Physical implementation and release evidence

The V2 release candidate uses the local TT checkpoint recovery of 447b3dfd0986d473c71afa28efafb3a5cff48da7 as its physical release basis. The repository owner selected this already uploaded layout on 18 October 2026 after the fresh 881495c hosted run reached its six-hour limit. This decision replaces the fresh-route requirement for this release while retaining all physical checks and the original source identities. No silicon/FPGA demonstration or competition submission is claimed.

The public recovery archive contains the layout and original evidence. Its SHA-256 is 2d742a7604d573617cf5f0471e435858afdd9729505c9c9ebb275c075942b466. The successful CI reporting run records that local evidence; it did not rerun physical verification.

Fixed configuration

The candidate remains 6×4 tiles at 40 ns clock period, with maximum transition 1.5 ns, clock uncertainty 0.25 ns, and 32 capture records. The anticipated 8×4 template is deferred; this plan includes no migration or re-hardening for it. The stock TT procedure binds original/prepared inputs and exact action, support tools, LibreLane, container and PDK versions to the results. Its repair-provenance table distinguishes local fresh routing, diode power reconnection and guarded fanout repair from operations performed by stock TT.

Evidence boundaries

Evidence	Source and meaning
Frozen baseline/migration	3d5210adf0edc53b635738850861f639996f2d4a: 583 pytest cases, 24 RTL scenarios, five proofs and six mutations; no deprecated cases removed
Published functional candidate	447b3dfd0986d473c71afa28efafb3a5cff48da7: 631 pytest cases and complete pre-reset-audit gate, locally and in hosted CI
Reset-audit checkpoint	c1c9be573005a022346c2eca6d425f17f815a69e: 667 pytest cases, all original RTL scenarios, both RTL reset scenarios, seven initialization elaborations, five induction proofs and six detected mutations
Software/package candidate	881495cb4d5a6270f5dd7625b1787d83f3059827: 699 pytest cases and the complete functional gate passed locally and in hosted CI; strict Sphinx HTML and all 14 executable examples passed, including installed-wheel checks
Original hosted timeout	Run 38155699349, source 447b3df; the local recovery retains this parent identity
Final hosted timeout	Run 39087676085, source 881495c: functional gate passed; GDS exceeded six hours, with gate-level/precheck/acceptance jobs skipped. Retained logs reached routing iteration 7 with 1,405 intermediate violations; this is not a completed failing physical result or a physical pass
Manual TT run	Run 38115576757, source 3d5210a, produced the V1 top; it does not qualify V2
Historical local acceptance	af573c5fef6f96b4c2fbd400e6565b3b41f54aef; immutable evidence index retains its original source identity
Selected physical release	Physical source 447b3df: recovery completed with zero required physical violations; nine stock prechecks, all 14 V2 gate-level pin scenarios, independent mapping and full-geometry checks passed. Supplemental reset audit from c1c9be5 also passed on that netlist. See recovery scope and provenance.

Layout and software correspondence

The physical source remains 447b3df; the package and functional-test source is 881495c. On 18 October, both versions of all five V2 RTL inputs were elaborated with the same Yosys executable (read_verilog, hierarchy -check, proc, opt_clean). Their complete JSON designs are identical after removing only src location attributes. Ports, cells, parameters, connections, memories, net names and all other attributes are retained in that comparison. Four RTL files are byte-identical; v2_project.v differs only in comments.

The effective configuration (root config plus V2 overrides) and the V2 TT metadata also match. This establishes the source correspondence for the selected layout; it does not pretend that the newer Git revision produced the old GDS. The machine-readable selection records the hashes and comparison method. Historical manifests retain their original release_acceptance: false reporting scope. The existing same-source acceptance checker is unchanged; this explicitly recorded release selection does not constitute a successful execution of that checker for 881495c.

The selected GDS SHA-256 is 902c78fce4a69fde56dbf209b34d5c809e2d44610e6b1e790f03963bb97322c6; the netlist SHA-256 is d99cc8992fe8fc71a5a7dc5c0fd40f162a5f84a1e2fc9cf7a1166c069bda1a84. The current physical checker was rerun successfully against those retained results and exact required test identities. Minimum extracted setup slack is 18.456291 ns; minimum hold slack is 0.054473 ns across the three corners.

Executable migration measurements preserve the same production executed lines and branch arcs: 2948/4745 lines and 1197/1912 branches at the Step 1 checkpoint. The frozen migration record accounts for every baseline pytest case and cocotb scenario. Those measurements are not a claim of complete RTL or instruction-path coverage.

Release assembly and publication checklist

• Retain the original recovered physical source, prepared inputs, checkpoint, pinned tools/PDK, GDS/netlist, full geometry and every required passing check.

• Retain both hosted timeouts and the failed parent route; the successful recovery and later selection supplement those records.

• Bind the 2.0.0rc1 software archives and their complete functional evidence to 881495c, with the exact source correspondence recorded above.

• Run the current full V2 gate-level pin suite and four-state reset/state-loss audit against the selected netlist, recording the test revision separately.

• Review the release-selection documentation, pass the complete functional gate and strict HTML build after those edits, then authorize its commit and gated push.

• Assemble source/package archives, updated Sphinx HTML, example programs/traces, checksums, release notes and an index naming both physical and software sources.

• Obtain separate authorization for merging, tagging, release publication and competition submission. Verify the recipient’s acceptance of the documented local checkpoint procedure before claiming competition acceptance.

No further route is required by the owner’s release selection. Future designs or physical-setting changes still require corresponding physical qualification; the selected evidence cannot be transferred to a changed implementation.