RAG Lab

A retrieval pipeline, measured rather than described. 93 pages of this site, chunked into 825 passages, embedded on a laptop CPU and searched exactly in pgvector. Twenty questions were written and committed before anything was graded; 16 of 20 retrieved the right page. The other 4 are on this page too, with the passages that came back instead.

Every figure below was produced by github.com/pigfox/rag-lab and is read from that repository's own eval/results.json, embedded here byte-for-byte. Nothing on this page runs a pipeline, embeds a model or contacts a server — it reports a run that already happened, and the repository is where you re-run it.

The result

0.800 hit rate @5 a gold page anywhere in the top 5 — 16 of 20
0.775 recall @5 the fraction of each question's gold pages found, averaged
0.782 MRR over the full depth-10 list, not the cut
0.627 nDCG @10 binary gain, ideal capped by the relevant chunks that exist
23.7 ms median latency mean 24.9 · p95 30.3 · max 31.6 ms

Hit rate and recall are two different numbers, and both are above. Hit rate asks whether any gold page reached the top 5; recall asks what fraction of a question's gold pages did. They are identical for the eighteen questions with a single source and differ for the two that name several — which is why they are printed separately. Quoting the first under the second's name is the ordinary way a retrieval result gets overstated.

How these numbers were produced

  • Local embeddings. sentence-transformers/all-MiniLM-L6-v2, 384 dimensions, running on CPU in a container. No API key, no account and no per-run cost, so re-running it costs a reader nothing.
  • The corpus is committed and frozen. 93 pages were fetched once and are stored in the repository. Every figure is computed over those exact bytes, so the same command gives the same answer tomorrow.
  • The question set was locked before grading. Twenty questions, each naming the page that answers it and stating what on that page answers it, committed before any retrieval ran against this index.
  • Graded by published code. Retrieval writes a predictions file; a separate grading step scores that file and never retrieves. A metric change therefore grades the same predictions a reader already has.
  • The raw output is published unedited — one graded question per line, every retrieved passage with its score, hits and misses alike. What is on this page is that file, not a summary of it.
  • The failures are shown. All 4 misses are below with the passages the pipeline returned instead of the right ones.

What was searched

93 documents became 825 chunks at 180 words each with 45 words of overlap. Search is exact — a full scan with an inner product, not an approximate index — because at this size exact is a few milliseconds and an approximate index would fold its own recall into the figure above. The question reaches the embedding model exactly as written: nothing is expanded, decomposed or stemmed.

Every question, and what it retrieved

Pick a question to see its full ranked list. Gold passages — the ones from a page the question set names as an answer — are marked, and the cut at rank 5 is drawn, so a near miss below it is visible as one.

Question
q01 HIT When a queue between two components fills up, what are the only things that can happen next?

Should have retrieved: https://pigfox.com/demos/backpressure-lab

recall 1.000 · nDCG 0.983 · first gold passage at rank 1 · 31.6 ms

  1. 1. https://pigfox.com/demos/backpressure-lab GOLD 0.593

    Skip to content Backpressure Lab A bounded queue you can push past its own capacity, and three different answers to what should happen next. Every queue between two components has a capacity. The interesting question is never how big it …

  2. 2. https://pigfox.com/demos/backpressure-lab GOLD 0.491

    changed — the numbers on this page are always the numbers that ran. Workers Arrivals per second Service time (ms) Queue capacity When the queue is full Block the producer Drop the newest Drop the oldest Run it Stop Idle. …

  3. 3. https://pigfox.com/demos/backpressure-lab GOLD 0.462

    page over a server-sent event stream ten times a second. That is the opposite of the keygrind demo next door, where everything happens in your tab — and it is why this page has limits that one does not need. …

  4. 4. https://pigfox.com/demos/kafka-lab 0.435

    is three inputs and one line of integration: the ceiling is min(consumers, 3) × 1000 ÷ work, and the backlog integrates the difference between the producer rate and that ceiling, floored at empty. The cap is the topic's partition count: …

  5. 5. https://pigfox.com/demos/backpressure-lab GOLD 0.423

    each figure is the top of the bucket its percentile falls in and never reads lower than the truth. What to try Leave the defaults and run it three times, changing only the policy. The arrival rate is above what …

  6. 6. https://pigfox.com/demos 0.413

    depots among sixteen real Aegean ports, solved two ways in your browser. Max-coverage greedy is submodular, so it is proven to reach 1 − 1/e of the optimum; the p-median heuristic beside it carries no guarantee at all — and …

  7. 7. https://pigfox.com/tools/go/channel-deadlock 0.410

    interleaving Start from one of these The grammar Published in full, because the parser refuses everything outside it. This block is the same string the tool itself checks against, so the two cannot drift apart. The semantics it models, and …

  8. 8. https://pigfox.com/demos 0.407

    all. JavaScript ARIA Go Open demo → Source WASM Keygrind EVM vanity addresses ground out in your browser by Go compiled to WebAssembly. Keys are derived from the browser's own cryptographic random source and never leave the tab — pull …

  9. 9. https://pigfox.com/demos/etl-lab 0.396

    ones such a book actually carries. Every fill is simulated — the profit and loss is arithmetic over those generated rates against a fixed 1000-unit notional, with seeded slippage of up to 85% of the theoretical edge subtracted from it. …

  10. 10. https://pigfox.com/demos/pprof-lab 0.394

    Across the committed grid rows, all three overload policies completed exactly 392 messages. On these captures the policy decided who paid for the overload rather than how much work got done. That is what these eight runs show — it …

The dashed rule sits after rank 5: everything below it was retrieved but falls outside the cut the hit rate and recall figures use.

q02 HIT How many simulation sessions does the backpressure demo run at the same time before it turns one away?

Should have retrieved: https://pigfox.com/demos/backpressure-lab

recall 1.000 · nDCG 0.805 · first gold passage at rank 1 · 30.3 ms

  1. 1. https://pigfox.com/demos/backpressure-lab GOLD 0.627

    page over a server-sent event stream ten times a second. That is the opposite of the keygrind demo next door, where everything happens in your tab — and it is why this page has limits that one does not need. …

  2. 2. https://pigfox.com/demos/backpressure-lab GOLD 0.584

    Skip to content Backpressure Lab A bounded queue you can push past its own capacity, and three different answers to what should happen next. Every queue between two components has a capacity. The interesting question is never how big it …

  3. 3. https://pigfox.com/demos/etl-lab 0.469

    a network, holds a credential or places an order. Where the work happens The simulation runs on the server, in Go, and the frames reach this page over a server-sent event stream. A run starts real goroutines on a real …

  4. 4. https://pigfox.com/demos/backpressure-lab GOLD 0.445

    changed — the numbers on this page are always the numbers that ran. Workers Arrivals per second Service time (ms) Queue capacity When the queue is full Block the producer Drop the newest Drop the oldest Run it Stop Idle. …

  5. 5. https://pigfox.com/demos/etl-lab 0.443

    — 40 nanoseconds per raw row — rather than from wall-clock time. A rate measured against the wall would describe whichever machine this happened to run on; a rate measured against an explicit charge is a property of the simulation, …

  6. 6. https://pigfox.com/demos/reconcile-lab 0.435

    over a server-sent event stream ten times a second. A session starts real goroutines on a real box and holds them, so the server admits 4 simulations at a time, one per address, and runs no more than 32 reconcile …

  7. 7. https://pigfox.com/demos 0.415

    depots among sixteen real Aegean ports, solved two ways in your browser. Max-coverage greedy is submodular, so it is proven to reach 1 − 1/e of the optimum; the p-median heuristic beside it carries no guarantee at all — and …

  8. 8. https://pigfox.com/demos/reconcile-lab 0.415

    nothing about the one you chose. One result did not survive the move from the test bench to this page, and it is worth knowing why. The same simulation runs under a virtual clock in its own test suite, where …

  9. 9. https://pigfox.com/demos/reconcile-lab 0.406

    retry at once against retry with backoff. Both get slower and both waste far more, because every competing write now lands on the object you are reconciling — but retrying at once stays the quicker way back, and buys it …

  10. 10. https://pigfox.com/demos/pprof-lab 0.391

    what is on-CPU at that instant. This workload is asleep almost all of the time, so a five-second run yields only a few dozen samples. The limitation is the sample count — you cannot meaningfully diff two trees built from …

The dashed rule sits after rank 5: everything below it was retrieved but falls outside the cut the hit rate and recall figures use.

q03 HIT Why does replacing the browser's own dropdown control mean losing its built-in required-field checking?

Should have retrieved: https://pigfox.com/demos/unselect2

recall 1.000 · nDCG 0.613 · first gold passage at rank 1 · 28.1 ms

  1. 1. https://pigfox.com/demos/unselect2 GOLD 0.566

    Skip to content unselect2 Replaces the <select> but looks like and behaves like the <select>. Watch: 30 seconds The native <select> cannot be styled, so there is not one here — not hidden, not one per field, zero. A listbox …

  2. 2. https://pigfox.com/demos 0.407

    is evidence for. Move a threshold and it re-runs in your browser, then print the evidence report. Go WebAssembly LLMs Open demo → Agent Gateway A policy gate in front of six AI agents, replaying 150 requested actions as allowed, …

  3. 3. https://pigfox.com/demos/agent-gateway 0.346

    <seq> <VERDICT> <one short sentence of reasoning> Adjudicate all 38. Do not skip any, and do not group them. What this page got wrong This page shipped three defects and its own tests caught all three before anybody else did. …

  4. 4. https://pigfox.com/demos/agent-gateway 0.340

    visible with nothing to print or download. Why it matters. The least serious of the three, and the most instructive about where defects live. This is the second page in this estate to ship it and the sixth to have …

  5. 5. https://pigfox.com/demos/agent-gateway 0.316

    would have moved it to a neighboring value. Why it matters. Smaller than the first, and the same shape: a control that quietly reports a setting nobody chose. Every other slider on the page was already aligned, which is why …

  6. 6. https://pigfox.com/demos/agent-gateway 0.316

    so. What caught it. The browser tests, on their first run against dev. Nothing on the server could have seen it — the markup was correct, the Go replay was correct, and the two agreed. Only a browser applies an …

  7. 7. https://pigfox.com/demos/agent-gateway 0.310

    — the bounds arrive from the engine, deliberately, so the page holds no copy of a Go constant — so $5,000.00 was clamped to $1.00 and snapped onto the step grid before a line of JavaScript executed. Setting the real …

  8. 8. https://pigfox.com/demos/agent-gateway 0.249

    a browser resolves a cascade, so the check now reads the computed style on every route. What this is not It is not a gate in front of anything real. The agents, the actions, the amounts and the resources are …

  9. 9. https://pigfox.com/demos/access-audit-lab 0.247

    shows you what the evidence looks like. This page prepares evidence. It issues no compliance opinion — that remains the act of an independent audit firm. A control citation says which control a finding speaks to. It does not say …

  10. 10. https://pigfox.com/demos/agent-gateway 0.245

    are data in that package, not conditionals in this page, which is what makes changing a threshold re-run a policy rather than re-render a table. Two rules read history, so the order matters. A rate cap and a spend cap …

The dashed rule sits after rank 5: everything below it was retrieved but falls outside the cut the hit rate and recall figures use.

q04 HIT How large is the vanity-address grinder's WebAssembly module, and why is it that size?

Should have retrieved: https://pigfox.com/demos/wasm-keygrind

recall 1.000 · nDCG 1.000 · first gold passage at rank 1 · 28.4 ms

  1. 1. https://pigfox.com/demos/wasm-keygrind GOLD 0.513

    Skip to content WASM Keygrind EVM vanity addresses, ground out in your browser by Go compiled to WebAssembly. Every key is derived in this tab, from the browser's own cryptographic random source. Nothing is transmitted: there is no endpoint behind …

  2. 2. https://pigfox.com/demos/wasm-keygrind GOLD 0.443

    Before you use a key from this page This is a demonstration. A private key generated in a web page has been held in browser memory, on a machine whose other tabs and extensions you are trusting — treat anything …

  3. 3. https://pigfox.com/demos 0.369

    all. JavaScript ARIA Go Open demo → Source WASM Keygrind EVM vanity addresses ground out in your browser by Go compiled to WebAssembly. Keys are derived from the browser's own cryptographic random source and never leave the tab — pull …

  4. 4. https://pigfox.com/tools/go 0.335

    your input, hands it over, and draws what comes back. Nothing is reimplemented in JavaScript, which is the only way the answer a page gives can be the answer Go gives. The section ships two modules rather than one per …

  5. 5. https://pigfox.com/blog 0.331

    system design more than it changes prompts. If you build with long-context models, inspect retrieval precision, prompt assembly, injection resistance, and output validation before you scale usage. When your executable also ships a SQLite database 2026-08-24 A binary which also …

  6. 6. https://pigfox.com/demos/mmr-log 0.324

    per-hash constant is essentially unchanged at 152.2 against v2's 151.2, while the fitted base moves from 3002.8 to 5392.8. The wire format did not make the Merkle work more expensive — it added a fixed overhead to every call. That …

  7. 7. https://pigfox.com/tools/go/slice-growth 0.321

    Skip to content Go Slice Growth Go stopped doubling slice capacities in 1.18, and the capacity you get is not the one the growth formula asked for. This shows both: the formula's answer, the allocator's rounding up to a size …

  8. 8. https://pigfox.com/tools/go/slice-growth 0.319

    for the same reason. The table above shows what the formula asked for and what you actually got, in separate columns, because the difference between them is the point. Where these numbers come from Not from memory, and not from …

  9. 9. https://pigfox.com/demos/mmr-log 0.317

    5392.8 base and 152.2 per hash, from an independent run against a fresh log. Where an interface description LIVES costs a program nothing at run time, because the chain never reads it: only explorers and clients do. The two repositories …

  10. 10. https://pigfox.com/tools/go/slice-growth 0.310

    the ratio column in the table above is full of numbers like 1.63 and 1.38 — values that appear in no blog post because nobody expects them. And the capacity you get is not the capacity it asked for This …

The dashed rule sits after rank 5: everything below it was retrieved but falls outside the cut the hit rate and recall figures use.

q05 HIT How can an append-only log with a million entries cost the same on-chain storage as one with ten?

Should have retrieved: https://pigfox.com/demos/mmr-log

recall 1.000 · nDCG 0.766 · first gold passage at rank 1 · 28.5 ms

  1. 1. https://pigfox.com/demos/mmr-log GOLD 0.493

    by strangers by design. This page used to say that an explorer shows this program's instructions as raw bytes, and that this was by design rather than an omission — the consequence of being a native program that publishes no …

  2. 2. https://pigfox.com/demos/mmr-log GOLD 0.466

    Same root Binary CU per append v1 → v2 byte for byte 28088 → 28760 bytes (+672) +403 to +522 (mean 484.1) v2 → v3 byte for byte 28760 → 48696 bytes (+19936) +2393 to +2401 (mean 2395.6) v3 → …

  3. 3. https://pigfox.com/demos/mmr-log GOLD 0.463

    Skip to content An append-only log that commits in 1040 bytes, however long it gets A Merkle Mountain Range on Solana devnet: the account keeps at most 32 peak hashes and a count — never the leaves, never the interior …

  4. 4. https://pigfox.com/demos 0.458

    was attacked — they lose it because the consuming contract used a price that was stale, negative, or from a round that never finished. A contract on Base Sepolia reads live Chainlink feeds and names which, with the raw round …

  5. 5. https://pigfox.com/demos/mmr-log GOLD 0.453

    as well as on the measured constants. With the model's own PER_HASH=350 the figure is 717,250 CU, which is already under 1,400,000 - so the sentence did not follow from the numbers printed two lines above it. A naive rebuild …

  6. 6. https://pigfox.com/demos/mmr-log GOLD 0.434

    and appending at n = 8 runs none. Every figure below was measured once against devnet and committed to the repository; this page performs no chain call, because a compute-unit cost is a property of a run and re-measuring on …

  7. 7. https://pigfox.com/demos/deposit-dispute 0.411

    field anywhere in the schema they answer in. That is the same division the contract enforces on chain. The live panel above runs all three slots against a single vendor, so it demonstrates per-item independence and the 2-of-3 threshold — …

  8. 8. https://pigfox.com/demos/mmr-log GOLD 0.395

    was uploaded. The IDL was not uploaded until it had decoded real transactions, and a checklist was not what decided that. A checklist is what let the two oldest versions freeze too early, twice. This file is hand-written, so nothing …

  9. 9. https://pigfox.com/tools/go/slice-growth 0.392

    []T. Anything else is what you passed to make. The element type contains a pointer A string, a slice, a map, an interface or an actual pointer. Past 512 bytes this adds an 8-byte allocation header, and the header can …

  10. 10. https://pigfox.com/demos/mmr-log GOLD 0.391

    measured on Solana devnet. Every row links to its transaction. n merges hashes predicted measured naive rebuild tx 7 → 8 3 5 3250 6153 7676 explorer 8 → 9 0 3 2550 5840 7980 explorer 9 → 10 1 …

The dashed rule sits after rank 5: everything below it was retrieved but falls outside the cut the hit rate and recall figures use.

q06 HIT Which test network were the Cosmos proofs captured from, and at what block height?

Should have retrieved: https://pigfox.com/demos/cosmos-lab

recall 1.000 · nDCG 1.000 · first gold passage at rank 1 · 23.7 ms

  1. 1. https://pigfox.com/demos/cosmos-lab GOLD 0.623

    Skip to content Cosmos Lab A Cosmos signed header and two chained ICS-23 proofs, captured once and replayed offline. The header, its validator set and both state proofs were captured from osmo-test-5 at height 68199363 on 2026-08-11T18:56:17Z and are frozen. …

  2. 2. https://pigfox.com/demos/cosmos-lab GOLD 0.599

    at 68199362 Module store bank The state is read at one height and committed by the header at the next one: a Cosmos header's AppHash is the root of the state the previous block left behind. Both heights are recorded …

  3. 3. https://pigfox.com/demos/cosmos-lab GOLD 0.543

    checked against a different chain ID. Expects: rejected ICS-23 membership As captured The key, its value and both proof layers exactly as captured. Expects: accepted One bit flipped in the value The proof is untouched; one bit of the value …

  4. 4. https://pigfox.com/demos/cosmos-lab GOLD 0.540

    flipped in one validator's ed25519 signature. Nothing else is touched. Expects: rejected A majority of validators, a minority of power The largest validators are marked absent, leaving a majority of the validators by head count signing — and far less …

  5. 5. https://pigfox.com/demos/cosmos-lab GOLD 0.497

    validators carry more than the threshold between them. So a minority of validators can pass this commit, and a majority of them can fail it. Two scenarios below do exactly that. What the proofs are about Both proofs are checked …

  6. 6. https://pigfox.com/ 0.429

    below is fetched from the network on load, and every address resolves on BaseScan. No model is involved in any of them. RWA Tokenization Explorer ERC-3643-style KYC-gated transfers. ZK KYC Pass Explorer Prove a credential without revealing an identity. Gas …

  7. 7. https://pigfox.com/demos 0.410

    second, whatever its rate limit says — drag the work delay, watch a backlog build, then add consumers and watch it drain. The runnable Kafka stack is linked underneath. Go WebAssembly Kafka Open demo → Source Cosmos Lab A signed …

  8. 8. https://pigfox.com/demos/orbit-lab 0.395

    at any point. Coastlines from Natural Earth at 1:110m, public domain, decimated to the tolerance recorded in the artifact's own header and reported below the globe. Positions are as accurate as the element sets behind them, and those were retrieved …

  9. 9. https://pigfox.com/github-repos 0.374

    per-call metering against a pluggable credit store, rate limiting, and cost ceilings. eth-bridge-go — A Go library and CLI for moving ETH across the OP Stack Standard Bridge. Same-chain transfers run on any EVM chain; deposits and withdrawals run between …

  10. 10. https://pigfox.com/golang 0.351

    leg and persist prove parameters; prove and finalize are not implemented. mcp-metered — authentication, per-call metering, rate limiting and cost ceilings for a Model Context Protocol server. cosmos-light-go — replays Cosmos consensus and state proofs offline, from chain data captured …

The dashed rule sits after rank 5: everything below it was retrieved but falls outside the cut the hit rate and recall figures use.

q07 HIT If two dependencies ask for different versions of the same module, which version does the build actually use?

Should have retrieved: https://pigfox.com/tools/go/mvs

recall 1.000 · nDCG 0.802 · first gold passage at rank 1 · 22.0 ms

  1. 1. https://pigfox.com/tools/go/mvs GOLD 0.556

    be resolving something else and not telling you. Version ordering, and the rule that catches people Versions compare by semantic-version precedence. Major, minor and patch numerically; then a pre-release ranks below the release it precedes, so v1.0.0-rc1 loses to v1.0.0. …

  2. 2. https://pigfox.com/tools/go/mvs GOLD 0.489

    the newest that exists on the proxy — the newest that something actually asked for. That is the whole of minimal version selection, and it is why a Go build is reproducible without a lock file: adding a dependency can …

  3. 3. https://pigfox.com/tools/go/mvs GOLD 0.455

    asked for that. So every module here shows every version that was required, which module required each one, and the shortest requirement path from your main module to the requirement that won. Contested modules are listed first, because those are …

  4. 4. https://pigfox.com/tools/go/mvs GOLD 0.452

    Skip to content Go Minimal Version Selection Your go.mod asks for one version and you build against another. Paste a `go mod graph` and see the selection worked out: every version required, which module asked for it, and the shortest …

  5. 5. https://pigfox.com/how-smart-contracts-are-tested 0.303

    matches the canonical one, and fails the build when it does not. Foundry. Unit, fuzz and invariant tests on a pinned toolchain, all passing and none skipped — a skipped test is a red build here, not a yellow one. …

  6. 6. https://pigfox.com/tools/go/mvs GOLD 0.278

    that end up selected. A module version that loses can still bring a dependency into your build, and an implementation that expanded only the winners would produce a build list the go command disagrees with. The other tools Go Time …

  7. 7. https://pigfox.com/demos/mmr-log 0.272

    in this lineage that records a replacement rather than a repair, and calling it a fault would be inventing a defect to justify a decision. Still upgradeable: v3 is the only entry here that is still upgradeable. Its authority was …

  8. 8. https://pigfox.com/tools/go 0.258

    your input, hands it over, and draws what comes back. Nothing is reimplemented in JavaScript, which is the only way the answer a page gives can be the answer Go gives. The section ships two modules rather than one per …

  9. 9. https://pigfox.com/demos/mmr-log 0.256

    dependency rather than a second demo: pure view functions; no transactions by design; exercised via the attestor. An address with no transactions is not a quiet contract, it is a contract nothing has ever asked a question of — so …

  10. 10. https://pigfox.com/demos/mmr-log 0.255

    can only be written by the program’s upgrade authority, so the right to describe the program and the right to change it are the same right. There were three earlier versions, and all of them are still there The first …

The dashed rule sits after rank 5: everything below it was retrieved but falls outside the cut the hit rate and recall figures use.

q08 HIT Why does the struct layout tool make you choose a target architecture instead of picking one for you?

Should have retrieved: https://pigfox.com/tools/go/struct-align

recall 1.000 · nDCG 0.746 · first gold passage at rank 1 · 22.1 ms

  1. 1. https://pigfox.com/tools/go/struct-align GOLD 0.600

    of the same shape. type Record struct { Active bool ID int64 Kind uint8 Count int64 } Target architecture A layout is not a property of the source. The same struct is a different size on a 32-bit target, so …

  2. 2. https://pigfox.com/tools/go/struct-align GOLD 0.562

    computed them is printed with the result, because the rules belong to a toolchain. A layout figure with no version beside it is a figure with no shelf life. Why the architecture is a question A struct's size is not …

  3. 3. https://pigfox.com/tools/go/struct-align GOLD 0.503

    Skip to content Go Struct Field Alignment Paste a struct and get every field's offset, the padding between them, the total size, and a reordered version with the saving — computed by go/types for the architecture you pick, not for …

  4. 4. https://pigfox.com/tools/go/struct-align GOLD 0.492

    a zero-sized field last, so the tool puts them first, where they cost nothing. This was found by permuting every ordering of a test struct and comparing — not by reasoning about it. What this deliberately does not tell you …

  5. 5. https://pigfox.com/tools/go/struct-align GOLD 0.492

    a, so the compiler inserts padding to get there. The struct itself is aligned to the largest alignment among its fields, and its size is rounded up to a multiple of that — which is where tail padding comes from. …

  6. 6. https://pigfox.com/tools/go/mvs 0.483

    that end up selected. A module version that loses can still bring a dependency into your build, and an implementation that expanded only the winners would produce a build list the go command disagrees with. The other tools Go Time …

  7. 7. https://pigfox.com/tools/go/method-sets 0.394

    error; it produces a confident wrong answer about whether your type satisfies an interface. So an import is named and refused, and what you do instead is paste the interface's declaration, which for almost every interface this question comes up …

  8. 8. https://pigfox.com/tools/go 0.388

    Go reference layout and back, directive by directive, with every unmappable directive named rather than guessed. Go Method Sets — why *T implements the interface and T does not Paste a type and an interface and get both method sets …

  9. 9. https://pigfox.com/tools/go/time-layout 0.370

    interface and T does not — Paste a type and an interface and get both method sets computed by go/types, with the compiler's own error and the rule it follows from. Go Struct Field Alignment — offsets, padding, and what …

  10. 10. https://pigfox.com/blog 0.369

    system design more than it changes prompts. If you build with long-context models, inspect retrieval precision, prompt assembly, injection resistance, and output validation before you scale usage. When your executable also ships a SQLite database 2026-08-24 A binary which also …

The dashed rule sits after rank 5: everything below it was retrieved but falls outside the cut the hit rate and recall figures use.

q09 HIT Can I get my money back after buying a pack of analyses?

Should have retrieved: https://pigfox.com/refunds

recall 1.000 · nDCG 0.949 · first gold passage at rank 1 · 21.4 ms

  1. 1. https://pigfox.com/refunds GOLD 0.521

    Skip to content Refund Policy What happens to your money, stated before you spend it. Last updated 2026-08-16. The position Purchases are not refundable. Both products are digital and are delivered immediately: an analysis credit is spendable the moment it …

  2. 2. https://pigfox.com/refunds GOLD 0.452

    buy: once performance has begun with your express request and your acknowledgement that you lose the withdrawal right, the right is lost for the performance already delivered. Buying a credit pack or starting a subscription and immediately running an analysis …

  3. 3. https://pigfox.com/refunds GOLD 0.397

    period you already paid for, and then stops. We do not pro-rate a partly-used month or year. Deleting your account Deleting your account erases your credit ledger, so unspent credits are forfeited and are not refunded. Deleting an account also …

  4. 4. https://pigfox.com/demos/access-audit-lab 0.355

    to marketing-relevant records. Certify — this is intended Revoke this access Configuration MEDIUM Elevated access on an account that has not signed in recently Fennara Marlhollow — Sales Fennara Marlhollow has not signed in for 137 days, past the 90-day …

  5. 5. https://pigfox.com/demos/agent-gateway 0.333

    of $5,000.00. No approval is offered above this line; it is the amount nobody has agreed software may commit. The action touches 1069 things, over the hard ceiling of 500. $9,511.06 is over the $250.00 a person has to sign …

  6. 6. https://pigfox.com/demos/agent-gateway 0.319

    cannot be undone once it returns. The cost of a needless approval is a minute; the cost of a wrong one is permanent. Requested at 10:22 UTC, outside the change window of 02:00 UTC to 06:00 UTC, and it cannot …

  7. 7. https://pigfox.com/terms 0.315

    your own service; use the site to break any applicable law, or to attack any system you are not authorized to test. Accounts You are responsible for what happens under your account and for keeping access to your email address, …

  8. 8. https://pigfox.com/privacy 0.314

    if you do not want that collection, block those scripts in your browser or use an extension that does, because this site will not stop them for you. Deleting your account You can delete your account yourself, at any time, …

  9. 9. https://pigfox.com/refunds GOLD 0.314

    or no signals has done its job. That is a result, not a fault, and it is not grounds for a refund. Chargebacks If something looks wrong, contact us before your bank does. We would rather refund a genuine mistake …

  10. 10. https://pigfox.com/demos/agent-gateway 0.307

    is a minute; the cost of a wrong one is permanent. IRREVERSIBLE_ACTION Allowed Place purchase order purchasing-bot Asked to reorder consumables from catalog 12. vendors/consumables/201 · $174.21 · touches 7 · internal · 03:51 UTC No rule matched, so the …

The dashed rule sits after rank 5: everything below it was retrieved but falls outside the cut the hit rate and recall figures use.

q10 HIT Which outside companies receive the text I paste into the AI-backed tools?

Should have retrieved: https://pigfox.com/subprocessors

recall 1.000 · nDCG 0.246 · first gold passage at rank 2 · 19.2 ms

  1. 1. https://pigfox.com/ai-access-checker 0.510

    Skip to content AI-Access Checker Free Free to use — no account needed. Enter a URL to see which AI crawlers your robots.txt allows or blocks, whether you publish an llms.txt, and whether your content is readable without JavaScript. What …

  2. 2. https://pigfox.com/subprocessors GOLD 0.414

    Skip to content Subprocessors Every third party that receives data from pigfox.com, and exactly what reaches them. Last updated 2026-08-16. This list is generated from the services the code actually calls, so it cannot quietly fall behind what the site …

  3. 3. https://pigfox.com/tools-how-to 0.402

    SEO & marketing analyzers 21. AI-Access Checker What it is. Enter a URL and this tool reads your robots.txt to show which AI crawlers (the bots behind assistants like ChatGPT and Perplexity) you allow or block, checks for an llms.txt …

  4. 4. https://pigfox.com/demos/access-audit-lab 0.392

    no compliance opinion — that remains the act of an independent audit firm. Nobody here has looked at your systems. The tool performs the review work and produces the evidence; the organization achieves compliance; the opinion is somebody else's to …

  5. 5. https://pigfox.com/ 0.389

    Explorer Groth16 release, AI-arbitrated disputes. AI Parametric Insurance Language models adjudicating a claim, checkably. Deposit Dispute Models rule on evidence; the contract does the arithmetic. Bench Lab A benchmark that says where it was wrong. Judge Lab An eval that …

  6. 6. https://pigfox.com/demos/agent-gateway 0.360

    verdict, the reason codes, and a line for an approver. This is the artifact that leaves the page, so it carries the fictional-data disclosure and the boundary sentence with it — a reader holding the file has no page around …

  7. 7. https://pigfox.com/ 0.359

    Same rule on-chain: the arbiter's authority is whatever the contract grants it, and its reasoning is public record. Why Pigfox is different → What Pigfox delivers Engagements start with an outcome rather than a stack: an AI integration that earns …

  8. 8. https://pigfox.com/demos/access-audit-lab 0.353

    none — naming one that does not resolve would be worse than naming nothing. The rules and their control citations are internal/accessauditlab, the company is built by cmd/genaccessauditorg, the assessments by cmd/genaccessauditnarratives, and the browser engine is cmd/access-audit-lab-wasm. Those package …

  9. 9. https://pigfox.com/demos 0.353

    AI Lab A Go bot answers mentions in Pigfox's own Slack workspace with one model call, and this page mirrors what it said as it says it. The question, the reply, the model and what the call cost — without …

  10. 10. https://pigfox.com/automation 0.343

    a human stays in the loop wherever judgment actually matters. The engineering behind it → Production LLM Engineering in Go. What Pigfox automates for you. Bolt-on automation. Your existing stack stays. Automation runs beside it: inbox triage, lead enrichment, content …

The dashed rule sits after rank 5: everything below it was retrieved but falls outside the cut the hit rate and recall figures use.

q11 HIT What does the sandbox for running untrusted repositories actually guarantee?

Should have retrieved: https://pigfox.com/repo-quarantine

recall 1.000 · nDCG 0.990 · first gold passage at rank 1 · 27.2 ms

  1. 1. https://pigfox.com/repo-quarantine GOLD 0.670

    machine. The repository is cloned and run inside a VirtualBox VM that is reset to a known-clean snapshot after every single run, so nothing a repo does can persist and nothing it does can reach your host's disk. What it …

  2. 2. https://pigfox.com/repo-quarantine GOLD 0.629

    for Anyone who reads other people's code and would rather not run it on their laptop: engineers evaluating dependencies, reviewers opening takehome submissions, and researchers triaging samples that are probably boring and occasionally not. It is a small amount of …

  3. 3. https://pigfox.com/repo-quarantine GOLD 0.609

    Skip to content Open source repo-quarantine A host-side shell toolkit that clones and runs an untrusted repository inside a disposable VirtualBox VM, as a non-sudo runner user, and restores a clean-base snapshot after every run. One guarantee, honestly stated: your …

  4. 4. https://pigfox.com/repo-quarantine GOLD 0.578

    inside it can exfiltrate anything it can read. Therefore: nothing secret may ever enter the VM. No credentials, no tokens, no SSH keys, no .env, no private repositories. Treat everything you put inside as already published. A clean run proves …

  5. 5. https://pigfox.com/ai-engineering 0.567

    and payments. MCP orchestration. I wire operational tooling into my LLM workflows over MCP — for example Honeybadger for error monitoring — so the systems I build are observable and debuggable, not black boxes. LLM security. The Prompt-Injection & System-Prompt-Leak …

  6. 6. https://pigfox.com/repo-quarantine GOLD 0.545

    first two happen once; the third repeats for every repository you look at. Harden once. Run vm-harden.sh against a fresh VM. It creates the non-sudo runner user and strips the VM down to the state you want every run to …

  7. 7. https://pigfox.com/github-repos 0.469

    Skip to content GitHub Open-source repositories and downloadable tools from Pigfox. Repo Quarantine — Run an untrusted repository inside a disposable VirtualBox VM, so whatever the code does stays in a sandbox you throw away. llm-evalkit — Evaluation harness for …

  8. 8. https://pigfox.com/tools-how-to 0.408

    network once the module has loaded and it keeps working. Open the Delivery Forecast → Beyond the hosted tools Not everything Pigfox publishes runs in your browser. repo-quarantine is a downloadable shell toolkit rather than a hosted tool: it runs …

  9. 9. https://pigfox.com/demos/deposit-dispute 0.403

    contract, test or threshold was changed to make either pass. The size gate had measured the property-fuzzing harness against EIP-170, a limit that binds only deployed bytecode, while the harness is built in process and deployed nowhere. The coverage gate …

  10. 10. https://pigfox.com/how-smart-contracts-are-tested 0.400

    Skip to content How the smart contracts are tested. Every contract behind the Pigfox demos passes the same gates before it reaches Base Sepolia: two linters, unit and invariant tests, static analysis, two independent property fuzzers, and a 100% coverage …

The dashed rule sits after rank 5: everything below it was retrieved but falls outside the cut the hit rate and recall figures use.

q12 HIT Why does a CPU profile report almost nothing for a workload whose producer is stalling?

Should have retrieved: https://pigfox.com/demos/pprof-lab

recall 1.000 · nDCG 1.000 · first gold passage at rank 1 · 24.3 ms

  1. 1. https://pigfox.com/demos/pprof-lab GOLD 0.682

    Skip to content pprof Lab The same bounded queue as the backpressure lab, profiled offline and compared two captures at a time. The finding. Under the blocking policy, Go's block profiler attributes the producer's stall to a named application function …

  2. 2. https://pigfox.com/demos/pprof-lab GOLD 0.642

    what is on-CPU at that instant. This workload is asleep almost all of the time, so a five-second run yields only a few dozen samples. The limitation is the sample count — you cannot meaningfully diff two trees built from …

  3. 3. https://pigfox.com/demos/pprof-lab GOLD 0.602

    because it is nearly empty, which is the result. Width is on-CPU time (seconds). Total: Block profile — time spent blocked on a synchronization primitive, in seconds. This is not CPU time. Width is block delay (seconds). Total: CPU profile …

  4. 4. https://pigfox.com/demos/pprof-lab GOLD 0.590

    the worker pool slow rather than the producer fast. Compare two captures Left capture Below capacity — 30/s into an 80/s pool Overload, block the producer Overload, drop the newest Overload, drop the oldest Worker sweep — 1 worker Worker …

  5. 5. https://pigfox.com/demos/pprof-lab GOLD 0.581

    and checked in as JSON. The harness is pigfox/pprof-lab at 56de858, and make capture regenerates the whole set in about forty seconds. Where these numbers came from Machine 8 logical CPUs, x86-64 Linux Go go1.26.5 (linux/amd64) GOMAXPROCS 8 GOGC 100 …

  6. 6. https://pigfox.com/demos/pprof-lab GOLD 0.558

    for itself, are measured by completely unrelated mechanisms. Across the eight captures below they agree to within 0.6%. Every other figure on this page rests on that. Nothing on this page touches a chain. These are recorded captures, not a …

  7. 7. https://pigfox.com/demos/pprof-lab GOLD 0.524

    Across the committed grid rows, all three overload policies completed exactly 392 messages. On these captures the policy decided who paid for the overload rather than how much work got done. That is what these eight runs show — it …

  8. 8. https://pigfox.com/demos 0.502

    depots among sixteen real Aegean ports, solved two ways in your browser. Max-coverage greedy is submodular, so it is proven to reach 1 − 1/e of the optimum; the p-median heuristic beside it carries no guarantee at all — and …

  9. 9. https://pigfox.com/services 0.489

    handled rather than discovered later. Diagnosis of an existing system. Profiling, tracing and reading the code you already have, to say plainly where the time and the risk are. The evidence The strongest thing I can offer is not a …

  10. 10. https://pigfox.com/demos/goroutine-visualizer 0.484

    as you are looking at it. What is running here, and what comes next Everything on this page is live. A producer, one instrumented channel and a consumer are running as real goroutines in this server right now: you can …

The dashed rule sits after rank 5: everything below it was retrieved but falls outside the cut the hit rate and recall figures use.

q13 MISS What academic degree does the founder hold, and which university awarded it?

Should have retrieved: https://pigfox.com/about

recall 0.000 · nDCG 0.000 · no gold passage in the top 10 · 23.1 ms

  1. 1. https://pigfox.com/demos/unselect2 0.226

    on this page touches a chain. City Greater Metropolitan Los Angeles New York Chicago Miami Seattle Occupation Engineering Healthcare Finance Clerical Legal Education High School Associate Degree Bachelors Degree Masters Degree Ph.D Region (optional) US — Pacific US — Atlantic …

  2. 2. https://pigfox.com/demos/access-audit-lab 0.174

    none — naming one that does not resolve would be worse than naming nothing. The rules and their control citations are internal/accessauditlab, the company is built by cmd/genaccessauditorg, the assessments by cmd/genaccessauditnarratives, and the browser engine is cmd/access-audit-lab-wasm. Those package …

  3. 3. https://pigfox.com/demos/mmr-log 0.147

    this program still holds. A freeze can only come after that upload, never before, and getting that order wrong is what left the two oldest versions unable to gain a record they can now never have. There is no third-party …

  4. 4. https://pigfox.com/the-pigfox 0.139

    Skip to content How Did Pigfox Come To Be? The critter was first sighted in Venice Beach, CA on a grey and dreary Sunday, sometime in the beginning of this millennium. Pigfox's founder, Peter, in addition to a friend, went …

  5. 5. https://pigfox.com/tools-how-to 0.130

    person's full name (for example, Jane Smith). In Claimed role, type the role they say they have — for example recruiter, VC partner, hiring manager, or a company name. Optionally, attach a profile photo they sent you. This runs a …

  6. 6. https://pigfox.com/demos/orbit-lab 0.122

    at any point. Coastlines from Natural Earth at 1:110m, public domain, decimated to the tolerance recorded in the artifact's own header and reported below the globe. Positions are as accurate as the element sets behind them, and those were retrieved …

  7. 7. https://pigfox.com/demos/rust-move-lab 0.120

    Skip to content Rust Move Lab Who owns this value? Three rejections from a real compiler, each beside the Go program that runs happily and shares the memory instead. The finding. Rust answers the ownership question by refusing to let …

  8. 8. https://pigfox.com/demos/mmr-log 0.117

    Hqc3B9Hac779qpF944Ns1yVjjC9dYtAegS7Ut3kmrW3G, which this program owns and allocated itself. Both are live, and neither is spare: the explorer reads the metadata for a transaction's own page and the Anchor account for the address history list, so dropping either one leaves half …

  9. 9. https://pigfox.com/ 0.115

    contracts, tested and fuzzed before they ship Go/Golang APIs & backend systems APIs & Proxies Web sites & applications Rebuilding or upgrading outdated business systems WASM applications Organic Search Engine Optimization Practical tools — OSINT and security checks, SEO and …

  10. 10. https://pigfox.com/demos/mmr-log 0.112

    account BxrToumKoEDmnCbVfXSbr8Vz9U6rVU9YWYAAhHxx5Jv9. v1's upgrade authority was revoked BEFORE it carried a security.txt and before its build was verified. An immutable program can gain neither: there is no upgrade path to add a contact record, and a verified build must be …

The dashed rule sits after rank 5: everything below it was retrieved but falls outside the cut the hit rate and recall figures use.

q14 MISS Why is this whole site written in a single language, and what makes that language a good fit for the work?

Should have retrieved: https://pigfox.com/golang

recall 0.000 · nDCG 0.000 · no gold passage in the top 10 · 27.4 ms

  1. 1. https://pigfox.com/ai-engineering 0.317

    languages — Playwright-verified across all ten subdomains. Autonomous content pipeline MarketVerdict's blog runs itself. A scheduled job discovers source material, an LLM drafts a post, and the draft goes out for approval by email — approve, edit, or reject in …

  2. 2. https://pigfox.com/demos/langdag-lab 0.297

    What comes back is appended below the node you chose, tagged as live, and it is gone when you reload — nothing you write is stored and no captured tree is changed by it. To be exact about what is …

  3. 3. https://pigfox.com/demos/ai-parametric-insurance 0.277

    Skip to content AI Parametric Insurance Parametric means the payout is triggered by a defined condition, not by an assessed loss. That distinction is the whole reason a language model can be in this loop at all: “did the condition …

  4. 4. https://pigfox.com/ai-engineering 0.269

    a good example precisely because the LLM is one stage in a grounded pipeline, not the whole product. Here's the flow: What makes it reliable: Grounded, not guessed. Location and competitor data come from live map providers with dual-provider fallbacks …

  5. 5. https://pigfox.com/tools/go/channel-deadlock 0.261

    Skip to content Go Channel Deadlock Visualizer Write a few goroutines and channels in the grammar below and this explores every interleaving of them. A deadlock comes with the shortest path to it and what each goroutine is stuck on; …

  6. 6. https://pigfox.com/subprocessors 0.255

    Skip to content Subprocessors Every third party that receives data from pigfox.com, and exactly what reaches them. Last updated 2026-08-16. This list is generated from the services the code actually calls, so it cannot quietly fall behind what the site …

  7. 7. https://pigfox.com/demos/langdag-lab 0.250

    captured against langdag v0.10.0. That commit is checkable rather than merely cited. The harness has a verify step that re-runs every capture and compares it byte for byte with what is committed, which is what turns a JSON file in …

  8. 8. https://pigfox.com/demos/langdag-lab 0.244

    the model resolution are exactly as the library wrote them. The risk in a normalization like that is doing too much rather than too little — one that flattened enough would make any two conversations compare equal while still reporting …

  9. 9. https://pigfox.com/demos/langdag-lab 0.233

    Skip to content LangDAG Lab An LLM conversation is a tree, not a list. Three captured ones: a model asking for a tool, one answer continued two ways, and a routing policy whose preferred deployment stops answering halfway through. The …

  10. 10. https://pigfox.com/automation 0.226

    Skip to content This site runs itself. Pigfox publishes a researched article every day at 08:00 Pacific with no one at the keyboard. The same engineering is for hire: automation that runs parts of your business while you sleep. The …

The dashed rule sits after rank 5: everything below it was retrieved but falls outside the cut the hit rate and recall figures use.

q15 MISS How does an AI client call a tool it did not ship with, and what address does it connect to?

Should have retrieved: https://pigfox.com/mcp

recall 0.000 · nDCG 0.000 · no gold passage in the top 10 · 25.2 ms

  1. 1. https://pigfox.com/demos/agent-gateway 0.469

    May name resources beginning: services/, clusters/ crm-writer (CRM agent) Purpose: Keeps customer records current from support conversations and billing events. May request: crm.record.update, crm.record.delete May name resources beginning: crm/ support-mailer (Support mailer) Purpose: Sends the follow-up a support conversation ended …

  2. 2. https://pigfox.com/demos/agent-gateway 0.454

    deploy.release, deploy.rollback, infra.scale May name resources beginning: services/, clusters/ crm-writer (CRM agent) Purpose: Keeps customer records current from support conversations and billing events. May request: crm.record.update, crm.record.delete May name resources beginning: crm/ support-mailer (Support mailer) Purpose: Sends the follow-up a …

  3. 3. https://pigfox.com/demos 0.447

    AI Lab A Go bot answers mentions in Pigfox's own Slack workspace with one model call, and this page mirrors what it said as it says it. The question, the reply, the model and what the call cost — without …

  4. 4. https://pigfox.com/ 0.415

    Explorer Groth16 release, AI-arbitrated disputes. AI Parametric Insurance Language models adjudicating a claim, checkably. Deposit Dispute Models rule on evidence; the contract does the arithmetic. Bench Lab A benchmark that says where it was wrong. Judge Lab An eval that …

  5. 5. https://pigfox.com/subprocessors 0.408

    Skip to content Subprocessors Every third party that receives data from pigfox.com, and exactly what reaches them. Last updated 2026-08-16. This list is generated from the services the code actually calls, so it cannot quietly fall behind what the site …

  6. 6. https://pigfox.com/demos/agent-gateway 0.405

    anyone. The gate reads the shape of the action — who asked, for what, how much, how wide, how sensitive, at what hour. It never reads the agent's own account of what it is doing, because a gate that graded …

  7. 7. https://pigfox.com/demos/ai-parametric-insurance 0.404

    gate, and its integration suite deploys these contracts into an in-process EVM it builds from source and drives the whole path — including a disagreement — through the same code the production binary uses. The full determinism procedure, including the …

  8. 8. https://pigfox.com/demos/agent-gateway 0.404

    they act on. Every action an agent wants to take is submitted here first, and you decide one of exactly three things about each one: ALLOW - the action proceeds with no person involved. REQUIRE_APPROVAL - the action is held …

  9. 9. https://pigfox.com/tools-how-to 0.404

    had a public certificate. Look for anything you didn't expect — old test sites, admin panels, or services you thought were retired — and make sure each is patched, protected, or shut down. Good to know. This is passive: it …

  10. 10. https://pigfox.com/ 0.403

    Same rule on-chain: the arbiter's authority is whatever the contract grants it, and its reasoning is public record. Why Pigfox is different → What Pigfox delivers Engagements start with an outcome rather than a stack: an AI integration that earns …

The dashed rule sits after rank 5: everything below it was retrieved but falls outside the cut the hit rate and recall figures use.

q16 HIT What checks does a smart contract have to pass here before it is deployed to a test network?

Should have retrieved: https://pigfox.com/how-smart-contracts-are-tested

recall 1.000 · nDCG 0.801 · first gold passage at rank 1 · 21.8 ms

  1. 1. https://pigfox.com/how-smart-contracts-are-tested GOLD 0.491

    a contract. A failing test or violated property means the deployed contract is not demo-worthy. The source is fixed; a plain contract is redeployed and re-verified, and a contract behind an upgradeable proxy has its implementation upgraded rather than replaced …

  2. 2. https://pigfox.com/how-smart-contracts-are-tested GOLD 0.454

    property set and reported a smaller green run. The count is declared as a literal in Solidity, and four independent things must agree with it before a build passes. Live chain. Every demo is deployed to Base Sepolia and verified …

  3. 3. https://pigfox.com/how-smart-contracts-are-tested GOLD 0.443

    code is excluded by name, printed on every run, never folded into a percentage. The check fails on a report it cannot parse, so a format change cannot pass by default; and it fails on an exclusion that no longer …

  4. 4. https://pigfox.com/how-smart-contracts-are-tested GOLD 0.365

    commit that produced them, so a displayed number is true by construction rather than true until the repository changes — is the work this table is a placeholder for. Slither runs on every repository at fail-on: low, against one shared …

  5. 5. https://pigfox.com/how-smart-contracts-are-tested GOLD 0.356

    required check that the rest report into. Grouped by what they catch: forge lint and Solhint. The pipeline’s first stage, and first because it is the cheapest: no fuzzing, no coverage run and no chain, so a style or natspec …

  6. 6. https://pigfox.com/how-smart-contracts-are-tested GOLD 0.353

    chain, against throwaway contracts the drill deploys itself so it structurally cannot touch the real proxy, then broadcast, then verified on-chain. Coverage without asterisks. The gate requires 100% line, statement, branch, and function coverage on every file under src/. The …

  7. 7. https://pigfox.com/how-smart-contracts-are-tested GOLD 0.336

    a number that rots, and rots quietly. So the figures live where they are produced and where they are enforced: in each repository, on every push. The pipeline asserts them there — the coverage gate fails below 100%, the property …

  8. 8. https://pigfox.com/website-legit-check 0.331

    Skip to content Website Legit Check Paid 1 free analysis per IP, then credits or a subscription. Watch: 30 seconds Check a website's trust signals in one pass — domain age, TLS certificate, redirects, email-security DNS, and security headers — …

  9. 9. https://pigfox.com/demos/oracle-health-explorer 0.328

    of the threshold on every run rather than waiting to stumble onto it, and a companion test demonstrates that an interior sample does not catch the substitution. A check that is still copied, and cannot fire Older Chainlink documentation suggested …

  10. 10. https://pigfox.com/tools-how-to 0.328

    signals, each with a category, a severity badge, and what was observed; any checks that couldn't run are listed separately. Weigh the signals together — one weak header isn't much, but a very young domain plus several weak signals is …

The dashed rule sits after rank 5: everything below it was retrieved but falls outside the cut the hit rate and recall figures use.

q17 HIT When a delivery is contested rather than confirmed, how does the escrow decide who gets the money?

Should have retrieved: https://pigfox.com/demos/zk-escrow

recall 1.000 · nDCG 0.802 · first gold passage at rank 1 · 25.0 ms

  1. 1. https://pigfox.com/demos/zk-escrow GOLD 0.581

    back from the chain, so the value sent always matches to the wei. Buyer only, and only while the escrow is Created. Fund Release with a proof Escrow id Nullifier Proof Anyone may send this transaction: the proof authorises the …

  2. 2. https://pigfox.com/demos/zk-escrow GOLD 0.516

    total pending withdrawals, plus anything still held in an open escrow. Escrows Featured The escrows worth looking at first. Each is fetched by id, so it stays here however many escrows the contract accumulates. # State Amount Buyer Seller Arbiter …

  3. 3. https://pigfox.com/demos/zk-escrow GOLD 0.460

    Skip to content ZK Escrow Explorer A live demo of the Pigfox zk-escrow contracts on Base Sepolia. Funds are released either by a Groth16 zero-knowledge proof of a delivery secret, or — when delivery is contested — by an AI …

  4. 4. https://pigfox.com/demos/zk-escrow GOLD 0.442

    escrow. Also emitted on chain in the clear. Submit evidence Withdraw your balance Settlements are never pushed to you. Every outcome credits a pull-payment balance that you withdraw yourself. Withdraw Connect a wallet to see your balance. Every check on …

  5. 5. https://pigfox.com/ 0.440

    delivery secret — the secret is never revealed, not even to the contract. When delivery is contested, the case goes to an AI arbiter: it reads the evidence both parties published on-chain, rules, and its full reasoning is written back …

  6. 6. https://pigfox.com/demos/zk-escrow GOLD 0.432

    contract behaved under real traffic. An earlier run assigned several disputes an arbiter address that had no private key anywhere — a keyless arbiter can never call resolveDispute, so nine escrows sat stranded in Disputed, their funds frozen (though never …

  7. 7. https://pigfox.com/demos/agent-gateway 0.402

    · touches 5 · internal · 03:56 UTC No rule matched, so the action proceeds unattended. Blocked Issue refund support-mailer Asked to issue a goodwill refund on order 6620. payments/refund/429 · $38,230.09 · touches 1112 · internal · 10:45 UTC …

  8. 8. https://pigfox.com/github-repos 0.402

    and nine properties fuzzed by three engines. zk-escrow — Upgradeable escrow settled by a zero-knowledge proof of delivery, with an AI arbiter ruling on disputes — on Base Sepolia. ai-parametric-insurance — Parametric insurance on Base Sepolia whose claims are adjudicated …

  9. 9. https://pigfox.com/demos/deposit-dispute 0.385

    while presenting them as if they were. The burden sits with the landlord. NotEstablished is the contract's zero value and there is deliberately no abstain state. An item the evidence does not carry is an item the tenant keeps the …

  10. 10. https://pigfox.com/demos/agent-gateway 0.371

    is a minute; the cost of a wrong one is permanent. IRREVERSIBLE_ACTION Blocked Issue refund support-mailer Asked to issue a goodwill refund on order 9903. payments/refund/729 · $17,800.51 · touches 414 · internal · 19:30 UTC Support mailer is not …

The dashed rule sits after rank 5: everything below it was retrieved but falls outside the cut the hit rate and recall figures use.

q18 HIT Why does throwing away the whole pipeline every cycle avoid the problem of stale data?

Should have retrieved: https://pigfox.com/demos/etl-lab

recall 1.000 · nDCG 0.668 · first gold passage at rank 1 · 22.6 ms

  1. 1. https://pigfox.com/demos/etl-lab GOLD 0.713

    Skip to content ETL Lab The whole pipeline is rebuilt from nothing every cycle, and everything it built is thrown away at the end of one. Poll, extract, transform, load, detect, execute, flush — then again, from an empty graph. …

  2. 2. https://pigfox.com/demos/etl-lab GOLD 0.510

    produced what is below, not a copy of it. Run it Seed Same seed, same run. Exchanges Pairs each Speed 1× Start Stop Press start to open a run. Poll Synthetic venues emit raw quotes. — raw rows ETL Parse …

  3. 3. https://pigfox.com/demos/etl-lab GOLD 0.467

    ones such a book actually carries. Every fill is simulated — the profit and loss is arithmetic over those generated rates against a fixed 1000-unit notional, with seeded slippage of up to 85% of the theoretical edge subtracted from it. …

  4. 4. https://pigfox.com/ai-engineering 0.449

    handlers, and a dead-letter path — so a timeout or a rate-limit retries and recovers on its own instead of failing a user. Human-in-the-loop gates. Autonomous is not the same as unsupervised. My pipelines pause for a person's approval before …

  5. 5. https://pigfox.com/demos/backpressure-lab 0.449

    Skip to content Backpressure Lab A bounded queue you can push past its own capacity, and three different answers to what should happen next. Every queue between two components has a capacity. The interesting question is never how big it …

  6. 6. https://pigfox.com/demos/langdag-lab 0.447

    user Where should I start optimizing a service that is slow under load? n1 assistant anthropic Start with the read path. A cache in front of the slowest query buys the most headroom for the least risk, because a stale …

  7. 7. https://pigfox.com/demos/reconcile-lab 0.415

    Skip to content Reconcile Lab Two writers reach for one object. The store checks the version each write was read at and refuses the one that lost, and what the loop does next is the only thing this page changes. …

  8. 8. https://pigfox.com/demos/reconcile-lab 0.408

    retry at once against retry with backoff. Both get slower and both waste far more, because every competing write now lands on the object you are reconciling — but retrying at once stays the quicker way back, and buys it …

  9. 9. https://pigfox.com/demos/etl-lab GOLD 0.406

    no token uses, a missing side, or a symbol this universe does not contain. Each cause is counted separately, because “bad row” is not a thing anyone can act on. Collapsed counts pairs a second venue had already quoted. They …

  10. 10. https://pigfox.com/ai-engineering 0.398

    Skip to content Production LLM Engineering in Go I'm Peter Sjolin. I design and run LLM systems in production — solo, end to end. The hard part of AI engineering isn't calling a model; it's making it reliable: job queues …

The dashed rule sits after rank 5: everything below it was retrieved but falls outside the cut the hit rate and recall figures use.

q19 HIT Do I need a password to sign in, and what is stored about my account?

Should have retrieved: https://pigfox.com/privacy, https://pigfox.com/why-pigfox

recall 0.500 · nDCG 0.220 · first gold passage at rank 1 · 22.4 ms

  1. 1. https://pigfox.com/privacy GOLD 0.434

    Skip to content Privacy Policy This describes what Pigfox LLC actually does with data on pigfox.com today — the retention windows are the ones the code enforces, not aspirations. Last updated 2026-08-16. Who we are Pigfox LLC is a Nevada …

  2. 2. https://pigfox.com/breach-check 0.400

    was part of a data leak from some service. It doesn't mean you did anything wrong — but you should change any reused password and enable two-factor authentication on important accounts. My email wasn't found — am I safe? It's …

  3. 3. https://pigfox.com/breach-check 0.399

    below still applies. What this breach check shows Which breaches your address appears in, by name and the site involved. When each breach happened. What was exposed — for example passwords, phone numbers, or addresses. What to do — concrete …

  4. 4. https://pigfox.com/tools-how-to 0.380

    never connects to the domain's servers. Try the DNS Security Check → 14. Data-Breach & Exposure Check What it is. Companies get hacked, and when they do, their users' email addresses and passwords often end up in public leaks. The …

  5. 5. https://pigfox.com/terms 0.342

    reporting agency. Nothing here may be used to make decisions about credit, insurance, housing, or employment eligibility. They read public information and third-party data that may be incomplete, stale, or simply wrong. A signal can fire on an innocent person, …

  6. 6. https://pigfox.com/tools-how-to 0.313

    turns up, don't panic — it usually means a site you signed up for was breached, not that you did anything wrong. Change any password you reused, starting with your email and bank, and turn on two-factor authentication. If nothing …

  7. 7. https://pigfox.com/demos/access-audit-lab 0.311

    could currently view sensitive financial data without the account owner's recent involvement. This combination warrants investigation to confirm whether the access is still needed or should be revoked. Suggested next step: Reach out to Fennara Marlhollow's manager or the account …

  8. 8. https://pigfox.com/demos/access-audit-lab 0.288

    to marketing-relevant records. Certify — this is intended Revoke this access Configuration MEDIUM Elevated access on an account that has not signed in recently Fennara Marlhollow — Sales Fennara Marlhollow has not signed in for 137 days, past the 90-day …

  9. 9. https://pigfox.com/breach-check 0.288

    Skip to content Free Data-Breach & Exposure Check Free Free to use — no account needed. Check whether your email address appears in known data breaches — the security incidents where a company's user data was stolen and leaked. You'll …

  10. 10. https://pigfox.com/username-check 0.271

    Skip to content Free Username Checker — Across Platforms Free Free to use — no account needed. Enter a username to see where it exists across a curated set of platforms, with a direct link to any profile found. It's …

The dashed rule sits after rank 5: everything below it was retrieved but falls outside the cut the hit rate and recall figures use.

q20 MISS What is the company's registered business address?

Should have retrieved: https://pigfox.com/privacy, https://pigfox.com/refunds, https://pigfox.com/subprocessors, https://pigfox.com/terms

recall 0.000 · nDCG 0.140 · first gold passage at rank 7 · 23.5 ms

A near miss: a gold page was retrieved, at rank 7, just below the cut of 5. That is a different failure from finding nothing relevant at all, and this page does not print the same thing for both.

  1. 1. https://pigfox.com/job-offer-check 0.336

    Skip to content Job-Offer Legitimacy Check AI · Paid 1 free analysis per IP, then credits or a subscription. Watch: 30 seconds Paste a job offer or recruiter message and get neutral signals worth investigating — drawn from the message …

  2. 2. https://pigfox.com/contact 0.332

    Skip to content Have A Question About Pigfox? Get In Touch Today! If you want to know more about Pigfox or just feel like saying hello, please use the contact form below for a prompt reply! Questions are always welcome, …

  3. 3. https://pigfox.com/domain-recon 0.312

    Skip to content Free Domain & Subdomain Recon Free Free to use — no account needed. Enter a domain to look up its public DNS records (A, AAAA, MX, NS, TXT) and discover its subdomains from certificate-transparency logs. It's free, …

  4. 4. https://pigfox.com/demos/access-audit-lab 0.311

    none — naming one that does not resolve would be worse than naming nothing. The rules and their control citations are internal/accessauditlab, the company is built by cmd/genaccessauditorg, the assessments by cmd/genaccessauditnarratives, and the browser engine is cmd/access-audit-lab-wasm. Those package …

  5. 5. https://pigfox.com/demos/sba-lab 0.291

    26.60% Live Oak Banking Company 18,078 7,538 3.36% Celtic Bank Corporation 15,087 7,301 21.59% Newtek Bank, National Association 13,758 457 0.44% BayFirst National Bank 13,751 3,466 34.62% BMO Bank National Association 13,575 10,447 11.06% CDC Small Business Finance Corp. 13,472 …

  6. 6. https://pigfox.com/about 0.290

    life. It includes a Go DEX arbitrage bot built in 2022, whose stateless mapping pipeline was load-tested at 37M+ token mappings per request cycle; that design runs as a live demo at ETL Lab. Pigfox's services span many sectors, including …

  7. 7. https://pigfox.com/privacy GOLD 0.289

    Skip to content Privacy Policy This describes what Pigfox LLC actually does with data on pigfox.com today — the retention windows are the ones the code enforces, not aspirations. Last updated 2026-08-16. Who we are Pigfox LLC is a Nevada …

  8. 8. https://pigfox.com/web-sites 0.286

    at a competitive price. The other is concierge-level service – Pigfox is always in reach, and you'll never be out of the loop on how your project is coming along. With Pigfox, you will never feel like just a “client” …

  9. 9. https://pigfox.com/terms GOLD 0.271

    on the Subprocessors page. Changes We may update these terms; the date at the top changes when we do. Continuing to use the site after a change means you accept the updated terms. Governing law These terms are governed by …

  10. 10. https://pigfox.com/demos/rwa-tokenization 0.266

    The document hash and URI are recorded on chain; the URI is operator-supplied metadata, rendered as plain text. # Status Document hash Metadata URI 0 Active 0x8ced…33bc ipfs://bafy-acme-tower-manhattan-office-deed 1 Active 0x00c2…4e12 ipfs://bafy-lakeside-logistics-warehouse-title 2 Active 0xf59b…94bb ipfs://bafy-solar-farm-nevada-ppa-bundle 3 Active 0x2aa4…4990 ipfs://bafy-vineyard-estate-napa-land-grant …

The dashed rule sits after rank 5: everything below it was retrieved but falls outside the cut the hit rate and recall figures use.

What this pipeline does not do

Stated as a boundary rather than left as a gap, because each of these would move the figures above and none of them is in the measurement. There is no hybrid search — retrieval is dense only, with no keyword channel to fall back on, which is why a question that avoids a page's distinctive vocabulary can miss it entirely. There is no re-ranking: the vector search result is the final order. There is no graph or structured retrieval and no query rewriting. Relevance is judged at the level of a page, not a passage — a chunk counts when it comes from a page the question set names — because passage-level labels would have to be re-authored by hand every time the chunk width moved.

Answer generation is optional and never graded. The repository can generate a grounded answer per question when a key is present in the operator's environment, but nothing on this page depends on it and no figure here reports on it: grading a generated answer needs a human or a judge model, which is a different experiment.

Run it yourself

The repository is a Docker Compose stack — Postgres with pgvector, a Python embedding sidecar with the model baked into the image, and a Go API. One command brings it up and a second reproduces the evaluation. It needs no API key and no account.

git clone https://github.com/pigfox/rag-lab
cd rag-lab
./run.sh
./evaluate.sh

View rag-lab on GitHub

This run retrieved at 2026-09-02T17:34:04Z and was graded at 2026-09-02T17:34:06Z, over an index built at 2026-09-02T17:31:04Z.