You're pledging to donate if the project hits its minimum goal and gets approved. If not, your funds will be returned.
$2,200 left to reach the funding goal.
I am seeking $2,200 to add two offline tools to EvidenceDesk: a read-only comparison of saved evidence packets and a reusable, versioned verification receipt. The project would also produce synthetic examples, instructions and an independent reproduction report.
Saved research can change in several ways: a source may be replaced, a reviewer may append a comment, or a readable report may be rendered differently. Those changes have different implications, but can be difficult to distinguish by opening two archives side by side. A record that a packet passed verification also needs enough detail to identify exactly which packet and verifier were checked.
The proposed tools would help researchers and reviewers inspect those changes and retain a more precise audit record. They would be designed for local use without sending the packets to an online service. The intended public benefit is reusable infrastructure for inspecting research evidence; whether it improves reviewers’ day-to-day work remains to be evaluated.
The work builds on EvidenceDesk, an existing public prototype for evidence review. Retained private development records describe further offline export and verification work. That completed baseline, prior planning and separate application/model integration work are excluded from this request. The comparator and reusable receipt below are proposed future additions.
1. Compare two saved packets without modifying them. The comparator would produce a readable report and structured JSON, separating changes to archive packaging, file bytes, parsed data, source evidence, reviews and rendered reports. For example, an archive timestamp change should be distinguishable from a changed source passage. A review-only append would require a longer review ledger with the original ledger preserved exactly at its beginning and all original evidence unchanged. Reused local identifiers would not be treated as proof that two items mean the same thing. Invalid or unsupported inputs would yield a clear failure rather than a verified comparison; differences caused by a historical renderer would remain visible without automatically invalidating a packet.
2. Record exactly what was verified. A versioned receipt would identify the packet hash, actual verifier-source identity, Python version, check outcomes and known limits. The hash and verification would use the same acquired bytes. Changing the verifier source would change its recorded identity even if its package version stayed the same. Missing source identity would leave the receipt incomplete, and a failed check would not turn later, unperformed checks into passes.
3. Make the additions reproducible. The work would include synthetic input pairs, expected results, exact commands, exit results and before/after input hashes, plus a short guide. The new commands would be exercised on existing Python 3.12 and a newer interpreter on macOS. The planned output is reusable code, fixtures, documentation and a candid report, with a compatible public licence and required notices determined before release. This proposal does not assert clearance for every private or third-party component.
4. Recruit an independent reviewer. I would seek a reviewer who did not implement the changes and has relevant Python and reproducibility experience. The budget reserves $360 for an estimated six hours to rerun the new cases and report discrepancies, environment and limits. No reviewer has been selected or committed. Recruitment would begin early in the project; reproduction would follow a reviewable build.
The estimated sequence is four part-time weeks: establish the implementation baseline, begin the comparator and recruit a reviewer in week one; develop the comparator and receipt in week two; complete the new cases and guide in week three; and conduct independent reproduction, bounded fixes and closeout in week four. This is a planning estimate, not a confirmed calendar commitment. Reviewer scheduling could extend elapsed time.
Success would mean a reviewer can reproduce the documented distinctions on the synthetic examples, inspect a receipt tied to the actual analyzed bytes and verifier source, and see failures and limitations without guessing what was checked.
The retained acceptance plan contains 17 future cases: 11 comparator cases, four receipt cases, one output-behavior case and one independent-reproduction case. Together they cover unchanged packets; packaging-only changes; review append versus rewrite; changed evidence; reused identifiers; altered rendering; equivalent data with different serialization; invalid or unsupported packets; receipt identity and first-error handling; and safe output behavior. These are planned checks, not completed results of this proposal.
The closeout would include the actual case outcomes, commands and environments, input-preservation hashes, a versioned output inventory, and the reviewer’s findings with any unresolved issues. The new commands should preserve existing protections against overwriting outputs and accurately report output failures. A missing independent review or unresolved acceptance failure would be reported as incomplete work.
These checks would establish bounded software behavior. They would not establish that a source is true, an interpretation is sound, an author is authentic, or a receipt proves that code was executed. No claim of broader adoption or measured improvement in research quality is made.
Proposed work | Estimated effort and rate | Amount
Read-only packet comparator | 18 author hours × $40 | $720
Versioned verification receipt | 16 author hours × $40 | $640
New cases, guide, review coordination and closeout | 12 author hours × $40 | $480
Independent reproduction | 6 reviewer hours × $60 | $360
Total | 46 author hours + 6 reviewer hours | $2,200
The 12-hour author allocation comprises six hours for new checks, two for instructions, two for review response/fixes, and two for reviewer coordination and the final inventory. Rates and hours are budget estimates, not agreed fees or incurred costs. The four-week estimate implies about 11.5 author hours per week.
The budget assumes suitable existing equipment and runtimes. It includes no equipment purchase, paid model inference, installation expense, travel or overhead. There is no separate contingency beyond the two-hour fix allowance. Minimum and goal are both $2,200 because this request covers the complete scope, including independent reproduction; no smaller automatic project is proposed. Completed development and planning would not be charged to this grant.
I am Zackary Loevseth. My public portfolio documents my role designing and directing EvidenceDesk, a public prototype for preserving source snapshots, checking exact quotations and exporting attributed review packets.
My public record also includes a credited security finding, CVE-2026-96573, and a contribution to generation-stream error propagation and failed-read cancellation included in KeeperHub v3.2.0. The KeeperHub contribution built on Yoh5’s original diagnosis and accepted plan. These are relevant examples of practical work; they do not establish the proposed tools’ effectiveness in advance.
I would lead implementation and documentation and recruit the independent reviewer described above. This proposal was prepared with AI assistance.
The principal risks are an unclear compatibility boundary between packet versions, more implementation or review work than the estimate allows, difficulty recruiting a reviewer within the allocation, and limited practical usefulness despite passing synthetic cases. The intended scope is deliberately narrow: offline comparison and receipts for supported packets, with no legacy migration, packet-schema redesign, broad cross-platform support or full application integration promised.
The implementation baseline and publication permissions would be checked before releasing materials. If a needed component could not be used or released, that constraint would be reported rather than treating the private baseline as blanket permission. Reviewer recruitment, time estimates and any material funding overlap would also be revisited before committing the project to delivery.
If review findings exceeded the fix allowance, I would record the remaining defects and seek any required agreement on a changed scope or schedule. I would not replace independent reproduction with self-review or describe unfinished deliverables as complete. Any material change or unused grant funds would be handled under the actual grant agreement.
For this project, I have received no grant funding in the past 12 months and have no overlapping funding. I am requesting $2,200 for the future work described above.