Bounded proof research evidence
Research and proposed PLAN only. No bounded-proof production API, implementation review or whole Kadek foundation-row closure is claimed.
Radio measurements use a96d86e28d384cdf656aadc4282c78c0c170c102 and Git Hemera d5f0a09539360934fb32db9b2e15a91995607db6. Ordinary fetch/fast-forward integrated traversal PR39, b933829cbe273d2f31fe75ca83125f3ddd8917a2, for the future proof implementation. merged-bridge.json compares 39 owning source/lock/consumer inputs: 38 unchanged; lib.rs adds traversal module wiring. The merged traversal source equals the retained unmerged candidate. Its accepted CODE review is copied from the actual committed audit, with invocation/completion; its gates are not relabeled as bounded-proof gates and hosted CI was not claimed green.
Seven research command receipts retain raw output, argv, environment, exit status and timings: initial lock/metadata/debug, legacy release, primary lock/debug/release. Two further receipts retain fetch and fast-forward. The initial standalone std probe characterizes old reader semantics. The separate primary-probe writes manual carriers from explicit small kernel expressions. Both Rust 1.95 debug/release pairs have identical stdout; generated fixtures agree byte-for-byte across profiles. The fixture manifest has 13 files / 54,126 bytes: eight real-file carriers and five explicitly synthetic parent-encoding carriers. The latter do not witness a complete file or particle authority. Work counts in the primary output are PLAN expectations, not executed new-API measurements. No timing, allocation or MSRV claim is inferred.
The initial 55 Git input records and both probe freezes are unchanged. Additional support records pin contracts/callers, committed Kadek anchor and retained soft3 registry. Thirteen exact primary excerpts are replayed from their committed source. Radio is declared part_of soft3; Kadek part_of Cyb. historical-inventory-reference.json names the earlier cached 52-repository inventory and its original archive/member hashes, including the later retained-Nu clarification. This archive does not embed or replay that entire historical capsule and does not claim a fresh network inventory.
Replay from a checkout holding the exact Radio a96d/b933, Hemera d5, Kadek4d9 and soft3 18fd Git objects, with sibling repositories available:
Without --live, current working tree/HEAD need not equal the research task; committed objects remain required. The verifier checks all local file hashes, committed blob identities, exact excerpts, the bridge, raw receipt hashes, fixture hashes and frozen task copies. It does not rerun Cargo or cryptographically recompute the fixtures. To rerun probes, use the captured argv from the Radio worktree and a writable new fixture output directory. Cargo requires the exact lock dependencies in its cache or an explicitly recorded acquisition. These are repository/Git-object replay requirements, not an archive-only or cache-free reproducibility claim.
freeze-research.rb records the original assembly procedure, with original private paths. verify.nu is the portable replay entry point. Initial source-freeze helper text/binary mismatch was corrected before these records; no production discrepancy is attributed to that helper failure. No Python tooling dependency is introduced.