On this page
Evidence and model-assurance pilot
The proposed pilot produces a traceable report showing what one pinned mathematical source supports for a software claim. Its commercial value would be fewer unsupported guarantees and less reviewer preparation/correction effort. The experiment is prepared and unperformed; no paid pilot, buyer demand or independently verified proof backend exists.
Current reviewable components
All 372 families and 719 manuscripts are inventoried at fd4aeeb2ee4fc729c18d98444fed42fd0529eeeb. The current eighteen-contract registry contains exact source file pointers and hashes. The subset-sum report joins 2,933 source hashes, static configuration/withdrawal metadata and 18 documentary gaps in one self-authored control; six exact input files are archived. Complete documentation grants no evidence-content or applicability certificate.
The existing ten-case packet retains eight self-authored scope-error scenarios and two synthetic narrower documentary controls, its separate answer key, source/report cards, blank reviewer/timing fields and archived thirteen-contract tool/input edition. Current registry expansion does not alter that packet or create an expert result. Its 25 controls test fixture/code consistency, not reviewer accuracy or buyer value.
User flow and records
A reviewer selects a source revision, precise statement and proposed use. The report exposes quantifiers, assumptions, selected formal coverage, implementation/resource obligations and unknown or contradicted claims. Source/model/backend/reviewer versions and exact evidence bytes accompany each decision. A complete reference list is separate from a verified theorem or software certificate.
| Record | Required evidence | Purpose |
|---|---|---|
| Source snapshot | commit/path/hash/retrieval time | Reproduce reviewed bytes |
| Statement and model | quantifiers, assumptions, exact selected scope | Prevent unsupported use/model extensions |
| Implementation bridge | code/version/encoding and validated correspondence | Separate theorem existence from actual program |
| Verification run | locally produced authenticated run/trust metadata | Prevent unsigned logs from becoming independent proof |
| Finite calculation | input/tool/runtime hashes, completed domain and budgets | Reproduce a bounded result without generalizing it |
| Reviewer decision | identity, independent adjudication, timing and corrections | Measure actual usefulness |
Use versioned JSON and local reports initially; add a SQLite read model only after query needs exist. Corrections create dated superseding records. Never infer valid evidence from a file's existence, a supplied supported status or a solver label.
New concrete workflows available for review
The public statistical report now has 586 live SciPy controls validating tiny fixed-event/law calculations against an existing API. It illustrates the exact consequence of replacing conditional independence by a uniform-table law. A live library result is not a performed expert comparison, experiment-design validation or source-FPRAS implementation.
The scheduling adapter checks explicit unit-job models and supplied schedules, computes tiny optima and compares existing CP-SAT proposals. Its 3,346 finite and 40 baseline controls expose feasible/optimal/source-machine distinctions. OR-Tools already reaches the generated optima, so any business value must arise from useful model/evidence review rather than a claimed solver improvement.
The bounded-flow module preserves signed bounds/private arc identities and tiny exact counts, while a public planning model exceeds its counting domain. Costs, commodities, unit conventions, circulations and actual scenario probabilities require separate decisions. Data, scheduling, flow, selection and allocation adapters overlap within one platform, not separate customers.
The exact-selection module preserves original IDs, exact tiny multiplicities, checked positive proposals and unknown negative/count states. 2,294 reference/guard/error and 150 existing-solver controls passed. Literal source guards/cutoffs defer direct backends; established vendor matching/manual-review capabilities mean commercial value needs an actual unresolved review decision, not a new reconciliation-category pitch. The error planner's independence/base-error prerequisites remain unvalidated.
Expert experiment and earning decision
Use independent reviewers or counterbalanced assignment for the frozen ten inputs. Record preparation time, independently adjudicated material gaps, false alarms, missed gaps, correction time and actual decision changes. Repeat-reading alone can improve the second attempt; do not attribute that to the report. Retain real identity and timing before claiming an expert result. Synthetic known-gap controls do not estimate general recall or market demand.
An illustrative AUD 10,000 integration fee with assumed specialist labor AUD 180/hour leaves AUD 3,700 before overhead at 35 hours, while 56 hours cost AUD 10,080. These are price/delivery hypotheses, not a quote or profitability result. A highly bespoke review may support consulting; recurring subscription value needs repeated decisions and measured support cost.
Remaining independent work
Review further companion manuscripts/imports, source practical constants, implementation extraction and realistic numerical workflows; preserve unknown budget states and historical byte archives. Precise selected-section records currently cover 28 of 719 manuscripts; complete proof/import/program acceptance remains open. The semantic queue records 65 declared entries across 10 interfaces but does not verify full imports.
No outside expert was contacted and no customer/vendor claim, paid compute, purchase, publication or deployment was performed. Continue local research autonomously; external outreach needs explicit authorization. Current priorities, prototype directory.