V0 — Evidence that remembers its reasons

Experimental evidence · V0Historical result reported by CONTROL; the displayed original console excerpt is withheld pending source-byte reconciliation. Exact execution timestamp not verified.
View qualified historical result ↓

From familiar examples to measured observations

First read the example. Next, use Next step → to move through the diagram. The colored boxes show which items still have support; arrows show which item relies on another. Finally, compare the explanation with the selected original console excerpt. The example teaches the idea; the console excerpt reports what was observed.

Why mistakes like this can matter →

RECORDED EXPERIMENT · V0

Evidence that remembers its reasons

Recorded experiment

First, an everyday example

A hospital needs dialysis supplies before morning treatments

At 6 a.m., a hospital needs dialysis filters delivered before scheduled treatments. A supplier has confirmed two possible transport routes: a truck on the highway and a freight train.

Read the complete real-world example

At 7 a.m., a closure makes the highway unusable. The train remains available, so the shipment might still arrive. When rail service is also cancelled, the original answer—“We have a delivery route”—can no longer be supported. A stale answer could lead staff to plan treatments around supplies that may not arrive.

How the example maps to experimental labels

In this picture: E17 = confirmed truck availability; C17 = truck delivery could work; E3 = confirmed rail availability; ALT = rail delivery could work (alternative path); TOP = at least one delivery route is available (overall conclusion). These are teaching labels; the real experiment used synthetic relationships, not actual transport records.

Where the analogy ends and the experiment begins

Now move from the logistics story to the recorded test.

In the hospital example, the truck and train are independent ways to obtain a shipment. In the recorded experiment, two explicitly represented supporting paths affected an overall conclusion. Watch the first path fail while the alternative survives, then watch the remaining support disappear.

Real test: specifically represented synthetic evidence and conclusions, not actual supply routes. Select “Next step” to see the states reported in the selected original console output.

▶ Interactive replay of the recorded observations
Guided graphical replay · v0

Follow the evidence through two corrections. The diagram shows which conclusions still have support at each step. This is a teaching illustration based on previously reported test results, not a new execution or a diagram of internal implementation.

Arrows show relationships, not good or bad status. Follow an arrow to see what relies on what; follow the box color to see its current state.

Green = supported / validRed = withdrawn / blockedYellow = does not apply here

Evidence boundary: The colored diagram is an explanatory reconstruction, not a screenshot or a live run. The recorded outcome is summarized below; unverified verbatim console lines are withheld pending source reconciliation.

Define the terms used in this demonstration

In plain language

When a supporting fact changes, a conclusion may need to change too. This test asks whether unaffected conclusions can survive rather than being erased along with everything else.

Why it matters

Accumulated conclusions are useful only if the system can revisit them when their evidence changes.

For engineers

The bounded dependency representation supported selective invalidation and preservation of alternative justification.

Tested question

Withdrawing upstream evidence from an explicitly linked synthetic scenario.

View scientific record and qualified result summary

Actual execution date/time: NOT VERIFIED — exact historical timestamp unavailable
First website publication: NOT PUBLISHED — update only when the page is publicly released
Test conditions: A small synthetic scenario with explicit supporting relationships. Initially ALT and TOP were valid; MISSING was unknown. Evidence E17 was invalidated, then E3.
Observed outcome: ALT and TOP remained VALID after the E17 correction and became INVALIDATED after E3 was invalidated.
Known limitations: Small synthetic dataset; explicit dependencies; no competing-system benchmark. Deterministic replay PASS was reported by the experimental runner.
Historical release-note date: NOT VERIFIED
Release note: Initial experimental foundation for evidence-sensitive correction.

Historical result summary — not verbatim console outputSelected original lines withheld pending byte-for-byte reconciliation with protected source evidence.

This public-draft summary is based on the recorded bounded experiment and does not represent a fresh run or a certified original transcript.