docs: VERIFICATION.md trust-receipt guide (closes #35)#136
Merged
Conversation
Adds docs/verification.md walking users through the auto-generated VERIFICATION.md artifact — the trust receipt proving an exported ONNX model matches its PyTorch source at machine precision. Structure verified against src/reflex/verification_report.py — every section + field name in the doc maps to a real line of the renderer. Adapted from #130: - Kept the fixture numbers (1.192e-07, 2.384e-07, etc.) — they're float32 ULP multiples, characteristic of correctly-exported models, and 2.384e-07 lines up with pi0.5's actual first-action max_abs (2.38e-07) from the README parity ledger. - Refreshed Reflex version (0.2.1 → 0.9.6) + added a note that the field auto-fills from reflex.__version__. - Cross-referenced the parity ledger explicitly so readers can sanity- check their own runs against the shipped numbers (SmolVLA 5.96e-07, pi0 2.09e-07, pi0.5 2.38e-07, GR00T 8.34e-07). - Renamed file from understanding_verification.md to verification.md for consistency with sibling docs (eval.md, embodiment_schema.md). - Replaced broken TROUBLESHOOTING.md cross-ref with reflex doctor flow. - Kept the 'Interpreting results by vertical' table — verticals match the FastCrest customer research vault's P0 tier. - Added 'Files' section preamble (Total: N files, X size) that #130 omitted but the live renderer emits. Supersedes #130. Co-Authored-By: Divyansh Rawat <186957976+DsThakurRawat@users.noreply.github.com> Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This was referenced May 16, 2026
rylinjames
added a commit
that referenced
this pull request
May 16, 2026
#63) (#137) Adds docs/troubleshooting.md with the most common error signatures and fixes Reflex users hit on edge devices, cloud GPUs, ROS2 robots, and drones. Structure: CUDA/GPU → Jetson → ROS2 bridge → Drone/MAVROS → Export/validation → Registry → Quick diagnostics. Adapted from #129 — kept all the original error signatures and fix patterns (they're real and useful). Only refreshed: - onnxruntime-gpu pin: 1.18.0 → >=1.25.1 (matches v0.9.2 floor) - cuDNN floor: 9.0+ → 9.5+ (matches v0.9.2 floor) - Driver floor: R525+ → R555+ for cuDNN 9.5+ (per v0.9.4 doctor guard) - Added Blackwell sm_120 section (RTX 5090 / B200) per v0.9.3 guard - Added the four v0.9.4 reflex doctor guards as the prescribed first step on any failure (multi-GPU arch, Jetson R35, cuDNN/driver skew, TRT EP empirical session test) - Replaced #129's reference to PR #121 with the actual shipped --state-msg-type flag from #133 (joint_state/imu/odom dispatch) - Updated cross-ref from understanding_verification.md → verification.md (renamed in #136) - Updated cross-ref to adding_a_robot.md (shipped in #135) - File renamed TROUBLESHOOTING.md → troubleshooting.md for consistency with sibling docs (lowercase, no ALL_CAPS) Supersedes #129. Co-authored-by: Divyansh Rawat <186957976+DsThakurRawat@users.noreply.github.com> Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Adds
docs/verification.md— plain-English guide to the auto-generatedVERIFICATION.mdtrust receipt. Closes #35.Verified against
src/reflex/verification_report.py— every section + field name in the doc maps to a real line of the renderer.Adapted from #130
@DsThakurRawat opened #130 with the same goal. The teaching structure (When is it created → Section-by-section breakdown → Key metrics → Interpreting by vertical → What to do when it fails) is theirs and good; preserved via
Co-Authored-Byon the commit.What's the same as #130
1.192e-07,2.384e-07,1.788e-07, etc.). Earlier I lumped these in with "fabricated content" — that was wrong. They're float32 ULP multiples characteristic of correctly-exported models, and the aggregate2.384e-07lines up with pi0.5's actual first-actionmax_abs(2.38e-07) from the README parity ledger. Most likely DsThakurRawat ranreflex validateon a real pi0.5 export.mean_abs_diff≈max_abs_diffframing — accurate teaching.What's different
0.2.1→0.9.6reflex.__version__"Total: N files, X sizeunderstanding_verification.md→verification.md<topic>.md(eval.md, embodiment_schema.md, etc.); theunderstanding_prefix was the only outlierTROUBLESHOOTING.mdreflex doctorflow that does existsrc/reflex/verification_report.pyas source-of-truthCloses / supersedes
good first issue)Test plan
_format_parity/write_verification_reportinsrc/reflex/verification_report.pyreflex.__version__(0.9.6)cli_reference.md(✓ from docs: comprehensive CLI command reference (closes #65) #134),eval.md(✓),doctor_check_list.md(✓),adding_a_robot.md(✓ from docs: Adding-a-Robot embodiment cookbook (closes #64) #135)