Imagine you're a building inspector who certifies houses by checking that at least 90% of the bricks are regulation-weight. You assume that if most bricks pass, the house is structurally sound. But what if someone builds a house where every single brick is regulation-weight, yet the walls are assembled wrong — load-bearing beams in the wrong places, mortar missing at key joints? Every local check passes, but the building is unsound. That's the gap this paper exposes: the Flowering IOPP confused 'most local pieces look fine' with 'the global structure is correct,' and those two things point in opposite directions. The committed claim is blunt: the soundness proofs in both Flowering IOPP papers (ISIT 2025, arXiv:2501.14337 and CANS 2025, arXiv:2508.10510) are false. The core error is that the proofs treat vertex distance of a word to the code as equivalent to the fraction of unsatisfied local constraints. Only one inequality holds between these quantities — and it's the wrong direction for the soundness argument. This isn't a minor tightening issue; the inequality goes the wrong way entirely, meaning the proof strategy cannot be patched by adjusting constants. The Flowering protocol belongs to the family of Interactive Oracle Proofs of Proximity (IOPPs), which are the cryptographic backbone behind scalable proof systems like STARKs. FRI (Fast Reed-Solomon IOPP) is the gold standard for Reed-Solomon codes: it recursively folds a codeword and checks proximity at each level. Flowering attempted to generalize FRI from Reed-Solomon codes to codes on graphs with Reed-Solomon local codes — a natural and important extension. The recursive structure mirrors FRI: cut the graph into isomorphic subgraphs, fold along the cuts, repeat. The ambition was sound; the proof was not. What makes this paper remarkable is its provenance. Hugo Delavenne is an author on both of the broken papers. This is a self-correction — the rarest and most valuable kind of erratum in cryptography, where flawed soundness proofs can silently propagate into deployed systems. The paper doesn't just wave at the error abstractly; it provides an explicit counterexample demonstrating that the soundness bound fails. You can check it yourself. The integrity story here is unusually clean precisely because it's a negative result. There's no benchmark gaming, no cherry-picked comparisons. The argument is mathematical: here is the inequality the proof assumed, here is why it doesn't hold, here is a concrete instance where it breaks. This is the kind of paper that makes the proof-system literature more trustworthy, not less, even though the news is bad. For practitioners building on STARK-like proof systems, the immediate practical question is: does any deployed system depend on the Flowering IOPP's soundness claim? As of now, Flowering was a research-stage protocol, not yet integrated into production proof systems. But the graph-code generalization of FRI is a live research direction with real engineering interest — anyone building toward it needs to know the foundation has a crack. The generalization problem (FRI for codes beyond Reed-Solomon) remains open. The deeper lesson is about the gap between local and global notions of distance in codes on graphs. The fraction of unsatisfied local constraints is an upper bound on vertex distance in one direction, but the proof needed it in the other. This is a structural feature of graph-based codes, not a bug in one proof attempt, which means the next attempt at a Flowering-like IOPP will need a fundamentally different soundness strategy — not a patch.
The analogy
Like a building inspector who checks every brick individually meets regulation weight, but misses that the walls are assembled wrong. Local checks all pass; the global structure is unsound. The Flowering proof confused 'most local pieces look fine' with 'the global codeword is close to valid' — and the inequality between those two quantities points in the wrong direction.
Brick inspector
Checks individual bricks meet spec, assumes the wall is therefore structurally sound
Flowering IOPP
Checks fraction of unsatisfied local constraints, assumes this bounds vertex distance to the code — but the inequality goes the wrong way
Honest assessment
This is a rare and valuable negative result: an original author proving their own soundness claims wrong with an explicit counterexample. It doesn't advance the graph-IOPP frontier — it resets it. Anyone working on STARK generalizations beyond Reed-Solomon needs to read this. The integrity is exemplary, but the milestone impact is a step backward by design. Most useful to cryptographers working on proof-system foundations, not to engineers using existing STARKs.
What Is CLAIMS?
73
/ 100
BedrockNews applies a structured analytical framework called CLAIMS to academic papers and technical research releases.
How CLAIMS worksClaim
4/5
Ladder
4/5
Architecture
3/5
Integrity
5/5
Milestone
2/5
Successor
4/5
What is this paper actually committing to?
Genuine negative result: proves two published soundness claims are false with an explicit counterexample. Not an incremental finding — it invalidates a research direction's foundation.
The field fight
Can FRI-style IOPPs be generalized from Reed-Solomon codes to richer code families (e.g., codes on graphs)?
Side A
Yes — recursive folding generalizes naturally; graph structure provides the needed algebraic scaffolding
Held by: Flowering (Delavenne, Medevielle, Roussel), Delavenne & Lallemand
Side B
Not proven — the local-to-global distance relationship in graph codes may not support FRI-style proximity arguments
Held by: This paper (Delavenne, self-correcting)
Milestone tracker
Unlocks: STARK-like proof systems over code families richer than Reed-Solomon, enabling more flexible verifiable computation
Current
0
sound graph IOPPs
post-Flowering
Next
1
sound graph IOPP
first provably sound construction
Unknown — requires new proof strategy, not incremental improvement
Verified Facts (5)
The Flowering protocol is from Delavenne, Medevielle and Roussel (ISIT 2025, arXiv:2501.14337)
The generalization is by Delavenne and Lallemand (CANS 2025, arXiv:2508.10510)
The commit soundness proofs identify the vertex distance of a word to the code with the fraction of its unsatisfied local constraints
Only one inequality holds, in the wrong direction
An explicit counterexample is provided
Practical Implications (4)
No deployed STARK system is affected — Flowering was research-stage only
Anyone building toward IOPPs for codes on graphs must find a new soundness strategy; the Flowering approach is dead
The local-to-global distance gap in graph codes is now a known structural obstacle, not just a proof difficulty
FRI for Reed-Solomon remains the only IOPP with proven soundness in its family
More latest news
Read all