Failure Patterns / What Broke

EchoNet uses failure in the lab sense: an observed break, trap, false success signal, measurement limit, or boundary leak that changed later work.

The records include method work, forensic passes, diagnostics, runtime experiments, game-lab work, and publication failures. Public pages only carry the pattern level. Private mechanisms and legal-risk details stay out.

Object Drift

The most repeated trap is object drift: the answer starts with one object and quietly acts on another.

Observed forms include a later recounting replacing the original event, a digest replacing the source body, a stable label hiding a changed referent, and a remembered project state replacing current evidence.

Success signal: name the object first, then name the source surface. If the object changed, preserve that fact instead of smoothing it away.

Source And Evidence Traps

Stored material is not authority by existence. Agreement is not corroboration. A cross-reference lead is not a confirmed fact. A local endpoint or private path is not public evidence.

Success signal: claims carry evidence type, source boundary, and proof scope. Tables come before theory. Unknowns stay visible.

Durable Drift

Cheap failures are caught before they reach durable files. Expensive failures reach a handoff, card, route, repo, or public page before the guard fires.

The dangerous shape is often not nonsense. It is a useful compressed truth that got one artifact layer ahead of its source.

Success signal: preserve source bodies, not only summaries. A guard counts only when it fires before the artifact becomes load-bearing.

Lab And Runtime Traps

The lab record contains recurring boundary failures: historical success becoming authority, renderability mistaken for existence, soft prompts mistaken for hard gates, diagnostic output leaking into user-facing text, parameter controls that do not reach the payload, and happy-path tests treated as stress tests.

Success signal: a test names the property it pays for. A local pass does not prove the neighboring architecture. A typed non-execution result is valid when the required boundary is absent.

Publication Boundary

The prior public surface also failed. It drifted toward article-shaped material that did not derive from project records at the record level required here.

The correction is not silence. The correction is publication from the record layer: failure class, trap indicated, success signal, source boundary, and unknowns.

Open Edges

Some old material may contain useful observations. It does not return automatically.

Each candidate needs a source check, a public-risk check, and a reason to be public. Anything not established stays open rather than resolved by narrative.