Everything below is running right now — a tool you can press, a demo you can click, or a repository you can read. No mockups, no "coming soon". If a number appears on this page, it is one that was measured, and the card says how.
Each one runs entirely in your browser. There is no server to send your file to — open the network tab on any of them and watch: after that tool loads, zero requests. That is architecture, not a privacy policy. (This index page carries ordinary web analytics; the tools themselves carry none, which is why the claim is about them and not about here.)
Every other tool here asks you to upload something first. This one asks for nothing — press a button and it builds you a working Excel tracker where unbilled jobs and invoices past 60 days flag themselves.
A real .xlsx assembled in your browser with no library and no upload — verified to open with formulas intact and amounts stored as numbers, not text.
One drop zone for any business export — Excel (.xlsx) works directly, no save-as-CSV step. It works out what the file is from its columns, runs every check that applies, and reconciles two files against each other if you drop a pair. The Excel file is parsed in your browser with no library and no upload — nobody else does that.
On its sample pair it returns 8 findings with zero configuration and puts a number on the damage: $1,240 of finished work that was never invoiced. See a real report before uploading anything.
Drop any CSV and get an honest report: columns that carry no information, status fields that never vary, mixed date formats, duplicate rows. Then download the file with the judgment-free problems already fixed.
On its sample: 6 findings down to 2 after the automatic fix — and the two that remain are exactly the ones a script must not touch.
Two exports every business already has — the work log and the invoices — lined up so only the mismatches show: finished jobs nobody billed, invoices sitting unpaid past 60 days.
Matches on exact ID only, and says so on the report. Nothing fuzzy, nothing guessed.
Acme Roofing. Acme Roofing LLC. acme roofing inc. Three rows, one customer — inflating your count, splitting revenue history, and sending the mail merge three times.
Every group states why it grouped: identical after ignoring case, punctuation and legal suffixes, or a near-miss typo worth a human glance.
For anyone who runs on a roster — senior living, clinics, restaurants, facilities. Finds days a role is uncovered, staff crossing 40 hours, roles only one person can fill, and turnarounds too short to be safe.
Reads the columns itself and shows which ones it used. Deliberately never proposes who should work when — that is the manager's call, not a CSV's.
Working Google Sheets with the formulas already in them. The copy link is on the page — there is nothing to submit and no follow-up sequence.
For recruiters and staffing agencies: log placements and it works out perm fees, contract margins, owner/sourcer splits, each recruiter's draw balance, and which deals are still inside their guarantee window and could be clawed back.
Five tabs with the formulas already in them and a Read Me explaining each one. Every formula was evaluated before shipping — a contract deal at $78/$52 across 36h × 26 weeks returns $24,336 gross and $17,035 to the owner at a 70% split, and all three sample deals correctly flag AT RISK inside their guarantee.
A one-board view of the numbers an owner actually checks — revenue, jobs, what is outstanding — already wired with the formulas, so you fill in the rows and it keeps itself current.
Copy it straight from the page. No email, no signup, no follow-up sequence — the copy link is right there, and you can take it and never speak to me.
Categories already matching a normal small-business chart of accounts, currency formatting done, header row frozen. Open it and start typing rather than building it first.
Same deal: a copy link on the page, nothing to submit. If you would rather I set it up on your real numbers, that offer exists separately and is also free.
For an export that arrived messy — inconsistent names, mixed date formats, duplicate rows. The template gives it a structure that stays clean as you add to it.
Copy link on the page, no email required.
The whole method in one page: what gets automated, what deliberately does not, and what each rung costs. Sensor → formulas → AI narrative → human approval → audit trail — with a live console showing how a chase email's tone is derived from invoice age rather than chosen.
Carries an honesty ledger that opens by stating there are zero paying clients, and status chips that say RUNNING (on my own business) rather than LIVE, because no customer is on it yet.
A real dashboard on sample data — revenue, receivables, weekly reports. Hover the chart, open any week's report, or press Draft chase on an overdue invoice and watch it write the email and then stop. Or drop your own Excel file or CSV on it and it rebuilds on your numbers.
Your file is read in your browser — .xlsx works as-is, nothing uploads, and the page makes zero network requests while it parses. On real data it drops every illustrative figure rather than inventing a finding about your business.
The flagship retainer: every Monday an AI reads your live numbers, writes a plain-English summary, and drafts chase emails for money you're owed.
Every draft waits for a human approval before anything sends. That gate is the product.
A live commission dashboard for a staffing desk that settles splits, draws and fall-off clawbacks by rule instead of by argument.
Built for a vertical where the math is the whole disagreement.
Three finished builds you can read and then download as a working .xlsx. Every number in them is invented — they show the shape of the output, not a client and not a result. Each page says so above its figures rather than in a footnote.
A contractor logs quoted price against real labour and materials, and every job shows its margin with the losers flagged. Read the finding, then take the workbook.
On invented sample data: $44,600 booked, $5,550 kept — and two jobs underwater by $1,100. Every figure on the page is stated as invented, and the workbook recomputes them by formula.
Rent in, expenses out, management fee, net payout per owner — with a flag and a unit-level reason on any owner whose cheque dropped.
On invented sample data: four of five owners down, and one negative because a $3,200 roof repair landed on a $1,750 unit. The page states plainly that the reason field assumes a single culprit, which is its real limit.
Two systems that should agree — the schedule and the invoices — matched row by row so only the disagreements show. Silence means everything matched.
On invented sample data: $165 in one week delivered and never invoiced, plus one invoice charged against a booking that does not exist. Matches on exact ID only and says so.
The recurring service: we watch your automations weekly and send an evidence report. Checks the two things nobody else does — whether the data inside a fresh file is actually current, and whether a report has quietly stopped changing.
Sold on a true story: our own trading digest served Friday's data under Monday's timestamp the day the service page was built. The embedded report is that catch, verbatim.
Drop two versions of the same report — last month's export and this month's — and see exactly what changed: which cells moved, which rows came and went, and whether a formula was quietly replaced by a typed number.
That last one is the finding it exists for. Somebody pastes a value over a calculation and the cell stops tracking its inputs, reporting the same answer forever while looking perfect. It is also built to survive the thing that makes most spreadsheet comparisons useless: rows are matched to each other BEFORE cells are compared, so inserting one row near the top does not report every row below it as changed. Reads the files in your browser; nothing is uploaded.
Drop a year of invoices and get a description of the business back: when the money actually arrives, how much of it rides on one client, whether the work repeats or has to be won again, and which regular customer has quietly stopped appearing.
It is built to refuse. Twelve months is one year, and one year cannot tell “busy every March” from “busy that March”, so it will not call a single year a season — it says so and asks for the second. A client counts as gone quiet only against its own rhythm, never for merely being quiet. Reads the file in your browser; nothing is uploaded.
Drop any Excel file and see how the workbook actually works: a map of which cells feed which, laid out by depth so raw inputs sit on the left and the deepest formulas on the right, with a plain-English note on anything that stands out.
The only tool here that shows you what you built rather than what is broken — the defects fall out of the picture as a consequence. Reads the file in your browser with no library and no upload; a two-sheet workbook maps in about 8ms.
An open-source reliability tool that finds signals carrying no information and jobs producing nothing: a check that fires on 100% of rows, a scheduled task that reports success while writing nothing, a branch that can never run.
Built after a real scheduled job discarded four days of data while reporting success every run. 136 tests, MIT, public — the page shows five ways to use it on your own stack.
A cost-governance gateway for telemetry. Metrics are billed on how many distinct label combinations you send, so one deploy that adds a user ID to a metric can multiply the bill overnight — and you find out weeks later on the invoice.
It refuses over-budget telemetry before it leaves your network, so the cost is never incurred rather than merely recorded as blocked. The large vendors bill runaway cardinality as a line item, which makes your mistake their revenue; this caps it instead, which is the one position they cannot copy. Sits in front of the backend you already pay for.
The always-on half of flatline. Point it at the folder where your exports already get saved and it checks each file the moment it changes — a column that is one value all the way down, a field empty on every row, a report that stopped moving weeks ago.
Every tool of this kind waits to be asked, which means the people who most need the check never run it. This one inverts the trigger. Wheels for both tools download straight from the page; it reads only, and nothing leaves the machine.
The primitive the others are built on. Wrap any scheduled command and it writes a signed, chained receipt of what the job claimed against what it actually produced — and exits non-zero when a tool reports success without touching its own output.
0.1.1 closes a gate that was passing silently: a 2.7MB file of 2019 dates satisfied a one-day freshness limit, because the date scan gives up above 2MB. The release notes lead with that, because it can refuse runs the previous version allowed.
Checks the four things between runs: does the output still exist, was it written recently, is the data inside it current, and has it quietly stopped changing.
Written the afternoon a digest of ours served Friday's data under Monday's timestamp. Its first production run caught the residue of that same incident — flagging its own author's fix as incomplete.
A chain of custody for work done by AI: what the model was allowed to read, whether that data was current, what it produced, which human signed it off, and whether it later turned out to be wrong.
Built because every company is now letting AI do real work and almost none can answer a board, an auditor or an insurer asking how they know it was right. Stale input means the model does not run at all; prompts and outputs are hashed, never stored.
Drop in any file and your browser hashes it and signs a receipt with an Ed25519 key it generates on your device. Later, drop the file back in — it tells you whether a single byte has changed, and hands you a public key so anyone else can check the signature too.
Hand-built on the Web Crypto API: no library, no account, no upload, and no shared secret. Verified by notarizing a report, changing one digit of the revenue figure, and watching it come back ALTERED with both hashes shown.
A running record of what actually gets built here, newest first. Every entry states how its claim was checked, because “we tested it” is not a test. It regenerates itself nightly, so it cannot quietly fall behind the work it describes.
The defects stay in — the dead CSS token that removed the keyboard focus ring, the check that could never fire, the gate that blocked a page for being right. A log with only wins is a brochure, and it would undercut the whole argument.
Every night our scheduled jobs write a signed receipt recording what each run claimed against what it actually produced. The receipts are hash-chained, and the chain head is published publicly so the record binds us too.
It includes the receipts where our own nightly audit produced nothing — kept deliberately, because a ledger holding only clean runs is one nobody should believe.
A national help directory — food, health care, mental-health and substance-use services — built on federal data (HRSA, SAMHSA, USDA, EPA/Hunger Free America) and refreshed automatically. Shelter and legal aid are not in it yet; it sends those needs to 211.
44,000+ live records year-round, more in summer while USDA meal sites are open. Its public health endpoint reports freshness per source, so a stale feed cannot hide behind a fresh one.
How the multi-tenant reporting pipeline behind Automated Workflow is actually assembled — n8n, Google Sheets, an AI narrative step, and a human approval gate.
Written for someone deciding whether this is real engineering or a wrapper.
Tell me the chore and I’ll rebuild a real piece of it on your own file, usually within a business day. Free, and yours to keep either way — there is no contract and no follow-up sequence.