Skip to content

Proof scope, baselines, coverage, prior art and qualification path - #18

Merged
wbnns merged 7 commits into
mainfrom
claims-and-pages
Oct 11, 2026
Merged

wbnns merged 7 commits into
mainfrom
claims-and-pages

Conversation

@wbnns

@wbnns wbnns commented Oct 11, 2026 •

Copy link
Copy Markdown
Member

Pages

  • Shrinking quorum: lede without "decides exactly as fixed TMR". "What it guarantees" is split into Proved (per tick) (four Kani harnesses, linked, with scope) and Tested (per run). New sections: Degradation-first (quoted from the whitepaper; the nav item now points here), Two replicas (what happens when the known-answer test passes on both, fails on both, or cannot tell), Command path, Open questions. The 136-year break-even now sits next to the per-mission risk metric.
  • quorum crate: Kani table with a scope column for all nine proofs, a "Not covered" list (Health, diagnosis, timing, the system), and a #gate example that gates commands on Verdict.mode (compiled and run; also in flight as quorum/examples/gate.rs).
  • Dusk: exact definitions of useful, service, wrong days and risk from src/model.rs; new Baselines section (TMR with an unchecked fallback; cold standby swept over spare aging 0.1 to 1.0); default output from a real run; ground support section.
  • Bench: new Known-answer test coverage section with scope, the coverage table and the test-period table (from flight bench/results/*-kat.json).
  • CLI: the stuckbit fault and the corrupt fix.
  • Architecture: layers L0 and L2 to L5 "Not built"; OPS-SAT "proposed".
  • New pages: Prior art and related work (eight references, each checked against Crossref or NTRS, DOIs linked); Qualification path (NPR 7150.2, NASA-STD-8739.8, ECSS-E-ST-40C, ECSS-Q-ST-80C; "No qualification evidence exists yet"; Rust toolchain plan: TODO).
  • Commands: every quickstart and page command was run as written. The quickstart failed as written (two binaries); it now uses --bin quorum / --bin dusk, with the real output.
  • llms.txt summary: TRL removed; nine Kani proofs (four core, five C interface).

Shared files

scripts/sync-tokens.mjs now also copies src/data/claims.json, src/data/claims.ts and src/components/Claim.astro from the website; check:tokens fails if they drift. The sync also brings src/styles/shared/site.css up to date with website PRs #31 and #32, which had not been synced.

Checks

npm run check passes locally against the local website. Visual check of 10 pages in light and dark at 1280 and 375 px: no overflow, no console errors.

Merge order

Merge after website #34: CI's check:tokens reads website main, so it fails here until then. The bench page links to bench/results/2026-10-11-darwin-arm64-kat.json, which exists on flight main after flight #9 merges.

wbnns added 7 commits October 11, 2026 13:16
…ints to its section

Prior art lists only verified references (von Neumann, Lyons and
Vanderkulk, STAR, Sklaroff, Meyer, Lamport et al., Spector and Gifford,
Rasmussen) with how Mru relates to each. Qualification path maps what
exists to NPR 7150.2, NASA-STD-8739.8, ECSS-E-ST-40C and ECSS-Q-ST-80C by
topic and makes no compliance claim. Metadata for both, and descriptions
updated for the shrinking quorum, quorum crate and Dusk pages.
…mmand path, open questions

New lede without 'decides exactly as fixed TMR'. Each Kani harness is
linked with its scope. Fixed TMR is framed as the baseline, not as what
spacecraft fly; the Shuttle and degradable systems are cited as prior art.
Adds the Degradation-First Principle and corollary (whitepaper section 5),
the KAT outcomes at two replicas with their gaps, the command path, and
open questions. The 1.34x states its no-ground-support assumption. A
RISK-METRIC marker holds the place for Dusk's per-mission risk.
…covered, and a command gate example

The gate example compiles and runs against v0.2.0; its output is pasted.
F´ page: name the four functions the C interface proofs cover; null
pointers are unit-tested, not proved.
…ground support section

cargo run needs --bin because the package has two programs. Times are
from a run on an Apple M4. Useful, service and wrong days are defined as
src/model.rs computes them. The ground table is pasted from a real run and
the 1.34x states that it falls to 1.00x when support never ends. A
DUSK-BASELINES marker holds the place for the new baselines.
…ed; llms.txt summary

Quickstart uses --bin quorum and --bin dusk (cargo run alone is
ambiguous in both repos) and shows the real output lines. Bench run times
measured on an Apple M4. Bench page no longer says both policies 'make the
same decision'. Architecture labels Layers 0 and 2 to 5 'Not built' and
the 1.34x states its ground assumption. llms.txt: no TRL, nine proofs (four
core, five C interface).
…d risk, known-answer test coverage on the bench page, stuckbit in the CLI page; shared styles synced
@wbnns wbnns mentioned this pull request Oct 11, 2026
@wbnns
wbnns merged commit 1596a35 into main Oct 11, 2026
1 of 3 checks passed
@wbnns
wbnns deleted the claims-and-pages branch October 11, 2026 14:28
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant