01
How criticism changes research
Lakatos asks whether templates, provenance labels and long reviews can produce increasingly polished packaging without stronger research. V6.4 turns criticism into counterexample programs, discriminating experiments or prior references; rival mechanisms bet on one observable, and unexplained anomalies create the next question. The unit broadens into a programme with a hard core, auxiliary hypotheses and history.
02
Engineering proofs and refutations
Provenance has operational consequences. A fallback_template may be archived, create questions and face cheap checks, but cannot promote itself to promising. Pure prose criticism remains a note; only valid executable forms affect gates. Programmes are classified and budgeted by settled predictions, requiring disagreements to produce discriminating consequences.
03
Duels, anomalies and endogenous agendas
Candidates receive provenance stamps, reinstantiation and null-model checks before paired bets with opposing predictions. Settlements update Elo; unresolved residuals enter an anomaly ledger and become QuestionGenome entries in a HammingList. Programmes record predictions and revisions, L operators alter auxiliary assumptions, and programme states guide budget reallocation.
04
Discriminating bets and executable criticism
ForkingObservable requires differing predictions, a task reference, adjudication rung and positive cost. PairedBet records two candidates, prior probability, provenance and evidence, allowing winner, undecidable and void states. Critique identifies its target and kind: programs name the failed claim, specifications provide rival predictions and prior references cite a ledger. Pure text has no gate effect.
05
Deliver agendas and states
Inputs include topic, candidate provenance mode, pool size, settlement budget and adjudicator selection. Outputs include provenance and reinstantiation checks, candidate verdicts, bets, settlement Elo, anomalies, questions, programme states and budget changes. The five local themes—provenance, paired settlement, executable criticism, programme ecology and endogenous agendas—are deterministic construction material.
06
Conceptual walkthrough: distinguish two mechanisms
Suppose two explanations predict an increase and a decrease under one intervention. The bet records the observable, predictions, task and cost before seeing the result. If the outcome is nondiscriminating, it remains undecidable rather than crediting either side; unresolved residuals can become questions. This is a contract example, not a real intervention study by V6.4: default anomalies come from synthetic smoke.
07
Stop templates from self-promotion
Relative to V6.3, V6.4 turns provenance ceilings, reinstantiation, null-model checks and settlement accounting into PG gates, while moving Elo from heuristic evaluation to settlement-based updates. Programmes and anomaly-derived questions extend the population approach. Feynman next asks whether provenance stamps and records actually show responses to fresh world challenges.
08
Wiring versus heuristics
Source wires PG gates, paired bets, Elo, anomaly ledgers, question generation, programme demarcation and budget reallocation. Default adjudication and residuals still include controlled demonstrations. The EIG queue is an engineering ordering from prior probability and cost, not a validated information-gain guarantee. Reinstantiation fields do not establish two independent external-model calls.
09
A historical programme-ecology prototype
This version connects failure, disagreement and the next agenda, but software states cannot substitute for findings. Provenance stamps require actual invocation records, observations require original tasks and settlement requires genuine rival predictions. The fallback ceiling is implemented; open-science novelty, research-resource efficiency and long-term programme progress still require independent task evidence.
Reading example · Explaining the workflow
Turn criticism into the next check
Two mechanisms make opposing predictions on one observable.
- 01
Record both identities, the task, cost and prior probability.
- 02
Express criticism as a discriminating specification or counterexample program.
- 03
Settle from evidence; keep nondiscriminating outcomes undecidable.
- 04
Turn an unexplained anomaly into a new QuestionGenome.
- 05
Record the revision in programme history and reallocate budget.
A continuous record of the bet, anomaly and next question.
Conceptual walkthrough; the default residual demonstration is not a real-world task result.
Changes in this version
Provenance
- Earlier approach
Templates can receive provisional high states
- This version’s approach
Fallback is capped at quarantine
Prevent structural self-promotion
Criticism
- Earlier approach
Prose criticism and scoring
- This version’s approach
Only executable criticism affects gates
Make disagreements inspectable
Agenda
- Earlier approach
Individual candidates and generation archives
- This version’s approach
Anomalies, questions and programme ecology
Organize the next research question