Case Studies
Four searches that returned nothing,
and why that was wrong.
A search that finds nothing looks the same whether the art is absent or the tool failed. Telling those apart is most of the work. Below are four matters where the difference was measurable — with the counts, so you can check them.
Every patent, publication and database named here is public. No client work or confidential material appears on this page.
Case 01 — Reading the examiner's own search
The recorded prior-art search was empty. The best reference failed one filter, on a word choice.
The situation
A granted patent on a distributed inverter architecture. Nine rounds of searching had produced nothing on its central limitation. Rather than search harder, I pulled the Search Notes from the file wrapper and read what the examiner had actually run.
What the numbers showed
| EAST Search History (Prior Art) | “This search history is empty” | |
| Every recorded query | restricted to .clm. | claims only — no specification searched |
| Classification constraint | none | though he classified the case himself |
| Filter L1 | inver$5.clm. | |
| “main controller” in the reference's claims | 9× | passes his filter L2 |
| “local controller” in the reference's claims | 5× | passes his filter L3 |
| “inverter” in the reference's claims | 0× | fails L1 — it says “generator units” |
Counts taken from the reference's own claim text. The word “inverter” appears 21 times in its description and never once in its claims.
The finding
The reference satisfied two of the examiner's three claim-level filters and was removed by the third, because it calls the units “generator units” rather than “inverters.” A claims-only search could not reach it. Neither could a classification search: it sits in G05B5/01, anti-hunting arrangements — a control-theory class, not a power class.
Why it matters to a matter
An examiner's search history tells you where he could not look. That is a map of what is still findable, and it is sitting in the public file wrapper of every granted patent.
Case 02 — A retired field code
A field code was quietly retired. Searches using it return a clean zero.
The situation
US classification searches were returning nothing across several unrelated matters. Nothing errored. The results simply read as though the art did not exist.
What the numbers showed
| Legacy field code .uspc. | retired | no error, no warning |
| Current field code | .ccls. | |
| Symptom | 0 results | indistinguishable from “no art exists” |
The same classification, the same database, the same day — one field code returns nothing and the other returns the art.
The finding
Every search written against the old code had been silently returning nothing. A negative result produced this way looks exactly like a thorough search that found nothing, which is the most expensive kind of error in this work.
Why it matters to a matter
A clearance opinion or an invalidity conclusion is only as good as the assumption that the tool answered the question. That assumption has to be tested, not trusted.
Case 03 — Function words
A phrase containing the word “of” returns zero. Silently.
The situation
A search strategy built on precise phrases — the arithmetic a controller performs, quoted as it would be written — was returning nothing. The phrases were correct. The concept was common. The result was zero.
What the numbers showed
| “the” as a standalone term | 400,177 | |
| “by” as a standalone term | 6,365,888 | |
| divided ADJ by | 0 | adjacent operator, function word inside |
| divided ADJ by ADJ the ADJ number | 0 | |
| divided ADJ number | 75,376 | same phrase, function words deleted |
| proportional ADJ to ADJ the ADJ total | 0 | |
| proportional ADJ total | 8,609 |
The function words are indexed as terms — they return millions of hits alone — but they are stripped from the positional index, so they cannot occupy a position inside a proximity expression.
The finding
Sixteen queries in one matter had used such a phrase. Every one contributed nothing. Part of a recorded negative result was an artifact of the tool, not evidence about the art. The fix is not a different operator — it is deleting the function words and letting the content words sit adjacent.
Why it matters to a matter
The failure mode that costs the most is not a search that errors. It is a search that returns a plausible zero and gets written down as a conclusion.
Case 04 — The wrong vocabulary
The examiner searched the claim's words. The field writes different ones.
The situation
A network virtualization patent. The examiner's recorded search was careful — correctly date-capped, plurals and British spellings enabled, across eight databases — and it returned five consecutive zero-hit queries.
What the numbers showed
| His queries | virtual near8 flow near8 statistics | the claim's own coined words |
| “logical port” in the reference | 133× | |
| “logical forwarding element” | 62× | |
| “virtual port” | 2× | |
| “virtual flow” | 0× |
Replaying the examiner's own query today returns ten results, and the reference is not among them.
The finding
The art existed, in volume, six years before the priority date. It was invisible because the industry that built the technology writes “logical” where the claim drafter wrote “virtual.” Decoding the claim into the field's vocabulary was the whole search.
Why it matters to a matter
Claim language is drafted to be broad, not to be searchable. Finding art means searching what the engineers wrote, not what the attorney claimed.
The through-line
A negative result is a claim about the world. It has to be earned.
Probe every clause alone
A conjunction returning zero tells you nothing about which clause killed it. Each one gets a count of its own before the result is believed.
Prove the search can find
A query shape is validated against a document already known to carry the concept. If it cannot retrieve that, its zero is not evidence.
Verify the quote, not the summary
Every passage relied on is checked against the original document, with a column and line cite. A reference is never carried on a paraphrase.
Who did this work
A former USPTO patent examiner.
Michael W. Smith examined applications at the USPTO, then worked as a patent engineer and a patent analyst — patentability, validity, freedom-to-operate and landscape work across US, EP, JP, WO, DE and CN collections. B.S. Mechanical Engineering, University of Oklahoma.
Case 01 on this page is the reason the credential is worth something. Reading an examiner's recorded search and working out what it could not reach is not a database skill. It is knowing how the search was built in the first place.
More about the background →If a search on your matter came back empty, it is worth knowing which kind of empty it was.
Invalidity and prior-art search, scoped and priced flat. Send the patent and the deadline.
Get a free screen →