Prepared invitations and announcements

Unsent drafts. Replace bracketed destinations with the approved public URLs before use. Tailor recipient, relationship, disclosure and request; do not bulk-send. No organization named here is a project partner or endorser.

Physics / astronomy methods seminar

Subject: A calibration correction as a test of a review protocol

We are preparing a research prototype of MCRP, the Minimum Credible Reproducibility Protocol. The question is narrow: when a calibration changes, can a downstream reader tell which checked claims and uses need reconsideration?

The attached synthetic astronomy case separates a reproducible scalar calculation from a shared calibration uncertainty. It then follows a material change through a dependent estimate. The toy code deliberately includes a missing-dependency case that the protocol cannot discover by itself.

Would a short methods discussion around that example be useful? The proposed exercise is to identify one dependency the record misses, one check that is too broad, and one field that could be removed. No adoption or transfer of review authority is requested. We would bring a one-page human exercise and runnable code for participants who prefer to start there.

The packet was developed with substantial AI drafting and adversarial-agent review; it is not externally peer reviewed or empirically validated. The desired result is a concrete counterexample from people who know the scientific work.

Prepared materials: [approved prototype URL]; [astronomy case URL].

Biology / reproducibility journal club

Subject: What does an agreeing rerun establish when batches are confounded?

Our MCRP prototype asks contributors to distinguish an offered claim, a specific check, a scoped decision to rely, and a material amendment. The biology teaching case starts with a calculation that reproduces exactly but cannot separate a treatment effect from a batch effect. Repeating the calculation does not create the missing experimental information.

We would like to test whether this distinction is easy to communicate without a large reporting burden. Participants would first write a normal review paragraph, then a short version-bound record, and finally respond to changed evidence. The useful output is a better example or an objection: where does the record imply more biological knowledge than the experiment supplies?

The exercise uses synthetic values and no patient or confidential laboratory data. It is a methods discussion, not a clinical recommendation or human study. The AI-assisted packet includes executable toy models and internal adversarial reviews, with their limits disclosed.

Prepared materials: [approved prototype URL]; [biology case URL].

Subject: A review-allocation model that fails after selective completion

A fair invitation lottery need not produce a fair set of completed reviews. Our MCRP prototype uses that gap as a worked mechanism-design example. Duplicated labels, strategic refusal, retries and scarce audit labor can each change who ultimately receives attention. Group-first selection repairs one representation problem under a known-control assumption; it does not solve strategic participation.

We are preparing a reproducible model-and-code packet and would value a short methods critique. The proposed contribution is one counterexample to a stated incentive or selection condition, or a simpler allocation rule that performs as well under the same labor budget. We count offered and unresolved work alongside completion and reliance, so exclusion cannot masquerade as efficiency.

For nonprogrammers there is a small causal-identification exercise and a review card. For implementers there are toy agents and exact model checks. All numerical results are synthetic; AI drafting and shared-operator adversarial review are disclosed. We are seeking criticism before making efficacy claims.

Prepared materials: [approved prototype URL]; [economics case URL].

Law / legal informatics clinic

Subject: Separating a verified citation from authority to rely on it

MCRP is a proposed record boundary for claims, performed checks, scoped reliance and later changes. Its legal teaching case asks what can be preserved when a source is accurately quoted but its authority, jurisdiction, timing or permitted use remains contested.

We would welcome a clinic-style critique of that example. Participants would identify which decision belongs to a researcher, editor, operator or legally authorized actor; what information an affected party must receive; and which deadlines cannot be represented as an ordinary review queue. The toy runtime deliberately cannot grant legal authority.

The exercise uses fictional facts. It requests no client information, sensitive dispute or legal opinion about a real person. The research packet was drafted with AI assistance and internal adversarial review, not professional clearance. The desired output is a corrected scenario and a clearer boundary between scientific disagreement and operational obligations.

Prepared materials: [approved prototype URL]; [law case URL].

Short public announcement

MCRP asks: what can we rely on, for which use, and what should change our mind? This research prototype offers four actions—offer, check, rely, amend—with toy agents, mathematical counterexamples and cases in astronomy, biology, economics and law. Try one claim; bring one failure. [approved URL]

AI-assisted research prototype; synthetic results; no field-efficacy claim.

Three-part launch sequence

The question: A review checked a version. After a material change, what happens to the downstream use? Introduce the four-action record and link the human exercise.
The counterexample: Show one plot where fair invitations become skewed completions. Link the assumptions, full denominators and runnable model.
The invitation: Ask each domain for one case the model mishandles. Link a bounded contribution template and the next correction milestone.