Read this before anything else
This sample is the prototype running on local test data, not a live read of any vendor. The five names below (api-alpha, billing-beta, sdk-gamma, cli-delta, queue-epsilon) are fixtures we wrote. Nobody fetched OpenAI, Stripe, or anyone else for this page. A public board of about 25 real APIs is the product; picking those sources and confirming each vendor’s written terms still needs a person, and it has not been done.
We cannot backfill. A verdict needs two dated readings we actually took. If you start today, the earliest we can tell you anything is the second reading. A buyer who starts in March cannot ask in June what happened in April.
The first week after you pay is one reading and no verdict. That is not a trial we forgot to make useful. One reading cannot prove a change. We would rather say that now.
Public samplePrototype on fixtures, as of 13 Apr 2026
The engine ran 15 weekly probes from 5 Jan 2026 to 13 Apr 2026 against five fixture dependencies. It then stored the readings in a file, stopped, and reloaded them. Verdicts, the change log, and the forecast scores were identical after the reload. Detection matched the fixture timetable on every name.
The last two weekly reads on each name matched, so the board’s current verdict is unchanged. The product is the dated diffs underneath, not that last line.
accuracy: 75.0% (3/4 settled forecasts, 2 unsettled excluded) — counted from forecasts whose window had closed and whose outcome was read back from stored evidence. The two still-open forecasts are not in the score. Until a live board has settled forecasts, this line will read accuracy: unknown (no settled forecasts yet). We will not put a made-up number there.
| Dependency | Current verdict | Reads stored | Last change | What last moved |
|---|---|---|---|---|
| api-alpha | unchanged | 15 | 30 Mar 2026 | version 1.2.0 → 1.3.0; auth api_key → oauth |
| billing-beta | unchanged | 15 | 2 Feb 2026 | price_cents 1200 → 1400; tier flat → metered |
| cli-delta | unchanged | 4 | none | Unreadable until 23 Mar 2026, then four identical reads. Never flagged changed. |
| queue-epsilon | unchanged | 15 | 30 Mar 2026 | version 3.3 → 3.4 |
| sdk-gamma | unchanged | 15 | none | Frozen on 9.9.9 for the whole run. Certified unchanged, never changed. |
Every dated diff the engine logged9 changes, 0 extras
Each row is a field-level difference between two stored reads, with both dates. These are all the changes the engine recorded. sdk-gamma and cli-delta have none, on purpose.
| Dependency | Date | What moved |
|---|---|---|
| billing-beta | 19 Jan 2026 | price_cents: 1000 → 1200 |
| queue-epsilon | 26 Jan 2026 | version: 3.0 → 3.1 |
| api-alpha | 2 Feb 2026 | version: 1.0.0 → 1.1.0 |
| billing-beta | 2 Feb 2026 | price_cents: 1200 → 1400; tier: flat → metered |
| queue-epsilon | 16 Feb 2026 | version: 3.1 → 3.2 |
| api-alpha | 2 Mar 2026 | version: 1.1.0 → 1.2.0 |
| queue-epsilon | 9 Mar 2026 | version: 3.2 → 3.3; ack: manual → auto |
| api-alpha | 30 Mar 2026 | version: 1.2.0 → 1.3.0; auth: api_key → oauth |
| queue-epsilon | 30 Mar 2026 | version: 3.3 → 3.4 |
How the accuracy number was counted3 of 4 settled, 2 still open
The engine also guesses when the next change will land, then scores itself after the window closes. A miss counts. An open guess does not. Folding the two open rows into the score would change the denominator, so leaving them out is a real exclusion, not a coincidence.
| Dependency | Guessed on | Window | Predicted a change | What happened | In the score? |
|---|---|---|---|---|---|
| billing-beta | 2 Feb 2026 | 12 Feb 2026 to 20 Feb 2026 | yes | No change in the window. Miss. Settled 23 Feb 2026. | yes, as a miss |
| queue-epsilon | 16 Feb 2026 | 4 Mar 2026 to 14 Mar 2026 | yes | A change landed. Hit. Settled 16 Mar 2026. | yes |
| api-alpha | 2 Mar 2026 | 23 Mar 2026 to 6 Apr 2026 | yes | A change landed. Hit. Settled 6 Apr 2026. | yes |
| queue-epsilon | 16 Mar 2026 | 25 Mar 2026 to 4 Apr 2026 | yes | A change landed. Hit. Settled 6 Apr 2026. | yes |
| api-alpha | 6 Apr 2026 | 20 Apr 2026 to 4 May 2026 | yes | Window has not closed. | no |
| queue-epsilon | 6 Apr 2026 | 15 Apr 2026 to 25 Apr 2026 | yes | Window has not closed. | no |
What you actually buy
The public board is free and stays free. The paid thing is a monthly watch of one named repo: up to 40 dependencies you choose, weekly readings, the full dated diffs, the verdicts, and any open forecast. A person emails you that file once a month for as long as the subscription runs.
A larger plan at $499 a month covers up to 200 dependencies with daily readings instead of weekly. That amount is not the rail price. Say so in the email if the repo is that size.
- Not Dependabot or Renovate.Those tell you a newer version exists in a package registry. We keep a dated record of version, price, and the behaviour flags we can see, whether or not anyone published a release note.
- Not a one-off audit.One reading cannot produce a verdict, so we will not sell a single-shot “audit” of your list.
- Not a live vendor read, yet.JSON sources only, and only where a person has recorded that the written terms allow automated reads. A robots.txt allow does not override a terms page that forbids it. We do not scrape HTML marketing pages: a layout change would look like a price change, which is the lie this product exists to refuse.
- Not unknown rounded up to unchanged.If we could not read it, or we only have one reading, the answer is unknown, with the reason.
Questions we are actually asked
What will I see in week one? Your first reading, and nothing else. From the second reading on, you get real comparisons.
Can you tell me what moved last quarter before I paid? No. We can only compare readings we took.
Why does it say unknown? We could not read the dependency, we only have one reading, or the evidence contradicted itself. All three are genuinely “we don’t know”.
How accurate are the forecasts? Whatever the board currently counts. On this fixture run it is 75.0 percent, 3 of 4, with 2 still open and excluded. A live board with no settled forecasts will say unknown.
Do you store our code? No. We store the list of dependencies you name and the readings we take of public sources.
Ask before you pay
No pay button on this page yet. Email operations@ustechautomations.com. Name the repo. We reply with whether we already watch its dependencies, before you spend anything.
What arrives after you pay: $199 a month, billed monthly, for one named repo. Cancel any month by email, with no account to close and no notice period. After you pay, a person emails you the first dated watch file within one working day, then a new one every month for as long as the subscription runs. The public board on this page is the free sample.
Email us the repoSay the repo. We will tell you whether we already watch those dependencies before you pay. There is no pay button on this page yet.