45 transactions
Inspect each transaction's declared permissions, recorded outcome, and instruction evidence.
Understand the contention
Follow a conflict back to its accounts
Two transactions have a declared account conflict when they share an account and at least one declares it writable. Read/read overlap creates no conflict edge. The microscope turns those declarations into account relationships, connected components, and estimated conflict-free waves that you can inspect transaction by transaction.
A concrete example: captured devnet slot 488638500
116 declared conflict pairs
Select an account to see which transactions share it and which pairs its permissions connect.
15–15 full-block wave bounds
A 15-member conflicting clique and a valid 15-wave grouping agree on the minimum for this declared-lock model.
45 of 45 compute values recorded
The supplied metadata totals 122,696 CU. Compute describes recorded computational work, not elapsed execution time.
A wave is a modeled group with no retained account conflicts. Wave counts describe estimated execution waves under the stated model; they do not reconstruct a validator's private schedule or measure runtime speedup.
Explore four bundled examples
- Real devnet: loader buffer writes
Trace a program-buffer upload to decoded write chunks and inspect hypothetical payer and byte-range models.
- Real devnet: cross-payer funding
Find different fee payers drawing from a shared funding source. Different payers alone do not establish independent application work.
- Real devnet: upload across nine blocks
Inspect payload byte ranges, gaps, overlaps, and exact instruction evidence. The captured range does not establish a complete upload lifecycle.
- Synthetic teaching: shared application state
A deliberately constructed fixture illustrates application-state contention. It is labeled synthetic and is not observed network activity.
From source to a report
Try the reproducible local workflow
Generate the demo
Download or clone the source, install the locked dependencies, and generate the reports from bundled fixtures. After installation, this demo works offline.
Trace the evidence
Search a transaction, select a shared account, or inspect a finding. Follow each relationship to declared permissions and available instruction evidence.
Read the bounds in context
Inspect the wave grouping and its lower-bound witness. Matching bounds establish a minimum within that model; different bounds leave the minimum unresolved.
Local quickstart
Use Node.js 22.12 or newer and npm. Run these commands from the project folder:
npm ci
npm run demoOpen output/demo/index.html in your browser. For example, the single-block loader report is written to output/demo/real-loader/report.html.
The tool analyzes public block data and writes local reports. The demo needs no wallet, API key, RPC account, database, or server.
Make the scope visible
Evidence you can check, with limits you can read
The independent verifier reconstructs declared conflicts from normalized account permissions, checks connected-component membership, and verifies that every transaction appears exactly once in a conflict-free wave grouping. It checks each pair in the saved clique witness, bound claims, status counts, and recorded compute coverage. It also prints the checked file's SHA-256.
npm run verify:analysis -- --analysis output/demo/real-loader/analysis.jsonThis certificate establishes internal consistency of the supplied declared-lock evidence. It does not authenticate the source RPC data, certify semantic explanations or counterfactual assumptions, or establish validator execution. A file hash identifies the checked bytes; it does not prove their origin.
- Compute coverage stays explicit. Recorded compute is computational work, not elapsed time. Missing compute remains unknown; a partial subtotal is not a full-block total, and comparisons suppress compute differences when coverage is incomplete.
- Conditional models retain their assumptions. Payer decoupling and byte-range refinement explore potential parallelism under hypothetical constraints. Their results do not establish a safe application redesign, current validator support, or a deployable performance gain.
- Captured and synthetic evidence stay distinct. Real examples use saved public devnet fixtures. The synthetic application-state case is teaching data. Range observation coverage and upload-lifecycle coverage are separate questions.
For a broader local check, npm run verify:portable runs strict compilation, deterministic portable tests, fixture checks, demo generation, and the bundled reviewer workflow. Saved demo data can be checked separately with npm run verify:demo; rendered HTML is outside the data certificate.