v0.2.1 - Provenance and release-hygiene correction

v0.2.1 - Provenance and release-hygiene correction

Released 30 July 2026.

This is a documentation, provenance, and release-hygiene correction to the
immutable v0.2.0 RB-003 release.

Mathematical result unchanged

This release changes no:

• mathematical claim or theorem statement;
• constant or extremizing family;
• exact finite certificate;
• proof conclusion;
• encoded mathematical instance; or
• verifier result.

In particular, the released RB-003 conclusion remains the unrefereed
fixed-gadget, two-scenario cost-nonincrease supremum

[
\beta_G^{(2\mathrm{sc})}=17/8,
]

with the same non-attainment statement and the same epsilon = 1/1000
certificate.

Corrections and additions

• Reclassifies the two role-separated hostile-review rounds accurately as
  AI-assisted model critiques, not external human mathematical review.
• Removes stale “sign-off candidate”, “external referee”, and pending
  re-signature language left in the v0.2.0 package.
• Publishes RB-003 paper revision 2 with the corrected review and release
  status. Its mathematics is unchanged from revision 1.
• Corrects the stale proof-integration checker filename in the audit record.
• Updates deterministic release metadata, PDF paths, tests, manifests, and the
  permanent read-only reproducibility workflow for v0.2.1.
• Retires the completed one-time write-enabled v0.2.0 publication workflow.
• Adds the stable-only agentic-mathematics case study, contribution guide, and
  bounded problem-reporting protocol.

Review status

No external human mathematical review has been requested or documented for
RB-003. The human-readable proof remains authoritative. Exact computation and
the AI-assisted critique records are narrower supporting evidence, not formal
peer review.

The immutable v0.1.0 and v0.2.0 tags and assets remain unchanged.

v0.2.0 - Exact two-scenario fixed-gadget supremum

v0.2.0 - Exact two-scenario fixed-gadget supremum

Release date: 28 July 2026

Main result

This release adds RB-003:

$$ \boxed{\beta_G^{(2\mathrm{sc})}=\frac{17}{8}=2.125} $$

for the released four-terminal planar acyclic gadget under two simultaneous
positive scenario-wise cost-nonincrease constraints.

The value is a non-attained supremum. A rational extremizing sequence has
intrinsic feasible C-family

$$ \uparrow{123,124,234}. $$

A finite exact certificate at epsilon = 1/1000 has normalized objective
1061/500 = 2.122; after integer scaling, every route respecting both scenario
budgets has upper deviation at least 8488 with maximum demand 4000.

Model semantics

A route is feasible when it does not increase either scenario cost:

$$ C_j(y)\le C_j(x),\qquad j=1,2. $$

Equality is not required. The objective is normalized additive upper arc
deviation, not a multiplicative total-cost or capacity ratio.

Review history

The theorem package underwent two hostile review rounds. The second proof-only
review found no theorem-level gap and recommended acceptance after local proof
revisions. The final source integrates those revisions, including:

• explicit two-colour matching logic;
• the generalized one-knapsack lemma and unit-profit consequence;
• the exact shared-baseline |A|=3 star-triangle core;
• the full non-attainment equality chain; and
• corrected scope and commercial language.

This remains unrefereed research, not formal peer review.

Reproducibility additions

• self-contained RB-003 paper and proof package;
• exact 16-routing graph-native certificate;
• exact threshold-recognition registry (149 + 18 + 1 = 168);
• deterministic replay authenticated before and after execution;
• proof-text regression and mutation guards;
• updated deterministic release tooling and manifests;
• explicit checker-assurance, AI-contribution, validation, and stewardship
  records.

Agentic-science context

The release records the division between human scientific responsibility and
agent implementation/reasoning work. It adopts externally checkable acceptance
targets, staged validation, explicit adjudication, and maintenance ownership.
OpenAI's 28 July 2026 field report on agentic scientific computing is cited as
contemporaneous methodological context only; it is not evidence for or an
endorsement of RB-003.

Preserved boundaries

• The immutable v0.1.0 tag and assets are unchanged.
• No open-source or open-content license is granted.
• No institutional ownership, sponsorship, or affiliation is asserted.
• No private Compliance Health product code, rule library, customer data, or
  uncleared product brand is included.
• The arbitrary one-scenario fixed-graph constant, many-scenario extension,
  and bounded-heterogeneity function remain open.

v0.1.0 — Four-terminal planar lower bound

v0.1.0 — immutable research disclosure

This release publishes the manuscript, exact finite certificate, symbolic
checks, and reproducibility materials for the four-terminal construction.

Included claims

• Exact finite certificate: 335/294, verified over all 16 routings and all
  13 arcs.
• Limiting lower-bound family:
  (299 - 41*sqrt(41))/32 = 1.139747070789....
• Restricted sharpness theorem for the fixed real equal-full-cost,
  two-cheap-choice model, with the same supremum for rational data.

Verification

• exact primary verifier and exhaustive routing table;
• symbolic family and restricted-optimality checks;
• representative mutation tests;
• separate clean-room graph, finite, symbolic, and deterministic stress
  cross-check;
• PDF text preflight, secret scan, manifest verification, and visual review.

The separate implementation is AI-assisted computational corroboration, not
independent human review. The manuscript is unrefereed.

Research provenance

The work grew out of a multi-week conversation between Matthew Protti and
OpenAI GPT-5.6 Pro about similarities among mathematical breakthroughs and
the kinds of difficult problems AI systems might help with. GPT-5.6 Pro did
most of the active mathematical heavy lifting; Matthew framed and directed
the inquiry, challenged the work, required exact verification, chose the
scope of the claims, and accepts responsibility for the release. The private
conversation transcript is not included.

Immutable assets

• ssuf-four-terminal-v0.1.0-source.tar.gz;
• ssuf_four_terminal_note_v5.pdf;
• SHA256SUMS.txt.

Later corrections would be issued as new versioned releases. Once the
repository's immutable-release setting applies to a release, its tag and
assets must not be altered.