AutoResearch/Innovation Generator/V6/Noether

VERSION / V6.6

Noether

Noether

Treats reports as evidence indexes: true audits, answer-free sealed challenges, adjudicator-minted settlements, stable hash identities and prosecutor replay must agree, with demonstrations isolated as PASS_DEMO.

Historical version · Design & implementation

Current: Daedalus v7.4.0

V6.6 / ARCHITECTURE

How does an exploration move forward?

01稳定候选Stable candidate02密封挑战Sealed challenge03真实门控True gates04事先预测与结算Prediction thensettlement05独立复算Independentrecomputation06分层判定Tiered verdict01稳定候选Stable candidate02密封挑战Sealed challenge03真实门控True gates04事先预测与结算Prediction then settlement05独立复算Independent recomputation06分层判定Tiered verdict
Architecture illustration derived from this version’s actual control flow.
  1. 01

    Stable candidate

    Identity and demo flag

  2. 02

    Sealed challenge

    Public inputs omit answers

  3. 03

    True gates

    Invoke V6.2 labs and gates

  4. 04

    Prediction then settlement

    Controlled claim-credit minting

  5. 05

    Independent recomputation

    Challenge, audit, settlement, hashes

  6. 06

    Tiered verdict

    Demo / conservative / eligible discovery

01

Does evidence survive independent recomputation?

Noether addresses Feynman's wiring and credit gaps: a report can name gates it never ran, and demonstrations can receive discovery credit. V6.6 requires an independent prosecutor to recompute challenges, true hard audits, settlements and index hashes. Recomputational conservation is a design metaphor for conclusions surviving rechecking, not experimental proof of a physical theorem.

02

True gates, sealed judging and zero demo credit

Candidate reports no longer control final confirmation. Public challenges contain tasks and inputs, not expected answers or their digests; judging is recomputed separately. Claim credit is minted only through the adjudicator and binds candidate, task, replay hash and a prediction made before execution. Demo and synthetic quality multipliers are zero, so interface success cannot become discovery.

03

Recomputation and tiered verdicts

The flow establishes stable identities, executes sealed challenges, invokes real V6.2 labs and gates for executable parts, registers predictions and settles them. Reports retain indexes and digests. The prosecutor recomputes challenges, audits, settlements and hashes and requires detection of known-bad fixtures. Only then does it distinguish demo, conservative and eligible discovery tiers.

04

Public challenges and controlled credit events

PublicChallenge carries challenge ID, world ID, task reference, round, query, input and judge-process reference while forbidding expected-answer fields. ClaimWEEvent is created through controlled minting, and settlement verifies candidate binding and prediction timing. ProsecutorReport lists discrepancies separately, making tampering and failed computation inspectable.

05

Conservative runs and explicit demonstrations

Inputs are the topic, mode, pool, external evidence bundles and an explicit demo_fixture flag. Default metadata-only candidates do not request claim credit; the sorting program appears only with the fixture. Outputs include challenge certificates, real gate reports, prediction and settlement ledgers, external evidence structures and independent recomputation, preserving the distinction between demonstration and discovery.

06

Conceptual walkthrough: after a demo passes

An explicit sorting demonstration may pass challenges and real algorithm gates, yet demo_fixture in the manifest keeps the verdict at PASS_DEMO and claim-layer bits at zero. Late predictions are rejected, and a forged V6.5-style imitation should fail negative checks. This illustrates resistance to invalid promotion, not the creation of a new sorting algorithm.

07

Repair Feynman's seams

Relative to V6.5, Noether replaces the three-gate imitation with a real V6.2 adapter, removes answer fields from public challenges and adds the prosecutor, controlled settlement and zero fixture credit. External bundles distinguish full recomputation, sampled checking and structural inspection. Poincaré extends these checks into event causality and multi-generation lineage.

08

Which errors the tests target

Code and test definitions compare real gates with the old imitation, reject direct construction of claim events and late predictions, detect a four-step bad fixture and keep demo credit at zero under PASS_DEMO. A content-free default is conservative. This account inspects those implementations without rerunning the archive; the negative cases specify software acceptance, not open-science effect comparisons.

09

Independent recomputation has trust boundaries

Independent recomputation improves report integrity, but the internal world, checker, task definitions and operator remain trust boundaries. Example algorithms and fixed challenges exercise interfaces, not general scientific discovery. Structural evidence-bundle checks must be distinguished from full re-execution, and any PASS_DISCOVERY label must be interpreted against a specific claim, task and verifier.

Reading example · Explaining the workflow

A passing demo remains a demo

Inputs

A sorting program with demo_fixture=True.

  1. 01

    Public challenges retain inputs; judging is recomputed separately.

  2. 02

    Invoke real V6.2 algorithm gates.

  3. 03

    Register predictions before execution and settle by candidate and task.

  4. 04

    The prosecutor independently recomputes key digests.

  5. 05

    Return PASS_DEMO with zero discovery credit.

Output

A demonstration package passing interface checks with no discovery credit.

Explained from source and test contracts; the historical package was not run here.

Changes in this version

hard audit

Earlier approach

V6.5 local three-gate imitation

This version’s approach

Real V6.2 lab and gate calls

Recheck actual executable objects

Challenges

Earlier approach

Public object carries an answer digest

This version’s approach

Separate public input from sealed judging

Reduce answer leakage

Credit

Earlier approach

Fixtures can receive discovery labels

This version’s approach

PASS_DEMO and zero multipliers

Preserve demonstration scope