Public editorial charter

The xResearch library centralizes and interprets scholarship bearing on the educational and measurement problems HeuriSight addresses. Its publications are field-facing scholarly works for faculty, teaching-and-learning leaders, administrators, researchers, and specialists. They are neither product marketing nor internal permission audits.

Each publication must:

1. Begin from a real problem or disagreement in the field.
2. Define its constructs before asking them to carry a finding.
3. Represent relevant prior work at full strength and in the authors' own terms.
4. Separate established evidence, interpretation, design proposal, current implementation status, and untested implication.
5. Report study conditions, comparison classes, effect sizes, uncertainty, and material criticism where the source provides them.
6. State null, mixed, or absent evidence without forcing a supportive conclusion.
7. Translate the evidence into consequences for the intended reader without changing the underlying finding.
8. End by locating the contribution inside the larger unfinished problem.

The provisional portfolio spine is:

As generative AI makes competent-looking artifacts abundant, higher education needs better evidence of the human decisions, reasoning, contribution, and independent performance behind them. Faculty expertise should remain visible and governable; AI assistance should preserve rather than substitute for consequential learner work; and any process trace used as evidence of learning must be validated against acquisition, transfer, and durability.

This spine is a program hypothesis to be tested against the full corpus. It does not authorize stronger claims than any individual paper's evidence supports.

The stable signature across publications is a sequence of intellectual moves rather than a sentence template:

1. Locate the gap in the field's own logic. State the canonical position fairly, then identify its unmeasured precondition, conflated level, or unresolved internal tension.
2. Reconcile by structure, not compromise. Where durable camps use one term for different objects, split the term, identify their relationship, and make the relationship testable.
3. Define borrowed structures in their source domain. Preserve provenance, internal distinctions, limitations, and the finite jobs the analogy will perform.
4. Build a portable analytic spine. Reuse a validated sequence deliberately; disclose textual and intellectual reuse at the point of reuse.
5. Keep one claim per paragraph. Paragraph length follows the apparatus the claim requires and the venue in which it appears.
6. Give every number its comparison class. Preserve denominators, uncertainty, distribution shape, and the precision actually identified by the design.
7. Match the display to the claim. Use prose for interpretation, tables for exact discrete comparison and verification, diagrams for relationship and process, and charts for magnitude, trend, and distribution.
8. Close the ring. Return to the opening construct with its meaning changed, widen only after the evidence has earned the stake, and give the final position to what becomes possible next.

A paper and its companion essay form one intellectual unit with two expressions.

• The paper carries the full evidence base, methods, prior-art distinctions, qualifications, tables, and validation requirements.
• The essay preserves the core claim while changing load order, examples, rhythm, and consequences for a broader faculty and institutional audience.
• The essay is not a compressed paper and may not invent a stronger or more dramatic finding.
• Reused wording, frameworks, tables, or figures are disclosed briefly where reused.
• The paper and essay may share a signature construct or figure only when its rhetorical job differs and the relationship is stated.

1. Peer-reviewed primary scholarship is preferred; preprints are clearly identified.
2. Vendor material is excluded as evidentiary support unless the publication's subject specifically requires analysis of vendor claims, in which case it is labeled as such.
3. Every material empirical statement resolves to a verified source-ledger record.
4. Search snippets, abstracts alone, agent summaries, and HAG guidance do not constitute evidence.
5. Effect sizes are reported with the population, design, comparator, outcome, interval or uncertainty if reported, and conditions.
6. Studies are not pooled into a new number without a prespecified and methodologically defensible meta-analysis.
7. Source counts describe coverage, not evidential strength.
8. A central claim involving learning, novelty, validity, fairness, or product capability receives independent verification by a second reader.
9. HeuriSight is never described as having demonstrated an effect that has not been demonstrated.
10. Current implementation status is stated once, plainly, where material to interpreting the construct.

Internal research may use adversarial language such as audit, verdict, occupied ground, failure mode, or claim test. Public papers translate that work into scholarship:

• field problem rather than company defense;
• evidence status rather than permission;
• strongest prior art rather than competitive dismissal;
• proposed synthesis rather than product assertion;
• limitations as the boundary of inference rather than a prosecutorial conclusion;
• a research agenda rather than a pass/fail verdict.

Honesty requires proportion. A paper should neither suppress inconvenient evidence nor make every section a critique of HeuriSight.

Canonical public front matter

Every publication uses a small common spine so readers can identify the work and its evidentiary status without decoding internal workflow labels:

1. title and stable publication identifier;
2. HeuriSight xResearch plus the public genre (working paper, research essay, or methods and authorship note);
3. publication or revision date and an evidence-current date where these differ;
4. a compact evidence-status statement in the vocabulary of scholarship—for example, narrative review, construct paper, empirical study, preregistration, or synthesis;
5. the linked companion publication and source ledger, where applicable; and
6. a brief reuse or provenance note wherever intellectual or textual material is shared across publications.

Working papers foreground research question, method, evidence status, and place in the research program. Essays foreground the interpretive question, the relevant evidence base, and the consequence for faculty or institutional practice. They do not imitate paper abstracts or methods sections.

Draft, internal verification edition, clearance language, handoff instructions, and temporary workflow labels do not appear in a public release. A genuine unresolved authorization, ethics, or reproducibility condition is stated as a substantive scope limit at the point where it affects inference. Event-specific framing appears only when the publication is genuinely an event artifact; a public research essay is not defined by the workshop at which it may be used.

The title block remains short. It does not combine conflicting labels such as draft, published, working paper, and essay. Status is controlled metadata; evidentiary limits belong in the argument.

Essential decisions concerning evidence priorities, construct interpretation, thesis, reframing, structure, audience, reuse, tables, figures, and conclusions are queried against the relevant HAG model before finalization.

For each material decision, record:

• model and query;
• retrieved heuristic name and ID;
• decision context and alternatives;
• applied, adapted, rejected, or unavailable status;
• evidence or reasoning supporting the disposition;
• editor decision and human-author checkpoint where required.

Retrieval is not obedience. A heuristic may be rejected when its domain, preconditions, evidence, or consequence does not fit. No long-term HAG update occurs without explicit human authorization.

The WP-09 Section 6 boundary applies throughout the program whenever HAG, heuristic retrieval, longitudinal graphs, or related mechanisms are discussed.

Public material may describe constructs and literature at the approved abstraction level. It must not reveal code, service or function names, pipeline internals, ranking or prompt structure, implementation thresholds, gating conditions, graph schemas or property names, private metrics, undisclosed volumes, or reconstructable parameter choices.

Release test:

Could a competent engineer materially reconstruct the implementation from this publication or figure?

If yes, remove or abstract the material. When a limitation cannot be described without revealing implementation, state it at the construct level rather than omitting it.

A paper/essay pair is complete only when:

• its shared core claim is explicit and evidence-calibrated;
• the paper and essay perform different genre-appropriate jobs;
• terminology matches the canonical glossary;
• all material citations resolve to verified sources;
• preprints and non-peer-reviewed records are labeled;
• tables and figures have distinct argumentative or evidence-store roles;
• every display has a standalone title, source, note, accessibility text, and transformation disclosure where relevant;
• reused material is disclosed at point of use;
• public-register, authorship, conflict, current-status, and trade-secret checks pass;
• the conclusion returns to the opening problem and identifies the next evidentiary step;
• the human author has reviewed the substantive changes and approved release.

Every publication carries an evidence-current date and a stable identifier. Material revisions preserve a revision note. Corrections distinguish factual correction, interpretive change, updated evidence, and formatting change. Silent replacement is avoided once a version has been made public.