Evidence machine

A research superorganism for evidence.

Amesemi is built for questions that should not evaporate after one answer. It recalls before searching, searches before claiming, argues against itself before closing, and preserves the receipts that let future work build on what came before.

recallMemory before search
forageSources before confidence
challengeObjections before closure
proveReceipts after artifacts

How it thinks in public

The evidence loop is the product.

The point is not a longer answer. The point is a working loop where every claim has a memory status, a source posture, a rival explanation, and a closure condition.

stage 01

Ask: the question becomes a bounded run.

Amesemi starts by naming the work: memory surface, source access, cost posture, authority, and what will count as done. Short questions stay short; serious questions get a thread.

inputquestion + desired closure
riskscope drift
receiptrun brief

Receipts over performance

Every serious answer should leave a packet behind.

This is a representative anatomy of an Amesemi evidence packet. It is not live inference on this page; it is the shape of the thing the platform should produce when a run closes.

A primary source moved.

A living record should not rely on memory alone. Amesemi marks the claim as time-sensitive, re-retrieves the primary source where possible, records retrieval gaps, and preserves the superseded state as evidence.

evidenceprimary source preferred
uncertaintyretrieval date matters
authorityresearch only
closureliving record updated

        

Why superorganism-like

It gets stronger because work is not discarded.

The language is ambitious, but the claim is practical: many small, bounded runs share memory, source ledgers, approval packets, and corrections. The whole becomes more capable than isolated conversations.

memory strata

Many layers of memory

Runs write different kinds of memory: facts, procedures, decisions, anomalies, counterfactuals, and identity commitments.

shared inquiry

Work feeds work

A short answer can become a standing brief; a standing brief can become a product; a product can create new questions.

source ecology

Sources stay attached

Claims remember how they were known, when they were checked, what failed, and what would change the conclusion.

approval gates

Power has thresholds

Local research can continue freely; money, legal, reputation, publishing, and outreach wait for explicit approval.

non-erasure

Corrections compound

Wrong or stale work is not hidden. It becomes the history that explains why the newer state is trusted.

completion proof

Done means proved

An answer is not complete because it sounds complete. It closes when the artifact and its receipt can be inspected.

Honest ambition

Powerful autonomy, not theatrical certainty.

What Amesemi should be allowed to do

  • Continue research across sessions without forgetting why the run mattered.
  • Use read-only web sources and local workspace evidence by default.
  • Patch and improve local artifacts when the run is scoped for building.
  • Prepare approval packets for consequential external actions instead of silently acting.
  • Move optional improvements to backlog so completion can happen without losing the future.
completion packetevidence required
{
  "run_kind": "research_thread",
  "claim_policy": "sources before confidence",
  "memory_policy": "supersede_without_erasing",
  "authority_policy": "approval_required_for_external_action",
  "closure": "artifact_exists_and_receipts_attached",
  "motto": "the memories keep themselves"
}
stewardship system · living memory · the memories keep themselves