DRACO7RT prototype · updated 23 October 2026

DRACO7RT Full Mission Test

Launch · Deployment · Autonomous Momentum Dump · Station-Keeping · GPS Spoof Response · IMU Fusion

cFS Draco 7.0 on RTEMS 6 / draco-s706 (MPC8548, QEMU), booted through the real u-boot/TFTP chain, fed by NASA 42 dynamics, GPS and IMU samples, commanded from COSMOS, verified by closed evidence runs in which every reported number is checked against a hashed file.

Evidence basis: run run_20261023_201111 · it covers all 75 requirements that existed on 23 October; requirements.csv now holds 83, all traced to code (TRACE.md), the 8 added since (5 CFDP, 3 table rules) are verified by the CFDP run or pending · results slides regenerate with gen_results.py

Orientation

The words in this deck, for a first-time reader

The vehicle side

  • cFS: NASA's core Flight System, the flight software framework; apps are its building blocks (ACS attitude control, NAV, IMU, and NASA's LC, SC, HS health-and-safety, CS checksum)
  • RTEMS: the real-time operating system. e500: the PowerPC processor. u-boot, TFTP, NFS: how the board boots and reads files
  • QEMU draco-s706: a software model of the computer board, so no hardware is needed
  • LC / SC / RTS: Limit Checker watches telemetry (a watchpoint, WP, is a limit; an actionpoint, AP, is the response); Stored Command runs pre-written command lists (relative-time sequences). Together: the onboard autonomy
  • MEKF: the attitude filter that blends gyro and star-tracker data

The ground and sim side

  • COSMOS (OpenC3): the ground system in a browser: telemetry screens, a command sender, and Script Runner, which runs the mission test. Reached through the front door https://draco.efsi.com/ (tile Mission Control, or Open COSMOS). Yamcs is a second ground station on the same downlink
  • NASA 42: a physics simulator that supplies the true attitude, orbit, GPS fixes and gyro rates. The bridge packs them into commands the vehicle accepts
  • Closed evidence run: a script that boots the vehicle from cold, runs every check against it, and writes a record with the exact console lines it read
  • dRT-XXX-nnn: a numbered requirement. A tag is that number written at the line of code that implements it, so a tool can trace both ways

The AI assistant

  • What it is: a program (Claude Code) that reads the repository, runs commands and tests, writes code and notes, and reports back in a text conversation with a person
  • What it did: most of the new code, the test script, the evidence tooling, the notes and this deck, under a person's direction and review
  • What it is not: nothing on the vehicle or in the ground system is AI. The flight software is ordinary C, verified the ordinary way
  • How to read its claims: every number in this deck is read from a file in the repository, never typed from memory. The file paths are on the slides

If a term is not here, the COSMOS Operator Guide deck walks the screens stage by stage, and the Latest for SMA deck covers the standards and the AI rules. All decks: front door tile Presentations, or https://draco.efsi.com/slides/ (open, no sign-in).

Overview

What this test demonstrates

  1. A full early-orbit mission timeline executed by the flight software. The ground only initiates.
  2. Real flight mechanisms under test: SC relative-time sequences with checksum-verified stored commands, LC watchpoint and actionpoint autonomy, SCH scheduling, EVS event reporting, CI and TO ground link.
  3. Deployment discipline: two-step ARM then FIRE pyro channels with microswitch confirmation, including the negative test that FIRE without ARM is rejected.
  4. Propulsion operations: autonomous momentum dump when wheel momentum exceeds the LC limit, plus ground-commanded station-keeping burns with propellant accounting.
  5. New: GPS ingest with a plausibility gate. A forged or replayed fix is rejected, latched, and answered onboard by LC actionpoint 1 starting RTS-006, with no ground in the loop.
  6. New: IMU ingest with a validation gate (validity, time, rate and acceleration limits, stuck axis, dropout). Validated rates feed the attitude filter; a fault or dropout falls back to the internal model within half a second, and a latched fault is answered by LC actionpoint 2 starting RTS-007.
  7. Mission assurance built in: requirements written before the code, traced by source tags, verified by a scripted closed run, reported with artifact-cited provenance.
  8. Operator and developer access: a telnet shell (the UART1 serial shell is off by design), a read-only serial console (front door › Demo control › Serial console tab), a live NFS-backed /cf in the dev config, and the 42 3D views in the browser next to COSMOS (front door tile Live 3D view, https://draco.efsi.com/42/, or Viewer › 3D view).

Environment

System under test

Context diagram: u-boot TFTP boot chain, QEMU draco-s706 board, cFS apps, NFS, COSMOS ground link

What is real

cFS Draco 7.0 flight software, RTEMS 6, PowerPC e500 big-endian, the u-boot boot chain over TFTP, CCSDS command and telemetry, the COSMOS 7.2 ground system.

What is modeled

The draco-s706 board in QEMU 11 with MPC8548 peripherals, built in a 22.04 container to match the vehicle image, interrupt-controller timers at the real 50 MHz; vehicle dynamics, GPS and IMU from NASA 42, bridged to the vehicle big-endian.

Host

Moved from a Windows laptop to the EFSI server in September. Snap Docker sees only the home directory, so the repository is bind-mounted under it; all stack commands run from that path.

The board's unmodified u-boot fetches the cFS uImage over TFTP. The OS swap is invisible to the boot infrastructure.

Conops

Mission timeline: test phases

PhaseWhat happensFSW mechanism
P0 pre-launchBoot to OPERATIONAL; RTS-001 auto-arms autonomy with one group command (RTS bank 2 to 7 enabled, LC ACTIVE)SC power-on RTS
P0b negativeFIRE without ARM: rejected, error countedACS two-step deploy
P1 launchGround: SC START_RTS 5 gives LIFTOFF (front door › Demo control › MISSION › LAUNCH (RTS-005), or the same button on Viewer › Mission › Limits & autonomy); separation at T+10 s with tipoff ratesSC RTS-005 to ACS
P2 deploymentSolar array and antenna: ARM, PYRO FIRE, DEPLOYING, microswitch confirms DEPLOYEDRTS-002 and 003 to ACS
P3 detumbleRates damped into wheels; automatic transition to POINT below 0.005 rad/sACS control mode
P4 momentumDisturbance torque accumulates wheel momentum in POINTACS dynamics
P5 auto dumpMomentum above 4.0 N m s: LC AP0 starts RTS-004, DELTA_H thrusters unload, AP re-armedLC, SC and ACS
P6 station-keepGround: 0.5 m/s burn; self-terminates at target delta-V; fuel decrementedACS maneuver mode
P7 IMU fusionWith 42 and the bridge up: IMU VALID, fault latch clear, ACS rate source IMU and steady for 10 s; skipped with a note when 42 is not runningIMU gate, ACS source select
Any time: GPS spoofImplausible fix rejected and latched; LC AP1 starts RTS-006; next genuine fix acceptedNAV gate, LC and SC
Any time: IMU faultBad sample rejected and latched; LC AP2 starts RTS-007, which pins the filter to the internal model until ground recoveryIMU gate, LC, SC and ACS

Autonomy

Three autonomous responses with no ground in the loop

Autonomy chain: ACS momentum telemetry to LC watchpoint to actionpoint to SC RTS-004 to ACS DELTA_H

Momentum dump

H_MAG (wheel momentum magnitude) above 4.0 trips LC watchpoint 0; actionpoint 0 starts RTS-004; DELTA_H unloads; RTS-004 re-arms the actionpoint so the sawtooth repeats.

GPS spoof response

NAV rejects and latches an implausible fix; LC watchpoint 1 sees the latch, actionpoint 1 starts RTS-006. Recovery by ground command only.

IMU fault response

IMU rejects and latches a bad sample; ACS drops to its internal gyro model within one cycle; LC watchpoint 2 sees the latch, actionpoint 2 starts RTS-007, which pins the filter to INTERNAL and re-asserts the gate. Ground clears the latch, re-arms AP2 and returns the source to AUTO.

Every hop is a real flight mechanism: telemetry-driven LC watchpoints, scheduled AP sampling, checksum-verified stored commands.

FSW

Flight software additions, all requirement-tagged

ComponentContentTraces
cFS/apps/acsModes SAFE, DETUMBLE, POINT, DELTA_H, STATIONKEEP; quaternion propagation; deploy channels; four thrusters; fuel and delta-V accounting; sensor model and MEKF estimator. Simulation-grade dynamics, labeled as suchdRT-ACS-001 to 022
cFS/apps/navGPS ingest on its own message ID; NO_FIX, FIX, STALE, REJECT states; plausibility gate (validity, time, altitude, speed, continuity); latched reject with clear command; rate-limited reject eventsdRT-NAV-001 to 009
cFS/apps/imuIMU sample ingest on its own message ID; NO_DATA, VALID, STALE, FAULT states; gate (validity bits, monotonic time, rate and accel limits, stuck axis); cumulative delta-angle and delta-V; latched fault with clear; rate-limited events. Pure imu_check module at 100% line coveragedRT-IMU-001 to 010
cFS/apps/acs (fusion)Rate source AUTO or INTERNAL: the filter runs on validated IMU rates when fresh, falls back to the internal model on stale or fault with 1 s re-selection hysteresis and bias-covariance re-inflation on every switchdRT-ACS-023, 024
SC RTS bankRTS-001 power-on arming with one group-enable command (five separate enables overflowed the SB pipe limit); RTS-005 launch; RTS-002 and 003 deploys; RTS-004 dump; RTS-006 spoof response; RTS-007 IMU fault responsedRT-ACS-012, NAV-009, IMU-009
LC tablesWP0: ACS H_MAG above 4.0, AP0 two fails start RTS-004. WP1: NAV REJECT_LATCHED, AP1 starts RTS-006. WP2: IMU FAULT_LATCHED, AP2 starts RTS-007dRT-ACS-008, NAV-009, IMU-009
TO_LAB subscriptionsDownlink table: EVS events, all app housekeeping, three ACS, two NAV and two IMU packets, with limits sized for the 10 Hz IMU stream; CI_LAB uplink poll cut from 500 to 50 msdRT-ACS-007, NAV-004, IMU-004
Ground and simCOSMOS fogi plugin with ACS, NAV, IMU, GPS42, IMU42 and SPACECOP targets and pages; 42 bridge packs GPS and IMU samples big-endian and injects fault modes; 42 3D views in the browserdRT-NAV-002, IMU-002

Test

The test script, COSMOS Script Runner

fogi_mission_test.rb phases, read from the script:

  • P0 pre-launch: verify OPERATIONAL, ACS SAFE, stowed
  • P0b negative test: FIRE without ARM must be rejected [dRT-ACS-003]
  • P1 launch: start RTS-005 (RTS-001 armed the autonomy at power-on)
  • P2 deployment: RTS-002/003 arm+fire; microswitches close [dRT-ACS-004]
  • P3 detumble to POINT [dRT-ACS-005]
  • P4/P5 autonomous momentum dump: LC WP0 trips AP0, RTS-004 [dRT-ACS-008]
  • P6 commanded station-keeping burn, 0.5 m/s [dRT-ACS-010]
  • P7 IMU fusion: ACS on the validated IMU stream when 42 and the bridge are up [dRT-ACS-023]
  • P8 GPS spoof response: forged fix rejected and latched, LC AP1 fires RTS-006, ground recovery [dRT-NAV-007, 009]
  • MISSION TEST COMPLETE, all phases passed

Every phase is cmd() plus wait_check() against live telemetry, with two timed holds in P7 and P8. No operator judgment calls.

The same sequence runs headless in run_evidence.py, writing a stamped run record either way.

Negative testing included: the FIRE-without-ARM rejection is a scripted assertion, not a footnote.

To run it: on the front door https://draco.efsi.com/, Demo control panel (bottom), group STACK: Fresh vehicle, then Run in the confirm box (about 70 s, the Job output tab ends “== OK”). Then group MISSION: Mission test (COSMOS), Run. The Job output tab prints one banner per phase and ends “== OK: mission test completed” with the evidence record path. Greyed button: hover it for the reason (usually “flown this boot”).

The latest run from each ground station, COSMOS and Yamcs, phase by phase: next slide, read from the run records.

Script updated 26 September 2026 with P7 and P8. It now asserts on ACS, LC, IMU and NAV; both phases skip with a note when 42 is not running, so the test still passes on a bare vehicle.

Two full mission tests

The same mission, flown once from each ground station

COSMOS: 9/9 PASS, 182 s

fogi_mission_test.rb in Script Runner, started by Mission test (COSMOS) (front door › Demo control › MISSION); commands over CFS_INT

cosmos_20261103_162908 · 3 Nov 2026 16:29 UTC · git d3d4cb2d

Yamcs: 9/9 PASS, 177 s

draco_mission_test.py, the line-for-line port, as a Yamcs script activity, started by Mission test (Yamcs) (front door › Demo control › YAMCS); commands over tc-out

ground_20261103_163655 · 3 Nov 2026 16:36 UTC · git d3d4cb2d

PhaseRequirementsCOSMOSYamcs
P0 pre-launch: verify OPERATIONAL, ACS SAFE, stowed—PASSPASS
P0b negative test: FIRE without ARM must be rejectedACS-003PASSPASS
P1 launch: start RTS-005 (RTS-001 armed the autonomy at power-on)ACS-012PASSPASS
P2 deployment: RTS-002/003 arm+fire; microswitches closeACS-004PASSPASS
P3 detumble -> POINTACS-005PASSPASS
P4/P5 autonomous momentum dump: LC WP0 trips AP0 -> RTS-004ACS-008PASSPASS
P6 commanded station-keeping burn, 0.5 m/sACS-010 ACS-011PASSPASS
P7 IMU fusion: ACS on the validated IMU stream when 42 + bridge are upACS-023PASSPASS
P8 GPS spoof response: forged fix -> NAV gate -> LC WP1/AP1 -> RTS-006, ground recoveryNAV-005 NAV-007 NAV-009PASSPASS

Each from a freshly booted vehicle with 42 running. Read from the run records in cFS/evidence/runs, never typed.

Verification, 1 of 4

Closed evidence run on the new host: platform checks (15)

CheckCoversVerdict, detail
V-BOOT-UIMAGEBOOT-001PASS: u-boot verified + booted the cFS uImage
V-BOOT-CHAINBOOT-004PASS: unmodified u-boot env fetched sp0 over TFTP and bootm'd it
V-EXEC-OPEREXEC-006PASS: cFE reached OPERATIONAL
V-FS-STARTUPFS-003PASS: startup script opened from /cf
V-EXEC-APPSEXEC-003PASS: all 13 static apps initialized
V-NET-TSECNET-002PASS: tsec0 attached via FDT and link UP
V-COMM-CIPORTCOMM-001PASS: CI_LAB listening on 1333 (PID-derived)
V-RMT-NFSMOUNTRMT-004PASS: NFS export mounted at /nfs during startup
CheckCoversVerdict, detail
V-FS-NFSCFFS-004PASS: dev config: /cf backed by live NFS (tarfs shadowed)
V-NET-PINGNET-003PASS: ping 10.5.5.38 from tftp container
V-COMM-UPLINKCOMM-001PASS: ENABLE_TELEMETRY accepted on 1333 (telemetry started => ...
V-COMM-DOWNLINKCOMM-002PASS: 30 TO_LAB packets captured on 2333 (TO_LAB marker present)
V-RMT-TELNETRMT-001PASS: telnet login shell reached (prompt + ls /)
V-RMT-NFSRWRMT-004PASS: NFS /nfs read OK, target->host write OK
V-RMT-SERSHELLRMT-006FAIL: UART1 serial shell accepts commands (ls / over the wire)

run_20261023_201111 · 64/66 PASS · git 4b8f3df7 · sp0 sha256 2d57f4b2aec1813e... · V-RMT-SERSHELL fails by configuration: the serial shell is suppressed by the tracked NOS3 flag file (documented deviation); SKIP: V-ART-ACS-RATE

Verification, 2 of 4

Closed evidence run on the new host: mission checks (14)

CheckCoversVerdict, detail
V-ACS-ARMACS-012PASS: power-on RTS-001 armed the autonomy (RTS bank + LC ACTIVE)
V-ACS-NEGFIREACS-003PASS: FIRE without ARM rejected
V-ACS-LAUNCHACS-012PASS: RTS-005 launch: LIFTOFF + SEPARATION + DETUMBLE
V-ACS-DEPLOYACS-004PASS: both deployments: ARM->FIRE->DEPLOYING->DEPLOYED w/ microswitch
V-ACS-DETUMBLEACS-005PASS: detumble complete -> POINT
V-ACS-DUMPACS-006 to 008PASS: AUTONOMOUS dump: LC AP -> RTS-004 -> DELTA_H -> unloaded
V-ACS-CMDDUMPACS-009PASS: ground-commanded dump enters/exits DELTA_H via the same path
CheckCoversVerdict, detail
V-ACS-SKACS-001, 010, 011PASS: commanded station-keeping burn to target delta-V
V-ACS-FUELWARNACS-015, 016PASS: latched LOW then CRITICAL warnings after SET_FUEL
V-ACS-FUELEXHACS-014PASS: exhaustion: event + SAFE; propulsive commands rejected
V-ACS-UTCOVACS-017 to 020, 022PASS: sensor+MEKF unit tests pass at 100% line coverage
V-ACS-EKFCONVACS-018, 020, 021PASS: live MEKF converged: attitude error 0.043 deg < 0.5
V-ACS-BIASACS-017, 019PASS: live gyro-bias estimate tracks injected truth [+0.0020 -0.0014 ...
V-ACS-APPCOVACS-001 to 018, 020, 021, 023, 024PASS: acs_app.c unit tests pass at 100.00% line coverage (floor 100%)

run_20261023_201111 · 64/66 PASS · git 4b8f3df7 · sp0 sha256 2d57f4b2aec1813e... · V-RMT-SERSHELL fails by configuration: the serial shell is suppressed by the tracked NOS3 flag file (documented deviation); SKIP: V-ART-ACS-RATE

Verification, 3 of 4

Closed evidence run on the new host: GPS, IMU and fusion checks (16)

CheckCoversVerdict, detail
V-IMU-UTCOVIMU-005, 010PASS: imu_check unit tests pass at 100% line coverage
V-NAV-UTCOVNAV-005, 008PASS: nav_gate unit tests pass at 100% line coverage
V-NAV-INGESTNAV-001 to 004PASS: live 42 GPS stream: fix acquired, HK STATE FIX, FIX_COUNT 7->14 over ...
V-NAV-GATENAV-005, 006PASS: gate ON at init; OFF: +500 km teleport ACCEPTED (POS_TLM FIX, flags ...
V-NAV-FAULTNAV-007, 009PASS: teleport REJECTED 0x08 -> LC AP1 PASS->FAIL (LC 58) -> LC 1101 RTS = ...
V-NAV-RATELIMITNAV-008PASS: 41 forged fixes: REJECT_COUNT 1->42, 3 REJECTED event(s) at totals ...
V-NAV-RECOVERNAV-007, 009PASS: ground recovery: CLEAR_REJECT -> REJECT_LATCHED CLEAR, LC 61 AP1 ...
V-IMU-INGESTIMU-001 to 004; ACS-023PASS: live stream: IMU VALID, ACS -> IMU, no STALE and a single source ...
CheckCoversVerdict, detail
V-IMU-FAULTIMU-005, 007 to 009PASS: spike -> REJECTED (rate-limited events) -> LC AP2 (LC 1102) -> ...
V-ACS-FAULTFBACS-024PASS: ACS fell back to INTERNAL (fault/stale) on the rejected stream
V-IMU-GATEIMU-006PASS: gate ON at init; OFF: injected |w|=86.603 rad/s spike ACCEPTED (IMU ...
V-IMU-RECOVERIMU-007, 009; ACS-023PASS: ground recovery: CLEAR_FAULT, LC AP2 re-armed, AUTO -> MEKF back on ...
V-ACS-FALLBACKACS-024PASS: stream stopped: IMU STALE within 0.5 s, ACS -> INTERNAL (stale)
V-NAV-STALENAV-003PASS: stream stopped: NAV STALE event within NAV_STALE_S, HK STATE STALE, ...
V-NAV-APPCOVNAV-001 to 008PASS: nav_app.c unit tests pass at 100.00% line coverage (floor 100%)
V-IMU-APPCOVACS-024; IMU-001 to 009PASS: imu_app.c unit tests pass at 100.00% line coverage (floor 100%)

run_20261023_201111 · 64/66 PASS · git 4b8f3df7 · sp0 sha256 2d57f4b2aec1813e... · V-RMT-SERSHELL fails by configuration: the serial shell is suppressed by the tracked NOS3 flag file (documented deviation); SKIP: V-ART-ACS-RATE

Verification, 4 of 4

Static artifact checks, no boot needed (21)

CheckCoversVerdict, detail
V-ART-BOOT-LOADADDRBOOT-002PASS: uImage load 0x4000 == entry 0x4000 == ELF ...
V-ART-BOOT-STARTGUARDBOOT-003PASS: deployed payload begins with 0x7c711b78; ...
V-ART-BOOT-FDTEMBEDBOOT-005PASS: ELF: bsp_fdt_get/bsp_fdt_copy are PSP functions ...
V-ART-BOOT-FDTCLOCKSBOOT-006PASS: embedded DTB (12736 B, valid FDT) cpu node ...
V-ART-BOOT-CCSRBOOT-007PASS: ELF absolute symbol qoriq = 0xe0000000 (CCSR ...
V-ART-BOOT-UARTPOLLBOOT-008PASS: console_device_table in the ELF binds ...
V-ART-EXEC-ABIEXEC-001PASS: ELF32 big-endian PowerPC; 'rtems-6.0.0 ...
V-ART-EXEC-STATICEXEC-002PASS: no SHT_DYNAMIC section; RTEMS_DYNAMIC_LOAD ...
V-ART-EXEC-OSLIMITSEXEC-004PASS: OSAL limits tasks/queues/modules/timers: ...
V-ART-EXEC-PROCIDEXEC-005PASS: targets.cmake cpu1_PROCESSORID 100; as-built ...
V-ART-FS-TARFSFS-001PASS: RTEMS_INCLUDE_TARFS; OSAL bsp_tarfs_setupfs.c ...
CheckCoversVerdict, detail
V-ART-FS-TARGENFS-002PASS: make_rtems_tarfs.sh stages ...
V-ART-NET-LIBBSDNET-001PASS: rtems-libbsd selected (OSAL_RTEMS_NETWORK_LIB ...
V-ART-NET-LO0ORDERNET-004PASS: network_ppce500_rtems_Init calls ...
V-ART-NET-FAILSAFENET-005PASS: all 2 failure branches of ...
V-ART-COMM-CFSINTCOMM-003PASS: COSMOS CFS_INT = udp 10.5.5.38 write 1333 read ...
V-ART-RMT-SHELLSETRMT-002PASS: as-built rtems_shell_Initial_commands = 23 ...
V-ART-RMT-NOAUTHRMT-003PASS: telnetd config and per-session shell env set ...
V-ART-RMT-NFSREGRMT-005PASS: draco_mount_nfs() (line 166) registers ...
V-ART-ACS-RATEACS-002SKIP: design cadence is a 100 ms SB pend timeout ...
V-ART-ACS-THRTLMACS-013PASS: ACS_ActTlm_Payload_t carries ThrFireCount[4] ...

run_20261023_201111 · 64/66 PASS · git 4b8f3df7 · sp0 sha256 2d57f4b2aec1813e... · V-RMT-SERSHELL fails by configuration: the serial shell is suppressed by the tracked NOS3 flag file (documented deviation); SKIP: V-ART-ACS-RATE

The mission test, extended to an hour

Planned in PlanDev, flown from COSMOS, compared afterwards

What ran

  • Plan 11 “DRACO launch day”: the mission test’s phases in its first ten minutes, then 45 minutes of pointing with 34 scheduled momentum dumps
  • Turned into a COSMOS script that sends each command at its planned second and makes the mission test’s own checks
  • 60 minutes on a freshly booted vehicle, 22 October 2026
ResultValue
Checks155 / 155 PASS
Command timingall within 0.1 s of plan
Momentum dumps41 planned, 41 flown
Detumble20.2 s planned, 19.9 s flown
Fuel at one hour9.445 kg planned, 9.442 flown

Not part of the closed evidence run and not tied to requirements. Record: claude-notes/PLAN-VS-ACTUAL-2026-10-22.md; tools in tools/plandev; the Ground System deck explains it. See the plan: front door, Ground stations panel, PlanDev card, Open Plan (https://plan.draco.efsi.com, new tab), or Plan ↗ in the viewer’s top bar.

SQA

Mission assurance with the new software

Requirements first

The 24 ACS, 9 NAV and 10 IMU requirements were written before the first line of their code; the IMU set came from a frozen interface contract; six assistant conversations, each driven and reviewed by a person, implemented it in parallel. The earlier port groups were reverse-derived.

Traceability

83 of 83 requirements carry [dRT-...] tags at the implementing line (75 of 75 when this slide was written, 23 October). gen_trace.py regenerates matrices and diagrams and fails the build on any code with no requirement.

Closed verification

run_evidence.py boots the real boot chain and executes every check against the live target. Results are stamped run records with captured console, telnet and packet logs.

Hard provenance and static analysis

gen_report.py builds the Design and Verification Report where every number cites an on-disk artifact in a SHA-256 manifest. cppcheck runs over the core and the mission apps with a dispositioned deviation log, and McCabe complexity is recorded per function.

This deck follows the same rule: phases from the test script, results from the run record, counts from requirements.csv and the regenerated trace.

NASA standards

How this evidence maps to the NASA standards

Standard, in plain wordsWhat this project does about itStatus
NPR 7150.2D: NASA's software rulebook, with rigor scaled by a class letterClassified Class E (a prototype exploring a design concept). Class E invokes only classification, records, reuse assessment, safety-criticality and assurance planning. The Class D practices on this deck, traceability, unit tests, test against requirements, target validation, static analysis, CM, are applied voluntarilyClass E set recorded 30 Sep 2026; voluntary set 3 met, 4 partial
NASA-STD-8739.8B: software assurance and software safetySafety-critical determination: no. No hazard analysis exists because nothing physical can be harmed; the vehicle is a software model. SWE-134 safety items are therefore not invoked, though several are implemented by designDetermination recorded; SMA concurrence not yet obtained
NASA-STD-7009B: how far to trust a model or simulationCapability self-assessed factor by factor: verification 2 of 4 (closed runs, checksummed baseline), validation 1 (no real-vehicle referent), data pedigree 1. Fidelity gaps are stated on the slides: QEMU board model, simulation-grade dynamics, synthetic IMU noiseSelf-assessment only; no formal record
NPR 7123.1D: verification (built it right) versus validation (built the right thing)Verification: 83 requirements traced to code (TRACE.md); the 75 that existed on 23 October all covered by the 66 closed-run checks, the 5 CFDP ones by the CFDP run, the 3 table rules pending; unit tests for every mission-app file. Validation: the intended use in MISSION_TEST_DESIGN.md is demonstrated end to end by the mission testVerification strong; validation internal only
SWE-146 and handbook topic 7.25: AI-written code is auto-generated codeEvery line the AI assistant wrote is under the same tests, static analysis and human review as hand code; the person who directed it is responsible; the instructions given to it are under configuration controlApplied; the generator itself cannot be validated

Record: cFS/evidence/rtems_reqs/CLASSIFICATION.md. The Latest for SMA deck carries the full scorecard, the SWE-134 item-by-item review and the NASA guidance on AI, quoted.

Findings, August

What the process caught while building the mission test

FindingCaught byClass
NFS mounted at /mnt shadowed the /cf tarfs; ES could not open the startup script, zero apps startedFirst closed evidence run (8 of 13)Flight defect
LC watchpoint offset assumed a 12-byte telemetry header; the header is 16 bytes, so autonomy never firedMission-sequence dry runConfig defect
SC walks RTS entries in 32-bit words with round-up; unpadded 10-byte commands broke table validationBoot events (RTS load reject)Table defect
Upstream RTS sample tables are endian-naive; arguments must be big-endian on this targetCode review during authoringLatent defect
TO_LAB default subscription table downlinks nothing; July's closed loop carried only TO_LAB's own housekeepingCOSMOS page designGap
The RTEMS shell init does not attach the shell to the named device; it inherits the caller's consoleSerial-shell bring-upLatent defect
LC sets a triggered actionpoint passive by design; after the first dump momentum grew unbounded. RTS-004 now re-arms itLong-duration live vehicleFlight defect
COSMOS packet definitions for MM, MD and FM were counter stubs with misaligned fields; rewritten from the flight structsLog Messages size errorsGround defect

Findings, September

What the process caught since the August deck

FindingCaught byClass
SC subscribes to its own command ID with the default pipe limit of four; RTS-001's fifth ENABLE_RTS was silently dropped with no counter moving. Replaced by one group-enable commandOne SB "Msg Limit Err" event in the boot consoleFlight defect
Restarting 42 rewinds simulation time; the NAV gate correctly rejects every fix as a replay until sim time passes the last accepted fix. Documented ground reset procedureLive vehicle after a 42 restartBehavior, documented
The downlink command in the runbook, TO ENABLE_OUTPUT, does not exist; the command is TO ENABLE_TELEMETRY with IP_ADDRBootstrap script on the new hostDocumentation defect
Snap Docker exposes only the home directory; every bind mount from the web root failed. Repository bind-mounted under home; scripts derive paths from their own locationFirst stack start after the moveInfrastructure
The 42 binary copied from the old host linked an older freeglut and crashed in text drawing; rebuilt on the target host, the same rule already applied to the flight softwareDebugger backtraceBuild discipline
The server firewall silently dropped the vehicle's telemetry to the host bridge address, failing the uplink and downlink checks with everything else green. One scoped rule on the internal bridge fixed itClosed evidence run on the new host; 32 firewall log entriesInfrastructure
The serial shell is suppressed by a tracked flag file from the NOS3 integration, yet the app that needs the UART is listed in the startup script without its object in the dev /cf. The serial-shell check now fails by configurationClosed evidence runConfig inconsistency

Findings, IMU integration

What the process caught while adding the IMU, the NAV verifiers and fixing the clock

FindingCaught byClass
CI_LAB services its uplink socket only on a 500 ms receive timeout, at most eleven packets per cycle. A 10 Hz stream backed up 12 s and kept growing. Mission override sets the timeout to 50 ms; path latency now 166 ms medianSequence-matched packet capture on the bridgeFlight config defect
The EVS long-event downlink limit was 32; the boot burst is exactly 32 events, so the 33rd and 34th were dropped every boot with no counter moving. Raised to 48"Msg Limit Err" on every earlier bootPre-existing flight defect
A comment line ending in "+" in the COSMOS plugin file is read as line continuation by the parser, so the next target definition vanished silentlyMissing IMU42 target after install; bisected plugin fileGround tool behavior
The IMU dropped its time reference on every stale entry, disabling replay and stuck detection after any gap. Reference now persists until a ground clearReview by a separate assistant conversation briefed on flight safety, read by a personDesign defect, fixed pre-release
Rate-source selection had no hysteresis; delivery hiccups made the filter thrash between sources. Added 1 s re-selection hold and bias-covariance re-inflation on switchReview by a separate assistant conversation briefed on flight safety, read by a personDesign defect, fixed pre-release
42 exports only true rates and accelerations over its IPC; its sensor noise model never reaches the wire. The bridge now adds gyro noise and announces the fidelity gap at start-upReading the 42 autocode; requirements-honesty reviewFidelity gap, disclosed
Ground recovery after an autonomous spoof response re-fired RTS-006 when the actionpoint was re-armed in the same second as the latch was cleared; LC had not yet sampled the cleared latch. Recovery now waits for the latch to read clearScripted P8 phase and V-NAV-RECOVER during authoringProcedure defect
The vehicle clock ran at 0.494x wall time. RTEMS ticks off the interrupt controller's global timers assuming bus/8 = 50 MHz; upstream QEMU hard-codes them at 25 MHz. Now a board property set to 50 MHz. The last 1% (1 Hz tone every 1.011 s) was a second defect, fixed 23 Oct: each tick re-armed from the late callback; now 1.000 sIMU rate counters vs wall clock, then a 60 s clock measurementBoard-model defect, fixed
A QEMU built on the Ubuntu 24.04 host needs glibc 2.38 and exits 127 inside the 22.04 vehicle container. The emulator is now built in a 22.04 container by run/build-qemu.shFirst boot on the rebuilt emulatorBuild discipline

Runbook

Running the demo: two tracks

Presenter, no programming: the front door

  1. Open https://draco.efsi.com/ and press Sign in with Microsoft (or, under it, type the shared COSMOS password in the box and press Sign in with password). Every button below is in the Demo control panel at the bottom of the page; each opens a confirm box: press Run, then read the live log in the Job output tab
  2. Stack down? Group STACK, Start stack: boots everything, enables telemetry. Check the status strip at the top: Vehicle, Stack, Commanding COSMOS
  3. Open the watch page in a second tab: tile Mission Control (the viewer, FOGI HOME), then Viewer › Home › Overview and Viewer › GNC › ACS
  4. Group MISSION, LAUNCH (RTS-005), and narrate the ACS page; or Mission test (COSMOS) for the scripted run
  5. Between shows: group STACK, Fresh vehicle (about 70 s, back on the pad). At the end: Shut down
  6. 3D view: group 3D VIEW, 42 start (also clears the IMU latch and resets the GPS gate), then tile Live 3D view (https://draco.efsi.com/42/); the filter switches to the IMU within a few seconds
  7. Greyed button: hover it for the reason. Values frozen: group HEALTH, Verify, then Downlink / map

Developer, terminal in /home/efsi/draco7_0/run

  • ./demo.sh: the same presenter steps as a menu (1 start, 2 LAUNCH, 3 status, 4 live events, 5 fresh vehicle, 6 shut down)
  • ./draco.sh rtems up | stop: checked bring-up and stop
  • ./draco.sh rtems fsw, then sync_cf: build the flight software and re-baseline the live /cf (required after every build)
  • ./bootstrap-draco.sh --verify | --per-boot: health check; re-map targets and enable downlink after each reboot
  • docker compose -p draco7 down, then python3 ../cFS/evidence/rtems_reqs/run_evidence.py --keep-up: the 66-check closed evidence run. The stack must be down first or the script never reboots the vehicle
  • gen_trace.py, gen_report.py: regenerate trace and report. gen_deck.py still builds the legacy .pptx by hand, but nothing runs it: the decks are powerpoint/decks (this one: gen_results.py after a closed run)
  • One-time COSMOS setup: install both plugin gems, map ACS, NAV and IMU to CFS_INT, never GPS42 or IMU42. One-time host setup: a ufw rule allowing UDP 2333 in on the fogi_net bridge, or the evidence run cannot capture the downlink

Changed since August: the run directory, the front door with its Demo control (replaces the public COSMOS address as the start), the per-boot bootstrap step, the NAV mapping, and the 42 browser view.

Validation

COSMOS: which screens, and how to trust the data

Screens: Viewer (tile Mission Control), group button, page button

  • Home › Mission home (FOGI HOME): top level, one box per subsystem with live values and Open buttons
  • FOGI pages: Mission › App health (APPS), GNC › Attitude & rates (GNC, live graphs), Mission › Deployments, Propulsion, Limits & autonomy (LIMITS, with the LAUNCH (RTS-005) button), Events, Sequences (ATS/RTS), Home › Overview; walls under Wall: Wall (3 screens), GNC wall, Sensors wall, Mission wall, All-in-one (WALL_ALL also via Select Screen)
  • GNC › ACS (ACS ACS): one-page combined mission view
  • Tools › Packet Viewer, target NAV, packets HK_TLM_PKT and POS_TLM_PKT: STATE, FIX_COUNT, REJECT_LATCHED, altitude, speed (or GNC › NAV (GPS gate))
  • GNC › IMU: stream state, sample rate, fault latch, gyro against ACS truth, the ACS rate source
  • Tools › Command Sender: TO ENABLE_TELEMETRY with IP_ADDR 10.5.5.200 for downlink (front door › Demo control › HEALTH › Downlink / map does it for you); ACS SET_FUEL for the fuel demo (or the fuel buttons on Mission › Propulsion)
  • Stage-by-stage walkthrough: the separate COSMOS Operator Guide deck

Validate the telemetry before presenting

  1. Tools › Cmd/Tlm Server, Interfaces tab: CFS_INT CONNECTED and Tlm Pkts climbing
  2. Same page, Log Messages at the bottom: no "received with actual packet length" errors. A size error means a definition no longer matches the flight structs. Report it
  3. Tools › Packet Viewer: any packet's RECEIVED_COUNT increments about once a second, and RECEIVED_TIMEFORMATTED is now. A first query right after mapping can return stale values
  4. Expected pad values: ACS MODE SAFE, FUEL_KG 12.0, FUEL_STATE OK, DEPLOYED_BITS 0, RATE_SOURCE INTERNAL; NAV NO_FIX and IMU NO_DATA until 42 feeds
  5. Quick check without COSMOS: the status strip at the top of the front door (Vehicle, CFE_ES age a second or two, Stack, Commanding)

All pages null? Target unmapped or downlink off: bootstrap-draco.sh --per-boot fixes both; on the front door it is Demo control › HEALTH › Downlink / map (then Verify).

Requirements backup · ACS-023 and 024 are on the IMU slide

The ACS requirement set: 23 of 25

IDShall (abbreviated)Ver
dRT-ACS-001The ACS application shall provide the control modes SAFE, DETUMBLE, POINT, ...T
dRT-ACS-002The ACS application shall propagate simulation-grade rigid-body attitude ...A
dRT-ACS-003The ACS application shall provide two deployment channels (solar array, ...T
dRT-ACS-004Upon pyro fire, a deployment channel shall transition FIRING to DEPLOYING to ...D
dRT-ACS-005In DETUMBLE mode the ACS shall damp body rates and shall autonomously ...T
dRT-ACS-006In POINT mode the ACS shall accumulate wheel momentum under a modeled ...A
dRT-ACS-007The ACS shall publish housekeeping, attitude, and actuator telemetry packets ...I
dRT-ACS-008When wheel momentum magnitude exceeds the LC watchpoint threshold, the system ...D
dRT-ACS-009The ACS shall accept a ground command to initiate a momentum dump, entering ...T
dRT-ACS-010The ACS shall accept a station-keeping command with a target delta-V ...T
dRT-ACS-011The ACS shall account for propellant: fuel mass shall decrease in proportion ...A
IDShall (abbreviated)Ver
dRT-ACS-012The launch sequence shall execute from SC relative-time sequences: RTS-001 ...D
dRT-ACS-013The ACS shall telemeter per-thruster fire counts and cumulative on-time.I
dRT-ACS-014When propellant is exhausted the thrusters shall be ineffective: no momentum ...T
dRT-ACS-015The ACS shall telemeter a fuel state (OK/LOW/CRITICAL/EXHAUSTED) and shall ...T
dRT-ACS-016The ACS shall provide a SET_FUEL maintenance command (test/sim use) that sets ...T
dRT-ACS-017The ACS shall model attitude sensors: a three-axis rate gyro whose ...T
dRT-ACS-018The ACS shall estimate spacecraft attitude with a Multiplicative Extended ...T
dRT-ACS-019The MEKF shall estimate the three-axis gyro bias as part of its state vector; ...A
dRT-ACS-020The ACS shall telemeter the filter attitude uncertainty (3-sigma, degrees) ...A
dRT-ACS-021After star-tracker acquisition the MEKF attitude estimation error (angle ...T
dRT-ACS-022The ACS attitude-estimation module (sensor model + MEKF) shall be implemented ...T
dRT-ACS-025The ACS's tunable parameters shall be loaded from cFE tables at initialization ... (added 27 Oct)T

Verification method: T test, A analysis, I inspection, D demonstration. Full text in cFS/evidence/rtems_reqs/requirements.csv and verify/requirements.yaml.

Requirements backup

The NAV requirement set, 10 requirements (001 to 009 new in September)

IDShall (abbreviated)Ver
dRT-NAV-001The flight software shall provide a navigation application (NAV) that maintains the onboard navigation state from an externally supplied GPS fix. NAV ...I
dRT-NAV-002NAV shall accept GPS fixes as NAV_GpsFixCmd_t messages on NAV_GPS_MID, carrying receiver validity, GPS rollover/week/seconds, ECI position and ...T
dRT-NAV-003NAV shall report a navigation state of NO_FIX, FIX, STALE or REJECT. The state shall become STALE when no fix has been accepted within 3 seconds, and ...T
dRT-NAV-004NAV shall telemeter the navigation solution (position, velocity, geodesy, radius, speed, fix age and counters) and housekeeping ...D
dRT-NAV-005NAV shall validate every received GPS fix against a plausibility gate before accepting it, rejecting a fix whose receiver validity flag is clear, ...T
dRT-NAV-006The NAV plausibility gate shall be enabled at initialization and shall be commandable off and on from the ground, so that the protected and ...T
dRT-NAV-007NAV shall latch an indication that a fix has been rejected since the last clear, and shall provide a ground command to clear that indication, so that ...T
dRT-NAV-008NAV shall rate-limit rejection events so that a sustained flood of forged fixes cannot exhaust or obscure the event log, while still annunciating the ...T
dRT-NAV-009A rejected GPS fix shall autonomously trigger an onboard response without ground action: Limit Checker watchpoint 1 shall monitor the NAV ...T
dRT-NAV-010NAV's tunable parameters (plausibility limits, stale time, event pacing, housekeeping cadence, gate state at boot) shall be loaded from the cFE table NAV.CFG_TBL ... (added 27 Oct)T

Verification method: T test, A analysis, I inspection, D demonstration. Every NAV requirement is tagged at its implementing line in cFS/apps/nav.

Requirements backup

The IMU and fusion requirement set, 13 requirements (12 new in September)

IDShall (abbreviated)Ver
dRT-IMU-001The flight software shall provide an inertial measurement unit application (IMU) ...I
dRT-IMU-002IMU shall accept samples as IMU_SampleCmd_t messages on IMU_SAMPLE_MID of exactly 80 ...T
dRT-IMU-003IMU shall report a sensor state of NO_DATA, VALID, STALE or FAULT. The state shall ...T
dRT-IMU-004IMU shall telemeter housekeeping (command and error counters, state, fault flags, ...D
dRT-IMU-005IMU shall validate every received sample against a plausibility gate before ...T
dRT-IMU-006The IMU plausibility gate shall be enabled at initialization with default limits of ...T
dRT-IMU-011IMU's tunable parameters (validation limits, stuck count, ...) shall be loaded from the cFE table IMU.CFG_TBL ... (added 27 Oct)T
IDShall (abbreviated)Ver
dRT-IMU-007IMU shall latch an indication that a sample has been rejected since the last clear. ...T
dRT-IMU-008IMU shall rate-limit rejection events so that a sustained stream of faulty samples ...T
dRT-IMU-009A latched IMU fault shall autonomously trigger an onboard response without ground ...D
dRT-IMU-010The IMU sample-validation module (imu_check) shall be implemented as a pure, ...T
dRT-ACS-023The ACS shall select its attitude-rate measurement source under a ground-commandable ...T
dRT-ACS-024When IMU data becomes stale (no VALID data for 0.5 seconds) or faulted (a FAULT data ...T

The September twelve were written before the code from a frozen interface contract (claude-notes/IMU-INTERFACE.md); every one is tagged at its implementing line and covered by a closed-run check or the unit-test harness. IMU-011 (table rule) is pending.

← All presentations