Research-Stack/6-Documentation/docs/stack_fail_closure_register_2026-05-09.md
2026-05-11 22:08:10 -05:00

7 KiB

Stack Fail Closure Register

Date: 2026-05-09

This register turns current failures into closure gates. It does not mark them solved.

Summary

  • Tickets: 7
  • Promotion decision: NO_PROMOTION
  • FPGA software status: PASS
  • FPGA hardware status: NOT_REQUESTED

Tickets

FAIL-FPGA-UART-001 Live fabric UART transport has no observable bytes

  • Class: fpga_transport_or_witness_debt
  • Status: BLOCKED
  • Owner surface: 6-Documentation/docs/fpga_uart_route_analysis_2026-05-09.md
  • Evidence: Q16 software witness passes
  • Evidence: Q16 hardware witness returns short receipt frames
  • Evidence: TX-only beacon standard pins produced zero bytes on ttyUSB0 and ttyUSB1
  • Evidence: TX-only beacon swapped pins produced zero bytes on ttyUSB0 and ttyUSB1
  • Evidence: Old faXX/ffXX direct-probe interpretation is superseded as bridge/MPSSE behavior, not fabric proof
  • Evidence: FPGA UART route analysis identifies the onboard BL702 bridge route as blocked and recommends external USB-UART
  • Evidence: Forced JTAG reset plus SRAM reload succeeds, but beacon/Q16 UART receipts remain empty
  • Evidence: Loopback-after-JTAG-clear diagnostic produced faXX-style bytes on ttyUSB0, consistent with bridge/MPSSE behavior rather than a valid fabric receipt
  • Evidence: PTY-backed virtual serial Q16 probe: PASS_VIRTUAL_SERIAL (3/3 matches)
  • Evidence: UART transport router active route: virtual://q16-pty (PASS_ACTIVE_VIRTUAL_ROUTE)
  • Closure gate: external USB-UART or verified onboard bridge captures beacon payload a6425131360a
  • Closure gate: loopback or beacon receipt passes before Q16 accelerator retry
  • Closure gate: Q16 hardware receipts match software receipts for shift, weighted, and monotone cases
  • Next action: Attach external USB-UART: adapter TX to fabric RX pin 18, adapter RX to fabric TX pin 17, and GND to GND
  • Next action: Probe the new adapter path, usually /dev/ttyUSB2 or /dev/ttyACM0, with the TX-only beacon before Q16
  • Next action: If external UART works, patch host default port or call scripts with --port and rerun Q16 hardware receipts
  • Next action: If external UART fails, inspect PNR pin placement and add LED-observed heartbeat fallback

FAIL-FPGA-FLASH-002 Durable flash programming readback fails

  • Class: fpga_transport_or_witness_debt
  • Status: HELD_SRAM_ONLY
  • Owner surface: 6-Documentation/docs/fpga_rrc_q16_accel_setup_2026-05-09.md
  • Evidence: SRAM load passes CRC
  • Evidence: Flash programming attempt had readback CRC failure
  • Closure gate: flash write and readback CRC pass for the exact Q16 bitstream
  • Closure gate: or documentation keeps SRAM-only boundary explicit
  • Next action: Keep SRAM-only claim boundary until flash command and board target are verified
  • Next action: Do not mark hardware install persistent

FAIL-SECURITY-BEAVER-003 Adaptive Beaver coefficients are not privacy-equivalent masks yet

  • Class: security_proof_debt
  • Status: HOLD
  • Owner surface: Network-Topology-Theory.md + Fundamental_Network_Topology_Equation.md
  • Evidence: 6 security proof debt surfaces found
  • Evidence: tri-cycle audit blocks promotion for adaptive mask claims
  • Evidence: mask freshness negative controls: PASS_NEGATIVE_CONTROLS
  • Evidence: mask freshness case count: 6
  • Closure gate: formal independence/freshness theorem exists
  • Closure gate: secret-sharing non-reuse receipt exists
  • Closure gate: negative control shows adapted coefficients do not leak party inputs
  • Next action: Add a Lean-facing finite-state mask freshness model
  • Next action: Generate negative-control fixtures for repeated coefficients and topology-derived coefficients

FAIL-COEFFICIENT-CALIBRATION-004 Numeric weights remain receipt-weighted priors, not calibrated coefficients

  • Class: coefficient_or_calibration_debt
  • Status: HOLD
  • Owner surface: shared-data/network_topology_database.json
  • Evidence: 20 coefficient/calibration debt surfaces found
  • Evidence: receipt reweighting exists, but coefficient calibration and negative controls remain open
  • Evidence: coefficient HOLD manifest rows: 9
  • Closure gate: dataset provenance receipt linked
  • Closure gate: coefficient calibration receipt linked
  • Closure gate: sensitivity sweep and negative controls pass
  • Next action: Create coefficient calibration fixture manifest
  • Next action: Separate hypothesis weights from calibrated weights in docs and JSON surfaces

FAIL-TOPOLOGY-PREDICTION-005 Topology predictions are not validation claims

  • Class: topology_prediction_debt
  • Status: HOLD
  • Owner surface: shared-data/network_topology_database.json
  • Evidence: 13 topology prediction debt surfaces found
  • Evidence: tri-cycle audit requires pre-registered target and independent comparison
  • Evidence: prediction HOLD registry rows: 15
  • Closure gate: pre-registered prediction target exists
  • Closure gate: outcome receipt exists
  • Closure gate: independent public map or measurement comparison exists
  • Next action: Create prediction-target registry with timestamps and immutable receipt hashes
  • Next action: Move existing predicted nodes into HOLD prediction queue, not validation table

FAIL-RECEIPT-GATE-006 Route promotion lacks complete receipt and rollback closure

  • Class: receipt_gate_debt
  • Status: HOLD
  • Owner surface: 4-Infrastructure/shim/rainbow_raccoon_compiler.py
  • Evidence: 11 receipt-gate debt surfaces found
  • Evidence: compiler keeps 6 objects HOLD and only 1 candidate
  • Evidence: compiler receipt hash: c006c48939fd12a280642e4fb70841fe502641d4c80233bef00b3c710e3f31ba
  • Evidence: optional sem entity probe: OPTIONAL_AUDIT_AID_READY
  • Evidence: HOLD closure checklist: 0 open documentation closures after RRC Gatekeeper refresh
  • Closure gate: validation receipt exists
  • Closure gate: rollback hash exists
  • Closure gate: exact replay or decode closure hash matches
  • Next action: Rerun compiler promotion against the refreshed rrc_hold_closure_checklist.json
  • Next action: Do not widen candidate admission beyond Q16 until Lean or independent replay closes

FAIL-WORKTREE-SCOPE-007 Broad worktree is too dirty for safe sweep commit

  • Class: release_hygiene_debt
  • Status: BLOCKED_FOR_BROAD_STAGE
  • Owner surface: 6-Documentation/docs/stack_solidification_staging_manifest_2026-05-09.md
  • Evidence: worktree status: DIRTY
  • Evidence: changed/untracked count: 949
  • Evidence: staging manifest created for the solidification slice
  • Evidence: optional sem entity probe: OPTIONAL_AUDIT_AID_READY
  • Closure gate: commit scope is explicit file list
  • Closure gate: generated artifacts are intentionally included or excluded
  • Closure gate: no unrelated modified Lean/doc/probe files are swept in
  • Closure gate: entity-level change list exists for staged or scoped files when sem is available
  • Next action: Use stack_solidification_staging_manifest_2026-05-09.md before any commit
  • Next action: Create separate manifests for CPU/logogram/wiki maturation slices if needed
  • Next action: Use /tmp/sem_probe/sem/crates/target/release/sem, not /usr/bin/sem, unless a durable sem binary is installed

Machine Receipt

  • shared-data/data/stack_solidification/stack_fail_closure_register.json