Method
Method and limits.
What I read, what I do not read, how to run it again, and where it stops.
What I read
Each entry of the registry rebuilds one incident from raw chain data (eth_getLogs, decoded transaction receipts, eth_call reads), not from press summaries.
Each folder keeps its Python script, its registre_hypotheses.csv (every claim broken into a hypothesis with a test that could prove it wrong) and the raw output of the script in resultats_*.txt. A figure I recompute can be checked against the file that produced it.
Role claims (admin, multisig, governance, infrastructure) come only from on-chain records and cited public sources. They are not allegations of wrongdoing, fraud or illegal conduct.
What I do not read
71 of the 72 entries rebuild their incident from public chain data, and their script reads the chain live. The exception is tectonic-cronos: according to Cronos’s report, the chain was rolled back past that attack, so the attack cannot be rebuilt from the chain.
18 entries give a figure that is a known floor, not a full total. The README of the repository lists each one with the reason.
13 figures are marked †. They come from the protocol’s own post-mortem, from the press or from DefiLlama, and I did not recompute them. The footnote of each names the source.
Funds that the attacker returned, or that were reversed on-chain, are deducted from the loss. Reimbursements that a protocol paid from its own reserves are not.
How to run it again
pip install web3
python3 sandbox-oft-delegate-hijack/reconstruct_exploit.pyThe historical figures should match sandbox-oft-delegate-hijack/resultats_reconstruction_2026-09-09.txt. Balances and roles that the script reads today may have moved since.
To check that the totals of the README add up from the Index, run python3 check_readme_consistency.py.
Open the repository Open the registry
Limits
This is independent research, not a security audit. I am not affiliated with any protocol, auditor or outlet cited in a folder.
A finding is a snapshot. The state of a chain changes after publication, and the reports are not updated by themselves.
No protocol or team discussed here commissioned or audited this work. It is not legal, financial or investment advice.
Disclaimer and licence
About this research program
An ongoing program of independent on-chain research, same discipline throughout: primary-source anchors, on-chain reconstruction, corrections issued openly when something's found wrong. Findings and on-chain sources are always public and free to reuse with credit. Ongoing work: X. Corrections and questions: contact@realspap.com.
Disclaimer
Independent, factual analysis of public on-chain data as of the date noted in each repo's Status/Method section. Role claims (admin, multisig, governance, infrastructure) come solely from on-chain records and cited public sources; they are not allegations of wrongdoing, fraud, or illegal conduct. Concentration of control is reported as a structural fact, not a moral or legal judgment; individual key holders are not named. Findings are a snapshot; on-chain state changes after publication and reports aren't auto-updated. Independent research, not commissioned or audited by any protocol or team discussed; not legal, financial, or investment advice. Believe something here is inaccurate? Contact via X with evidence and it's corrected promptly. Verify all cited addresses, transactions, and figures independently.
onchain-postmortems uses its own disclaimer, tailored to individual exploit reconstructions rather than portfolio-wide screening; see that repo's own Disclaimer section.
License note
Unless a repo states otherwise, findings, data and documentation are published under CC BY 4.0: reuse them freely, including commercially, with credit to RealSpap and a link to the source repo. Code in those repos and on this site is MIT. onchain-postmortems is MIT in full, including its reconstruction scripts and its dataset, since reproducibility on already-public historical incidents is the point. The private tooling that runs the multisig and admin-key checks is not published.
To report a mistake: Find my errors.