradictory) to Reconcile it: an evidence branch that ends honestly as REVISED, DISPUTED, or UNRESOLVED — with a stability margin that flags echo chambers, never quietly tuning a contradiction to the center.
Synthetic route data — not factual search results. ↑ evidence/inference break · ↓ rejected or repeated work · height = severity · x = accumulated cost.
Hover or click a spike (or a chip) to inspect the fault and its fix.
A reflectogram debugs where answer-building breaks, not which document ranks — the unit is the answer commitment, not the note. Each spike is a synthetic discontinuity; the path return loss = −20·log₁₀(RMS Γ), so higher dB = a cleaner route. Send the impulse to watch the query travel the route.