v0.17.0: Policy crosswalk and arXiv-ready preprint

v0.17.0: Policy crosswalk and arXiv-ready preprint

Why this release matters

The paper argues that assigning a human role is not enough to prove practical human control. Institutions also need evidence that the person received usable information, could understand it, had authority and time to intervene, exercised judgment, and affected execution. Version 0.17.0 shows how selected public policy and standards sources relate to those six questions without treating a policy requirement as proof that control worked in a specific event.

What was added

• A bounded official-source review covering EU AI Act Articles 14 and 57, selected NIST AI RMF functions, the public ISO/IEC 42001 overview, and selected United States oversight rules.
• A crosswalk from those sources to the six-stage practical-control chain.
• A prospective EU AI Act Article 57 regulatory-sandbox design for testing the method on contemporary evidence.
• Five new claim-evidence records, eight official-source records, one author-review attestation, and nine controlled mutations.
• An updated 30-page v0.17.0 preprint, deterministic 12-member arXiv source archive, source manifest, compile receipt, citation metadata, and reviewer guide.

Integrity result

All nine prespecified policy-crosswalk corruptions were detected. The checks test whether source identities, scope labels, evidence links, claim permissions, and method-output relationships remain consistent. They do not determine whether a law has been interpreted conclusively or whether an institution complies with it.

Formatting result

The exact source compiles successfully in Overleaf with XeLaTeX. The final PDF uses the approved single-column navy design, keeps all displays before References, labels the work as a preprint, and formats references alphabetically with no bullets and a 0.5-inch hanging indent.

Preserved findings and limits

The historical case findings do not change: Oko remains unresolved, both Patriot cases fail the complete event-control rule, and none of the three selected cases passes. The cases remain purposive, retrospective, and assessed by one author. This release makes no prevalence, causal-effect, legal-compliance, ISO-conformity, policy-effectiveness, contemporary-transfer, or publication-acceptance claim.

The v0.14.0 Zenodo DOI identifies an earlier public preprint. It is preserved as version history and must not be assigned to v0.17.0.

v0.16.2: Paper workspace organization

v0.16.2: Paper workspace organization

Problem addressed

The paper/preprints/ directory placed the current v0.16.0 paper beside three retired v0.15.0 delivery files. A reader had to infer the latest paper from filenames and release history. This patch gives the paper workspace one explicit starting point and separates the retired package.

Current paper

The latest research paper remains v0.16.0. Its 30-page PDF, LaTeX source, deterministic source archive, metadata, source manifest, and compile receipts remain in paper/preprints/.

paper/README.md now opens with direct links to the current PDF, review and endorsement guide, current package, LaTeX source, GitHub reader edition, and archive index. paper/REVIEW.md gives a reviewer or prospective arXiv endorser one current PDF, a cs.CY category-fit summary, a version guide, supporting evidence paths, and five focused questions for any separate substantive review. paper/preprints/README.md identifies the PDF in its first sentence. The historical paper/arxiv/README.md now directs readers to the current review guide before they share or submit a paper.

Archived files

The following v0.15.0 delivery files move to paper/archive/v0.15.0/:

• preprints-compiled-v0.15.0.pdf
• preprints-source-v0.15.0.zip
• overleaf-compile-receipt-v0.15.0.json

The v0.15.0 Git tag and release manifest preserve their original release paths and checksums. The archive copies preserve their bytes. The v0.14.0 public preprint remains in paper/arxiv/ because its builders, validators, and released manifests use that path.

Historical archive repair

Validation found three historical delivery artifacts whose working-tree bytes differed from the immutable v0.14.0 tag. The source ZIP was truncated, and both preserved PDF paths contained the same 11-page replacement file. This patch restores the three exact released files. The source archive digest, 3bbee7693c6dec68b5b183d44893e907affa5d9bb94cecb5e628edb4f688b6d3, matches paper/arxiv/source-manifest.json and passes a complete ZIP integrity test. Both 25-page PDFs have digest 10d6a2a2000826ad07851595fa6a1a666829e556dd35be4bcb576a36e5d04107, which matches paper/arxiv/overleaf-compile-receipt.json.

The historical validator previously read the current paper/manuscript.md while checking v0.14.0 wording. It now reads paper/arxiv/main.tex, so version-specific abstract and marker checks stay attached to the package being validated.

Archive boundary

The archive receives completed delivery packages. Versioned research protocols, source-review decisions, screening ledgers, evidence tables, and review records remain in the main paper/ workspace because the current manuscript and integrity controls still depend on them.

Research boundary

This maintenance patch changes no manuscript text, source permission, case packet, assessment state, case-level event-control result, figure data, or interpretation boundary. The v0.16.0 result remains zero passing cases, two failing cases, and one unresolved case among the three purposively selected packets.

v0.16.1: Repository Alignment for the v0.16.0 Manuscript

v0.16.1: Repository alignment for the v0.16.0 manuscript

Why this patch exists

The v0.16.0 release contains the rebuilt manuscript, formal event-control rule, deterministic case results, figures, and integrity audit. Several current-state repository files still described earlier versions. The citation section named v0.12.0, the figure metadata named v0.15.0 as its source release, the audit protocol linked to v0.14.0 outputs, and the Overleaf receipt described v0.15.0.

Changing those sealed files under the existing v0.16.0 tag would invalidate its published checksums. This patch corrects the live repository while preserving the v0.16.0 release as an immutable historical snapshot.

What changed

• Corrects the reader-facing and machine-readable citation instructions.
• Identifies the v0.16.0 GitHub release as the current working-paper source.
• Preserves 10.5281/zenodo.21926005 as the DOI for the earlier v0.14.0 preprint and prevents that DOI from being assigned to v0.16.0.
• Adds a formula register for eight v0.16.0 formulas, a JSON schema, reusable LaTeX, plain-language explanations, implementation paths, and interpretation limits.
• Aligns the core figure manifest and figure register with source release v0.16.0.
• Corrects the Figure A3 description from 20 to 40 material claims and updates the current manifest language.
• Updates the Chain-of-Evidence protocol to the 40-claim v0.16.0 audit and its current attestation.
• Records the 30-page v0.16.0 XeLaTeX compilation in Overleaf and preserves the v0.15.0 receipt separately.
• Extends repository validation to check formula schema conformance, unique identifiers, source markers, implementation paths, and LaTeX identifiers.

What did not change

The patch changes no manuscript claim, source permission, case packet, assessment state, case-level event-control result, figure data, or interpretation boundary. The v0.16.0 result remains zero passing cases, two failing cases, and one unresolved case among the three purposively selected packets.

Citation boundary

The v0.16.0 working paper has no version-specific DOI. Cite its versioned GitHub release. The Zenodo v0.14.0 DOI remains a valid identifier for that earlier preprint. The machine-readable preferred citation in CITATION.cff makes this relationship explicit.

Remaining limits

This maintenance patch improves version clarity, traceability, and reproducibility. It supplies no new historical evidence, independent assessment, reliability estimate, population inference, contemporary-system transfer result, institutional-effectiveness finding, peer review, or publication acceptance.

v0.16.0: Practical Human Control Manuscript Rebuild

v0.16.0: Practical Human Control Manuscript Rebuild

Decision

This release rebuilds the paper around the question an institutional reader must answer: does a named human reviewer have practical power over one consequential automated decision? It preserves the released case states and search counts, then makes the case-level rule and its consequences explicit.

Why the revision matters

A role assignment shows who carries responsibility on paper. It does not show that the person received usable information, understood it, could intervene in time, exercised judgment, or changed execution. When an institution treats the role assignment as proof of control, it can count an ineffective safeguard as functioning and place responsibility on someone who lacked practical power.

Version 0.16.0 turns that broad concern into a connected documentary test. It also explains why each research control exists, what inference error it prevents, and which conclusion remains outside the evidence.

Formal result

The event-control rule requires six supported stages:

1. information access;
2. comprehension capacity;
3. intervention authority;
4. intervention feasibility;
5. exercised judgment; and
6. execution propagation.

The rule returns pass when all six stages are supported, fail when any required stage is unsupported, and unresolved when no stage is unsupported but at least one is partially supported or indeterminate.

Applied to the released states:

• Oko is unresolved because all six required stages are partially supported.
• Patriot ZG710 fails because four required stages are unsupported.
• Patriot F/A-18C fails because execution propagation is unsupported and three other stages are indeterminate.
• No selected case passes.

These are results for three purposively selected public packets. They provide no prevalence, causal, reliability, safety, legal, or contemporary-system transfer estimate.

What changed

• Rewrote the abstract, introduction, research gap, contribution, method, results, discussion, institutional implications, and conclusion in problem-mechanism-evidence-limit form.
• Added a construct-derivation table that shows why each stage is required, its minimum evidence, and the inference error it prevents.
• Added a formal authority versus event-control table and an institutional interpretation table.
• Added a deterministic result builder and machine-readable v0.16.0 case-level output.
• Added a proposed timing margin for future trace-rich systems and recorded that the historical packets cannot supply its inputs.
• Replaced the reader-facing term protective effect with execution propagation; the legacy JSON field remains effect.
• Rebuilt Figures 2, 6, A3, and A4, their data, specifications, and reading guides.
• Extended the claim map from 35 to 40 material claims.
• Extended the negative-control suite from 33 to 39 controlled corruptions.
• Detected all 39 corruptions and published five exceptions.
• Recorded the Preprints.org decline as an external venue state without claiming which screening criterion controlled the decision.

Integrity result

The v0.16.0 Chain-of-Evidence audit returns PASS_WITH_EXCEPTIONS. Traceability, exact-value verification, reference resolution, method-code alignment, evidence fitness, dependency closure, and negative controls cover the new formal claims. This means the declared artifacts agree and the controls respond to seeded errors. It does not mean the historical interpretation is independently correct.

Open evidence gates

• independent assessor agreement;
• one contemporary learned or agentic system with sufficient trace data;
• validation of the proposed timing margin;
• 980 unrecorded retrieval outcomes;
• the frozen 177-record backward-reference stratum;
• authenticated and disciplinary database searching;
• construct validity, institutional effects, and peer review.

Version relationship

The v0.14.0 preprint remains public under Zenodo DOI 10.5281/zenodo.21926005. The v0.15.0 venue package remains version history. The exact v0.16.0 source compiles to a 30-page PDF with zero errors and no overfull or underfull boxes. Representative-page inspection confirmed display order and found no table or figure after References. Version 0.16.0 remains a working paper until Mark Julius Banasihan completes full author review.

v0.15.0: Preprints.org Submission Candidate

v0.15.0: Preprints.org Submission Candidate

Decision

This release adapts the public v0.14.0 preprint for a second preprint venue while preserving the research record. It does not change the research question, case selection, case assessments, quantitative findings, or bounded conclusions. The earlier public preprint remains citable under Zenodo DOI 10.5281/zenodo.21926005.

What changed

• Identifies Mark Julius Banasihan as Independent Researcher, Node & Norm, United States.
• Preserves the ORCID and both author-approved correspondence addresses.
• Records Harvard University student status in a separate author note without claiming University sponsorship, supervision, endorsement, or representation.
• Adds venue-aligned version, author-contribution, conflicts-of-interest, AI-assistance, and data-availability statements.
• Synchronizes the repository with the author-edited Overleaf source and preserves the exact 25-page compiled PDF.
• Packages a deterministic 12-member source archive containing the LaTeX source and ten figures.
• Extends the claim map to 35 material claims and keeps the final upload and submission decision visible as an author-controlled gate.

Control result

The release applies claim traceability, five-part evidence fitness, dependency closure, controlled mutations, reference verification, and method-code alignment to the venue package. These checks establish whether the declared artifacts agree and whether the controls detect the seeded errors. They do not establish source truth, independent assessor agreement, construct validity, population prevalence, causal effect, institutional effectiveness, or transfer to current learned systems.

Publication state

The v0.14.0 preprint is public on Zenodo. Version 0.15.0 is a repository and Overleaf submission candidate. It does not claim Preprints.org acceptance, moderation, publication, or peer review. The exact uploaded files and the final submit action remain under Mark Julius Banasihan's control.

Included venue artifacts

• paper/preprints/main.tex
• paper/preprints/preprints-source-v0.15.0.zip
• paper/preprints/preprints-compiled-v0.15.0.pdf
• paper/preprints/source-manifest.json
• paper/preprints/overleaf-compile-receipt.json
• release/v0.15.0-manifest.json
• evidence/claim-evidence-map.json
• evidence/human-review-attestation-v0.15.0.json
• audits/v0.15.0/

v0.14.0: Proposition Review and Public Preprint

v0.14.0: Proposition Review and Public Preprint

Decision

This version moves the paper from source screening to proposition-level permission and packages a bounded public preprint. It carries the v0.5 Chain-of-Evidence, evidence-fitness, ScientistOne-inspired lineage, mutation testing, and CoE Integrity Audit controls through the manuscript and source archive. The exact v0.14.0 preprint is archived on Zenodo under DOI 10.5281/zenodo.21926005.

What changed

• Reviewed all 13 close forward-citation sources under a frozen proposition protocol.
• Granted five exact, locator-bounded manuscript permissions.
• Kept two sources as background-only and quarantined six from manuscript claims.
• Closed the fifth direct-query screening decision while preserving zero proposition permission for its unreadable source text and unresolved author identity.
• Updated the manuscript, literature matrix, reference file, tables, reader edition, claim register, crosswalk, and research lineage.
• Added a 230-word abstract, professional single-column LaTeX source, canonical 25-page review PDF, navy-and-black journal styling, ten color figures, ten hash-traced grayscale alternatives, deterministic Overleaf and arXiv source archive, and source manifests.
• Imported the archive into Overleaf, compiled the canonical 25-page PDF with zero errors and no box-overflow notice, visually checked 14 representative text, table, figure, and References pages, and recorded source-to-output hashes and display locations in a compile receipt.
• Added major-section float barriers and an automated placement audit so all ten figures and seven tables must remain before References.
• Expanded the current claim map to 32 claims and the mutation suite to 33 negative controls.

Control result

The current CoE Integrity Audit returns PASS_WITH_EXCEPTIONS: 32 claims are traceable and all 33 mutations are detected. The result means the repository catches the corruption conditions it declares. It does not establish independent reliability, field validity, search completeness, institutional effectiveness, legal sufficiency, safety, source truth, or originality.

Open decisions

Zenodo publication closed the public-archive and version-DOI gates. arXiv remains blocked by its category-endorsement requirement. Preprints.org venue review is being prepared as v0.15.0 with the v0.14.0 Zenodo relationship disclosed. Authenticated database coverage, 980 retrieval outcomes, the 177-record backward-reference tranche, and independent assessment remain open limitations.

Release posture

The preprint itself is public on Zenodo. arXiv submission remains a separate platform process and is not implied by the Zenodo publication.

Repository-tag provenance

The GitHub tag was created retroactively from the first bundled v0.14/v0.15 preparation commit because the v0.14 repository work and the initial venue adaptation entered draft PR #26 together. The tag therefore contains early, unreleased v0.15 preparation files. The citable v0.14 research artifact is the exact PDF and source archive preserved under paper/arxiv/ and Zenodo DOI 10.5281/zenodo.21926005. The resealed manifest records the actual tag contents instead of representing the tag as a pure pre-v0.15 state.

v0.13.0: Forward-Citation Author Screening

What changed

This release closes the author-screening gate for the 71 forward citations whose content was recovered in v0.12.0. The decisions retain 13 close sources and 22 background sources, exclude 11 single-component sources and 25 off-topic sources, and preserve a record-level reason for every decision.

Why it matters

The repository now distinguishes three questions: whether a source was recovered, whether it belongs in the research corpus, and whether it supports a manuscript proposition. The first two questions are closed for this 71-record queue. The third remains open. No retained source has claim permission until a separate locator-level review connects an exact proposition to inspected evidence.

Evidence controls

• The screening protocol was fixed before the decisions were recorded.
• The input queue hash, membership, order, and 71 decisions are checked by executable controls.
• Every row records its rationale, inspected basis, source locator, decision owner, date, assistance disclosure, and claim-permission state.
• The claim map contains 28 material claims.
• The integrity audit detects all 26 prespecified corruptions and publishes four exceptions.
• The release manifest seals 198 artifacts with SHA-256 digests.

Limits

The result is an author-authorized, AI-assisted single-assessor screening pass. It supplies no inter-assessor reliability estimate, proposition-level support from the retained sources, close-source prevalence estimate, completed multidisciplinary search, originality finding, institutional effect, or transfer result. Another 980 retrieval outcomes, one direct-query screening decision, backward-reference retrieval, authenticated database searches, and retained-source proposition review remain open.

Author: Mark Julius Banasihan
ORCID: 0009-0001-8121-2878

v0.12.0: Forward-Citation Retrieval and Open Author Gate

v0.12.0 Release Notes

Decision served

This release candidate records one retrieval outcome for every record in the frozen 102-record forward-citation stratum. It makes the next author decision gate visible before any newly recovered source can affect the manuscript.

Added evidence

• 34 full-text recoveries;
• 37 abstract recoveries;
• 26 metadata-only outcomes;
• 3 duplicate-version outcomes;
• 2 unavailable outcomes;
• a 71-record author queue with blank decisions and no claim permission;
• route, source-observation, review-basis, assistance, and claim-limit fields for every selected record;
• a deterministic builder and validator cross-check against the frozen sample and population ledger.

Integrity result

The v0.12 claim map contains 27 material claims. PAPER-C33 records the forward-citation workflow counts. Its human support gate remains pending, so the claim is not conclusion-eligible. The audit detects all 22 existing controlled corruptions and publishes five exceptions.

Why the open gate matters

Retrieval shows that source material can be inspected. Screening decides whether a source is close, background, single-component, off topic, or ineligible by record type. Treating those stages as one decision would allow machine-assisted retrieval to become an unrecorded author judgment.

Unresolved work

• author decisions for 71 recovered forward-citation records;
• 980 retrieval outcomes in the full inaccessible population;
• 177 records in the frozen backward-reference stratum;
• two direct-query source limits;
• five authenticated or disciplinary-interface searches;
• independent assessment, ethics guidance, and venue packaging.

The three frozen topics, project question, public cases, and released assessments do not change.

v0.11.0: Direct-Query Evidence and Claim-Control Checkpoint

Why this release matters

Version 0.11.0 applies the claim controls introduced in v0.5.0 to the newest research work. It does not treat a recovered source, a screening decision, or a plausible comparison as a conclusion merely because it appears in the repository.

The release separates six questions: Was the source recovered? What part was inspected? What decision did the author record? Which bounded claim does the evidence support? Is the evidence fit for that exact claim? Do the prerequisites close before the claim enters a conclusion?

What this release adds

• A frozen 284-record residual-risk sample drawn from the 1,087-record recovery population.
• Five direct-query retrieval outcomes with route, locator, review basis, source observation, screening state, decision owner, assistance disclosure, and limits.
• Four recorded screening decisions: three background sources and one close source. One decision remains open.
• A 60-source literature matrix and an updated contribution audit.
• A human-review attestation covering the five direct-query records and six added claims.
• A deterministic 26-claim evidence map derived from the preserved v0.9 map.
• A v0.11 integrity audit that detects all 22 controlled corruptions.
• Four published exceptions covering independent assessment, incomplete search coverage, two source limits, and institutional transfer.
• A SHA-256 manifest sealing 171 research artifacts.

Audit result

The audit returns PASS_WITH_EXCEPTIONS. The five-record workflow claim and four bounded source descriptions pass the declared gates. PAPER-C32, a proposed transfer from a model-level residual-path result to a wider control mechanism, remains conclusion-ineligible because independence and completeness are indeterminate.

A passing control means the repository detected the corruption it was designed to detect. It does not mean a source is true, the method is independently reliable, or an institution will obtain a better outcome.

Standing research rule

Exploration may begin with incomplete notes. Once data, a source observation, a figure, or a finding is used to support a material claim or release statement, it must enter the claim register and pass traceability, integrity, human support review, evidence fitness, and dependency closure. Failed or indeterminate required gates must remain visible and must block the affected conclusion.

Next action

Resolve the two direct-query source limits, then retrieve and screen the 102 forward-citation records. The remaining 1,082 recovery outcomes, five authenticated or disciplinary-interface searches, independent assessment, and institutional testing remain open.

Full release notes

v0.10.0: Evidence-Gate Protocol Checkpoint

Why release this checkpoint now

The repository completed all 89 author-screening decisions in v0.9. That result exposed a second problem: screening relevance and manuscript support require different evidence. Of the 27 retained-close sources, 8 have a recorded full-text review basis and 19 remain supported by title and abstract.

Version 0.10.0 freezes the next rules before those 19 decisions and the wider search-coverage results are known. The release therefore supplies a dated, inspectable record of the method that will govern the next evidence cycle.

What this release adds

• A 27-source full-text ledger with 8 verified and 19 open records.
• Terminal states that separate verified full text, quarantined abstract-only use, post-review exclusion, and inaccessible sources.
• A recovery protocol for all 1,087 records that lacked abstracts in the formal search.
• A reproducible 284-record residual-risk sample when unresolved records remain.
• Accountable logs for Scopus or Web of Science, ACM Digital Library, IEEE Xplore, PhilPapers, and HeinOnline or Harvard HOLLIS.
• A discovery log recording eight findings that changed the research sequence while preserving the three topics and frozen project question.
• An executable validator and a SHA-256 manifest sealing 154 research artifacts.

Evidence boundary

This release establishes protocol fixation, source-state accounting, executable gate behavior, and artifact integrity for the declared checkpoint. It establishes no completed full-text review, exhaustive search coverage, independent reliability, field validity, institutional effectiveness, universal originality, or improved outcome claim.

The released v0.9 claim audit remains immutable. A v0.10 claim audit will follow after the new evidence gates advance.

Next action

The next cycle will verify or quarantine the 19 retained-close sources, then begin recovery of the 1,087 inaccessible records under the frozen protocol.