So now what can you do with the USPTO bundle?

Why this patent

This walkthrough runs on my own patent. US 7,971,071 issued in 2011 to my late father and me as co-inventors; the patent is now mine alone, I never licensed it, and its term runs out in 2027. That makes it the rare demonstration subject nobody can object to: every document in its file wrapper is public, every wart in its prosecution is mine to show, and when the final steps draft a mock attack on its own claims, the only patent owner taking the hit is the one writing this page.

It also occupies a genuinely unusual position in patent history, and every section of this walkthrough leans on some part of that position.

First, it is still in force. A patent examined in 2009 and expiring in 2027 sits at the tail end of its generation. The population of live pre-AIA patents shrinks every week, and the tools you are about to use handle both generations, so this file is a chance to watch them work the older one while it still matters.

Second, it is pre-AIA. Congress gated the America Invents Act by effective filing date, so this patent lives its whole life under first-to-invent law. You will watch that world in action: an attempt to swear behind a reference by proving an earlier invention date, a tactic that no longer exists for anything filed today.

Third, it straddles KSR. The provisional went in under the old teaching-suggestion-motivation test. The Supreme Court decided KSR eleven months later, and the application followed three weeks after that. Unlike the AIA, a Supreme Court decision reaches backward, so the earlier provisional bought no shelter at all. One family, two kinds of legal change, and only one of them respected the filing date.

One prosecution, three clocks running through it. That is why the tools' insistence on dates, on which rule governed on which day, is not pedantry. It is the whole game.

The walkthrough

The patent itself covers a secure hardware device for delivering and protecting digital objects: think of a locked USB device that carries software or media, enforces the seller's rules on it, and can be disabled remotely. It granted after four years of argument with the examiner, two requests for continued examination, and that failed swear-behind attempt, and I already know where the bodies are buried because I buried them.

Below is a real prosecution question, broken into twelve prompts you can paste straight into Claude once the connectors are set up. Read the worked result.

The question: what did the examiner actually make us give up to get this patent, and if someone wanted to knock it out today, what would they file, and where?

The closing steps answer that literally: you will draft a mock anonymous reexamination request against my patent, and then discover why the weakest part of the patent cannot go in it.

One thing that closing exercise turns on is worth knowing before you start, because it is the kind of document ordinary retrieval walks straight past. On 6 February 2026 the Office published an update regarding the filing of anonymous requests for ex parte reexamination, carried in the Official Gazette as 1544 OG 128. Where at least one claim of the target patent was previously challenged in an IPR or PGR that reached a final written decision, an anonymous requester must include an affirmative statement that the real party in interest is not the prior petitioner or its privy. Step 10 surfaces it, and how it does is the lesson: the strategic search, asked for the reexamination domain, does not return it, because it is a Director's notice rather than a piece of the reexamination framework; asked for by name, the same server returns it first. The worked result shows both calls, the miss and the hit. Then the honest ending: the requirement does not bite on this patent, because Step 4 will already have established that its Board record is empty, so there is no final written decision to be estopped from. A good answer connects those two facts itself.

How to read this page

Each step says what you should get back, described as a shape rather than a set of values. On the EDGAR walkthrough that convention exists because live filings move. Here the reason is different and worth stating: the prosecution record of a patent granted in 2011 is closed. Mathers, Qawami, and the withdrawn rejections are permanent facts of the file, and this page names them. What still moves is the tooling around the record: the USPTO republishes datasets, coverage floors shift, and the citation indexes gain rows years after the fact (one lane's records for this application carry a 2025 load stamp). So where a step's result depends on a dataset rather than the file itself, the page promises the shape, and anything shown as a specific count from an evolving dataset is labeled as a capture of a day, not an expectation.

Under each prompt is a fenced block showing what the two recorded runs actually called and the MCP App panel those calls drew, rendered from the same captured payloads the worked result replays. The panels work: filter them, switch their tabs, click a node in the graph. They are a dated record of two runs on 5 September 2026 rather than a promise about your run, which is why the figures in them are not repeated as expectations in the prose around them.

One caveat about the panels' buttons, which are the real ones. Open in Patent Center opens an application only for a signed-in reader: without a free USPTO.gov account the link lands on Patent Center's own front page rather than on 11/752,072. Google Patents needs no account. Every external link on this page opens in a new tab.

Three clocks, and the tools that respect them

The three clocks in "Why this patent" are not background color; they are operating instructions. Half of patent law's greatest hits (Alice, Nautilus, the modern eligibility guidance) postdate my entire prosecution, and reading a 2009 rejection through 2026 law is the quietest way to get every conclusion on this page wrong. The law servers used here are built to prevent exactly that: the hypergraph refuses to answer without an as-of date, and its answers come back sorted into what was in force on that date, what had not yet come into effect, and what has been displaced since. Steps 3, 6, 8, and 11 each cash a clock: Step 3 ends on a document code with the AIA boundary printed in its own description, Step 6 demonstrates the KSR straddle on the examiner's own sentences, Step 8 reads 112 under its pre-AIA paragraphs, and the finale turns the AIA line into the reason one entire forum is closed to an attack on this patent.

One other thing. The assistant chooses which tools to call, and these six servers deliberately overlap: an office action can be reached through the text lane, the document bag, or the citation indexes. Part of what this page teaches is which lane answers which question, because the fast lane and the expensive lane return the same document.

Step 1: pin the application, and tell father from son

Using the Patent File Wrapper tools, find the granted patent invented by John Walkoe, a device for delivering and protecting digital objects. Give me the application number, patent number, filing and grant dates, examiner, and art unit. Then check its family: is there a provisional behind it? And pull the patent term adjustment so we know where it stands in its term. Tell me about anything in the inventor search results that is NOT this patent. Give me the application number in slash form (series/serial, the way the USPTO prints it) and use that form in every later file-wrapper call.

What you should get: one granted patent, with an application number in the 11-series, a 2007 filing date, a 2011 grant date, and a provisional application one year earlier in the family. The term-adjustment tool should report the USPTO's own PTA accounting in days, broken into delay components; note that it computes no expiration date, and a good answer will say so rather than invent one.

The slash-form instruction is not fussiness. The file-wrapper tools read a bare eight-digit number as a patent number first, and this application's serial is also the number of a patent granted in 2020 to someone else, so the bare form would silently answer about the wrong file. The responses echo the application number as bare digits, which is exactly how the mistake creeps in on a later step. Every response says which way it read the identifier; a good answer keeps the slash.

The last instruction is the point of the step. The inventor search will also return my father's telecom family: a 2006 continuation on which Wilbur John Walkoe, Jr. is the first inventor, with ten co-inventors. Both of us are inventors and we share this patent, so the risk is not a stranger with a similar name. It is merging a father's and a son's portfolios, or deciding that a real relative is unrelated. A good answer separates them and says how it knows. On EDGAR the equivalent mistake is picking the wrong CIK; here it is getting the family wrong, and it poisons everything downstream just as quietly.

One shape the panel above makes visible and the prose could not: the inventor search reports how many applications matched and how many it actually returned, and those two numbers differ. The response carries a bound note saying it hit the client response-size limit and telling you how to narrow the call. Expect that note rather than a silent truncation, and expect a good answer to say it is working from a partial list before it starts sorting out whose filings are whose.

Step 2: map the record before reading any of it

Get the document list for that application, but do not dump it: first tell me how many documents the file wrapper holds in total, then filter to just the claims documents (document code CLM) and list those with dates. What does the number of claims documents already tell us about how this prosecution went?

What you should get: a total document count well past a hundred, and a much shorter dated list of claims documents: one at filing and several more spread across the following four years, growing in page count. Eight claim sets before grant is not a patent that sailed through. The filter matters as much as the answer: an unfiltered document bag on a heavily prosecuted application is the token-explosion failure mode, and the tool's own guidance says to filter. Watch whether the assistant does.

Step 3: trace claim 1 from filing to grant

Now show me how independent claim 1 changed between filing and grant, without downloading any PDFs.

1. Pull the granted claims from the patent XML. How many claims issued, and how many are independent? Quote granted claim 1.
2. Pull the application's published (pre-grant) XML and quote claim 1 as published.
3. Pull the office action text for the whole prosecution (all office actions, not just the latest) and reconstruct the story: which rejections drove which amendments? The examiner's own "Response to Amendment" sections narrate it. I want the arc: which prior-art reference did the examiner start with, which one did he switch to, and what did we add to claim 1 to get past them?
4. Finally, find the swear-behind affidavits themselves in the file wrapper. Fair warning: guessing the document code from the rule number will fail; filter to my incoming documents and scan for affidavit-type entries. How many affidavit documents are there, filed on which dates, with which page counts? And read the document code's own description carefully: what does it tell you about the era this practice belongs to?

What you should get: a granted claim set with exactly one independent claim, and a granted claim 1 several times the length of the published one: the added length is the story. The office-action trail should show a handful of actions alternating non-final and final, two continued examinations, an early anticipation rejection on one reference that we argued past, a new rejection on a second reference the examiner found, and (this is the detail worth pausing on) an affidavit attempting to prove we invented before that second reference's date, which the examiner found insufficient. We could not swear behind it, so we amended around it instead. The hardware-segmentation and remote-disable language in granted claim 1 is what "around it" looked like. Granted claim 1 is long enough that this page excerpts it rather than quoting it whole: its two load-bearing limitations are the four non-volatile memory segments with differentiated access levels and the distributor-controlled remote disable, and the full claim text is one XML tool call away, the same call the prompt already makes.

Two tool-discipline notes this step teaches. First, no PDFs and no OCR were needed: granted claims come from the grant XML, published claims from the application XML, and examiner reasoning from the office-action text lane. The as-filed claim set exists only as a scanned PDF, and this step deliberately does not open it; the published XML is the free, faithful stand-in. Second, office actions come back newest first, and the response says so; the arc reads back to front unless the answer reorders by date. Third, long office actions come back windowed with a continuation offset, and when several are returned at once they share one budget; a good answer pages through rather than quoting the first window as if it were the whole document.

Part 4 turns the swear-behind story physical, and it checks a family memory against the record. Expect the obvious code guess to come back empty (the affidavits are not filed under a code named for the rule) and the incoming sweep to find them under AF/D, a code whose description is a small monument: "Affidavit-submitted prior to Mar 15, 2013." The USPTO's own document taxonomy carries the regime line inside it, because swearing behind a reference is a first-to-invent tactic. Be precise about what that means, because the obvious reading is wrong: the mechanism is not gone. A 37 CFR 1.131 declaration is still available to pre-AIA applications and patents, including in a reexamination and in defending a pre-AIA patent at the Board. What is true is that it is not available to AIA first-inventor-to-file applications at all, which is why the document code stamps a date on its own description. My patent sits on the side of that line where the tactic still exists. Expect five affidavit documents in three batches: two with the after-final response, the same two refiled a month later with the continued examination, and one more with the last amendment before allowance. Now the part worth being careful about. My father and I signed the inventor affidavit and a friend signed a separate corroborating declaration. The record is consistent with that (each swear-behind round contains two distinct affidavit documents of different lengths), but the record does not by itself prove it: the metadata names no declarants, the documents themselves are scanned images this walkthrough deliberately does not extract text from, and the examiner's written analysis engages only the inventors' affidavit and its Exhibit A. Whatever the second document said, nothing in the examiner's text relies on it. That is the same discipline as the KSR-catchphrase hunt in Step 6: the page reports what the record shows, even about its own author.

Optional, if you want to test the path this page does not take. The five AF/D documents are scanned images, and nothing above needed them: the dates, the page counts and the document code all came from metadata, and the examiner's own analysis of the affidavits came from the office-action text. If you want to see what is inside them anyway, add a thirteenth prompt of your own and ask for the extracted text of the five AF/D documents, or for download links to them so you can read the originals yourself. Both are one call. The second is the more conservative habit and the one worth building: ask for the download link and read the document, rather than reasoning on a transcription of it.

Step 4: the PTAB check, where empty is the answer

Has this patent ever been challenged at the Patent Trial and Appeal Board? Check trials (IPR, PGR, CBM) by patent number AND ex parte appeals by application number. Then tell me precisely what a zero result does and does not establish.

What you should get: zero, twice. No trials against the patent, no appeal during prosecution. This is a step about resisting the urge to pad an empty result: a clean zero from the trials index is an answer, not a failure, and the server even tells you that the underlying USPTO API reports "no results" as an HTTP error it has normalized away.

What the zero establishes: nobody has used the Board against this patent, and we never appealed a rejection (we filed continued examinations instead; Step 3 already showed you that route). What it does not establish: that the patent is strong. Nobody challenges a patent nobody is asserting. An unlicensed, unlitigated patent near end of term has a spotless Board record for the least flattering reason available, and a good answer will say so unprompted. Hold on to this zero: it comes back in Step 10 with a procedural payoff I did not expect when I designed this page.

Step 5: what the examiner actually held against it under 103

Now the obviousness record, from two independent sources.

1. Try the structured office-action rejections tool first. If it comes back empty, tell me why (read the coverage note) and fall back to the office-action text lane with the section filter set to 103. Which claims were rejected as obvious, over which reference combinations, in which office actions? Pay attention to whether each result is a real 103 section or a fallback to the full document; the response labels it.
2. Then pull this application's records from BOTH citation lanes (enriched and office-action citations). The two lanes spell the same reference differently, so join them on the reference key the server stamps on every record rather than on the raw identifier, and use the examiner-cited flag to separate the examiner's citations from my own disclosures. Compare: which references do the citation indexes say were cited, and how does that list differ from the references the rejections were actually BUILT on?

What you should get from part 1: an empty structured result with a dated coverage floor (that dataset starts in late 2017 and this prosecution ended in 2011), followed by a successful fallback. That failure is deliberate and it is the most transferable lesson on this page: every USPTO dataset has a floor, the floors differ per dataset, and the discipline is to read the floor note and reroute rather than conclude the data does not exist. The text lane reaches back years further than the structured lane. Expect the section filter to behave two ways: some office actions have a separately indexed 103 section and return just that; others return the full body with a note saying no 103 sub-document exists. The note is the tool being honest, and the honest tool is the one you can build on.

What you should get from part 2: a handful of records from each citation lane, overlapping but not identical, one lane carrying the statutory basis code per reference. Every record on both lanes carries the same normalized reference key, so the overlap is a join, not a guess, and the examiner-cited flag is a true or false value, not a word. And a discrepancy worth thinking about: the citation indexes list the references the examiner put on the citation form, which is not the same list as the references the rejections were argued from. Cited art and applied art are different questions. The citation lanes answer "what was on the examiner's radar"; the office-action text answers "what was used against the claims." An analysis that treats the 892 form as the rejection record has quietly answered the wrong question.

Dated capture: on 5 September 2026 both citation lanes returned five records for this application, with office-action dates from 2008-2009, years before the lanes' documented 2017 coverage start. That is a record of a day, not a promise: these indexes are documented as starting in 2017 and have been observed serving older prosecution anyway, and their load stamps show rows being added as recently as 2025. Query first, assume never.

Step 6: the KSR straddle, the law in force on the day, in the examiner's own words

Here is a date coincidence I did not engineer. Our provisional was filed in May 2006, eleven months before the Supreme Court decided KSR v. Teleflex. The non-provisional was filed three weeks after KSR came down. The entire examination ran under post-KSR obviousness law, applied to an invention with a pre-KSR priority date. This step uses that straddle to teach the sharpest date distinction in patent work, and then checks the theory against the examiner's actual sentences.

A date puzzle. My provisional was filed 24 May 2006. KSR v. Teleflex was decided 30 April 2007. My non-provisional was filed 22 May 2007, and every office action came in 2008-2010.

1. Confirm those family dates from the file wrapper's continuity data. Then answer precisely: which of my dates governs WHAT COUNTS as prior art against my claims, and which dates govern WHICH OBVIOUSNESS FRAMEWORK the examiner applies? Those are different questions with different answers.

2. Run the hypergraph's section_103_obviousness pattern with the office-action date (13 October 2010, the final rejection of claim 1's lineage) as the as-of date, and read me its three regime buckets: in force, not yet in effect, already superseded. Was KSR in force? How fresh was the newest KSR guidance on the examiner's desk that day? And which parts of today's obviousness-and-eligibility framework should I NOT read these rejections through? Then run the MPEP RAG server's strategic search with that same key as the domain for the governing guidance documents themselves.

3. Now the examiner's own words as the evidence. Pull the complete 103 sections from the 2009 and 2010 office actions and hunt the rationale language: do you find the KSR catchphrases (combining familiar elements by known methods, predictable results, obvious to try), strict TSM phrasing, or something else? Quote every "it would have been obvious" sentence in full and classify each against the MPEP's post-KSR rationales. Quote, do not paraphrase.

What you should get: the framework-shift report, which is the payoff of the three-clocks framing and the reason the as-of parameter is mandatory. Expect three buckets. In force: Graham (1966), KSR (2007), and the chain of USPTO guidance implementing KSR, and look at the newest entry's effective date, because the Office's updated KSR guidelines took effect weeks before this office action was mailed. My claims were rejected under guidance fresh off the press. Not yet in effect: everything a modern reader reflexively reaches for (the 2011 112 guidelines, Alice, the eligibility guidance line, the 2024 obviousness updates), all explicitly flagged as postdating my prosecution. Already superseded: empty, because in 2010 nothing governing this examination had yet been displaced. Also expect the pre-AIA statutory framing to surface on its own: the rejections quote 103(a) with "at the time the invention was made," language the AIA later replaced with the effective-filing-date standard.

The hypergraph call is the one to watch: it refuses to run without an as-of date, on the principle that a rule applied to the wrong side of a regime change is wrong silently, and it echoes back which event its date should be (for this pattern, the examination date). That refusal is a feature. The RAG server should hand back the actual KSR guidelines document and the MPEP's prima-facie obviousness rationales, ranked by how many of the domain's sub-queries found each one.

One expectation this page had to correct after watching two runs. The retrieval server's domain search returns fused documents, not links: the panel above shows what it gives you, which is a ranked list with a fusion score and a corpus name per row. A link per record arrives only from its other lane, the plain semantic search, and Step 10 is where that difference stops being a detail and becomes the answer.

Part 1's answer should land the contrast cleanly: the priority and invention dates govern what counts as prior art (and this prosecution literally fought that fight, because the Step 3 swear-behind affidavit was an attempt to move the invention date to before a reference), while the examination dates govern the framework the rejections are built with. A pre-KSR invention examined in 2009 gets post-KSR law, and there is nothing improper about that: framework changes apply to the examination, not the filing.

And the provisional deserves its own sentence, because I learned this the hard way: the earlier provisional bought me nothing against KSR. A provisional fixes the priority date (what counts as prior art) but it does not freeze the law. KSR is a judicial interpretation of section 103, a statement of what the statute always meant, so it reached every application still pending and every patent still alive, mine included, regardless of what the test was the day I filed. Contrast the AIA: Congress gated that statutory regime change prospectively by effective filing date, so a pre-AIA priority date DOES keep you on the old statute. Two kinds of legal change, and only one of them respects your priority date. My patent sits on the protected side of the AIA line and the exposed side of the KSR line.

Part 3 is designed around what the record actually says, and it is more instructive than the version I expected to write. Do not expect the famous KSR catchphrases; they are not in these office actions. Expect every combination to take the older, articulated form (a named modification, a stated benefit, "as taught by" the secondary reference), and then let the law servers resolve the apparent anachronism: KSR did not abolish that style of reasoning, it demoted it from a rigid requirement to one acceptable rationale among several, so an examiner articulating a motivation to combine in 2010 is applying KSR-era law in pre-KSR vocabulary. The evidence of the regime is structural (Graham quoted up front, an explicit reason attached to every combination), not the buzzwords. If your assistant claims to find "predictable results" in this record, it is paraphrasing, not quoting. Make it show the sentence.

Dated capture: one of the examiner's combination sentences, quoted whole as the office-action text lane returned it on 5 September 2026: "It would have been obvious to one of ordinary skill in the art at the time the invention was made for the smart cards of Qawami to be disabled remotely in order to discourage theft and counterfeiting as taught by Rohrbach (Col. 2, lines 8-11)." The underlying office action is a closed 2010 document, but the text lane serving it is a living dataset, which is why the capture carries a date. Note the pre-AIA tell inside the sentence itself: "at the time the invention was made."

Expect the two servers to answer differently under the same domain key: one retrieves the text of the law, the other dates which rules governed. They are designed as a pair, and this step is where the pairing becomes visible.

Step 7: the contrast beat, a famous neighbor

For contrast, look at a famous patent in the same technology neighborhood: IOENGINE's US 8,539,047 (a McNulty portable-device patent: a secure USB device that delivers its own code to the terminal it plugs into). Run the same two checks you ran on mine: what does the PTAB trials index show against it, and what does its record look like next to mine? This is an illustrative contrast, not a validity comparison; say so in your answer.

What you should get: the mirror image of Step 4. Where my patent's PTAB check returned zero, this one should return actual inter partes review proceedings with petitioners, dates, and statuses: a portable-device patent that was worth enforcing, and therefore worth attacking. The point of the contrast is not that one patent is better; it is that the PTAB record measures commercial friction, not technical merit. Two patents from the same technology moment, one licensed and litigated into a Board docket, one parked, and the tools show you the difference in one call each.

Keep the answer honest: nothing in this step compares claim scope or validity, and an answer that drifts into "whose patent would survive" has left the record and started speculating.

Expect two acts in the result set: a cluster of petitions from one petitioner, all filed the same day and all denied institution on the same day, and a separate petition from another petitioner that was instituted and ran all the way to a final written decision. Denied-versus-instituted on one screen is the whole PTAB lifecycle in miniature, on a patent one technology shelf over from mine.

Dated capture: on 5 September 2026 the trials index returned five IPRs against US 8,539,047: four PayPal petitions filed together and all denied institution (IPR2019-00884 through -00887), and Ingenico's IPR2019-00416, instituted in 2019 and terminated in 2020 with a final written decision. That patent's docket is unlikely to move again, but the capture is labeled a capture all the same.

Step 8: the 112 read, where I think my patent is weak

Now the part I have been putting off. I have always suspected this patent's claims are weak under section 112. Check the record:

1. Pull the office-action text with the section filter set to 112, across all office actions. What 112 rejections were made, on which claims, on what theory (indefiniteness vs written description), and what happened to each: argued away, amended around, or fatal to the claims involved? One caution: where the response says it fell back to the full document because no indexed 112 section exists, search that full text yourself before concluding the office action made no 112 rejection.
2. Route the law to the MPEP servers under the section_112_requirements key, with the same as-of discipline as Step 6: the definiteness and written description standards as they stood at my office-action dates. Pre-AIA 112 first and second paragraphs, not today's 112(a) and 112(b), and not Nautilus, which postdates my grant by three years. The MPEP quotes the pre-AIA and AIA versions side by side; use that.
3. Finish with one honest paragraph: what does this prosecution record actually signal about 112 exposure, and what would it take to turn a signal into a conclusion? Neither you nor this page is offering that conclusion.

What you should get from part 1: three distinct 112 events with three different endings, and only two of them are served as indexed 112 sections. First, an early written-description rejection against a handful of amended and new claims, sitting inside an office action whose 112 material is NOT separately indexed, so it surfaces only if you search the full-body fallback text, which is exactly why the prompt tells you to. Second, an indefiniteness rejection sweeping most of the claim set mid-prosecution, withdrawn an action later after our arguments were found persuasive. Third, a written-description rejection late in prosecution against a small group of newly added claims, where the examiner rejected our pointer into the specification as insufficient support, and those claim numbers are absent from the granted patent. One 112 problem we amended through, one we argued away, one we resolved by letting the claims die. The tooling lesson rides along with the legal one: a fallback response is not an empty response, and the count of indexed sections is not the count of rejections.

The strip above is the connector answering that caution itself, and it is worth knowing before you run the step. Ask for one section and the response now reports what it actually handed you: which office actions had an indexed 112 sub-document and which fell back to the full body, and, for a served section, whether that excerpt holds every 112 ground the full text holds. USPTO indexes a first-paragraph and a second-paragraph rejection separately, so an excerpt can look complete and be half the story. Watch for a partial-excerpt marker and re-read with the section filter off when you see one. An assistant that reports the number of 112 rejections without mentioning that marker has answered from the excerpt rather than from the office action.

What you should get from parts 2 and 3: the MPEP's two-requirements framing of pre-AIA 112 second paragraph and the possession standard for written description, with citations, and then the boundary line, stated plainly: this is an analysis of prosecution-record signals. A withdrawn rejection is not a clean bill of health, an abandoned claim family is not an admission, and nothing here is a legal opinion on the validity of the granted claims. The step ends on that line on purpose, because Step 11 is about to show you where a real 112 challenge would have to be brought.

Step 9: the petitions lane, an empty with an explanation, then the pivot

One connector in the bundle has not spoken yet: Final Petition Decisions, the published final decisions of the Office of Petitions. My file wrapper even contains a petition-shaped paper (the document sweep can show a "Petition for review by the Office of Petitions" filed in 2007), which makes what happens next worth watching.

One connector left: Final Petition Decisions. Check my application (11752072; that connector takes digits only, no slash or comma) for petition decisions. My file wrapper contains a 2007 document coded as a petition to the Office of Petitions, so be precise about what an empty FPD result does and does not mean here; state the FPD corpus's documented coverage as the tool itself reports it. Then pivot: search petitions by my art unit, 2432, and characterize petition practice in my technology space: petition types, rules invoked, deciding office, outcomes.

What you should get: an empty by-application result that explains itself. The response's own guidance states the coverage: applications filed in 2001 or later, loaded monthly from a set the USPTO describes as starting with 2022 decisions, so 2022 is a completeness floor and anything earlier is partial. That is the whole lesson of the first half: a petition filed in a wrapper and a published final petition decision are different objects, the corpus's own floor sits fifteen years after my petition, and absence from the second implies nothing improper about the first. My 2007 petition was the routine kind that accompanies a filing, resolved without a published final decision, and the tool says in terms that a zero on prosecution that ended before 2022 says little either way. This is the third honest empty on this page, after the PTAB pair, and by now the reflex should be automatic: read the note, name the scope, do not pad.

The pivot should return a small set of petitions for the art unit, each row carrying a petition type, the rule invoked, the deciding office, and an outcome. Read the shape rather than the rows: what kinds of petitions actually arise in this technology space, and how did they fare?

Dated capture: on 5 September 2026 the art-unit query returned five petitions for AU 2432, and every one was the same species: a petition to make special under the Patent Prosecution Highway, 37 CFR 1.102(a), all DENIED by the Office of Petitions, mail dates 2011 to 2024, applicants including Huawei and Honeywell. Five-for-five one petition type is a fact of a day and of a corpus whose current posture is all-denials, not a fact about the art unit, which is exactly why this paragraph is a labeled capture and not an expectation.

Step 10: the finale, part one: draft the attack (the part that CAN be filed anonymously)

This is where the walkthrough turns the tools around. Everything so far read the record; now the reader drafts against it. Against my patent, with my blessing: it is unlicensed, near expiry, and I would rather my own walkthrough stress-test it than leave the exercise theoretical.

Using everything on the record so far, draft a MOCK ex parte reexamination request skeleton against US 7,971,071, clearly labeled as an educational exercise, never to be filed.

First the law, from the MPEP servers under the ex_parte_reexam_snq key with today's date as the as-of date: what are the ground limits of ex parte reexamination under 35 U.S.C. 302-303 and 37 CFR 1.510? What is the substantial new question standard, and what does the Office's February 2026 Official Gazette notice on filing anonymous reexamination requests against previously challenged patents say? If the strategic search does not surface that notice, search the MPEP RAG server for it by that title; it is there.

Then the draft skeleton, from the record: the claims to be reexamined (start with claim 1), the proposed art, beginning with what the examiner himself applied (Qawami, Rohrbach, Mathers, Girard) and anything additional the two citation lanes surfaced, and for each proposed combination, the "new light" argument the SNQ standard requires, given that art already considered by the Office needs a materially new presentation to count. Be honest about which of these references would struggle to clear that bar, and why. Cite the office action, date, and claim numbers behind every assertion.

What you should get: first, the ground rules, retrieved rather than recalled: ex parte reexamination takes ONLY grounds based on prior patents and printed publications: 102 and 103 over documents. The MPEP says in terms that other questions "should not be included... and will not be considered." Anyone at any time during enforceability may file, and a requester can stay anonymous by filing through a registered practitioner. Anonymity here is ordinary, regulated practice: the Office's Director notice of February 2026 is the document that governs it, and it adds an extra affirmative statement expected when the patent was previously challenged at the Board. Here is the Step 4 payoff: my patent's Board record is empty, so that condition is not triggered, and the assistant should connect those two facts itself.

Watch the two retrieval calls in the panel above, because the prompt is written to force the second one. The domain search does not surface the February 2026 notice: it is fused against a pattern key, and the notice is a Director's Official Gazette item rather than a piece of the reexamination framework. Asked for by name, the same server returns it first, with a similarity score near the top of the scale and a URL field pointing at uspto.gov. That URL field is the only one anywhere in these twelve steps: every other link either page shows is built from an identifier. So the shape to expect is a miss, a reroute, and a hit that arrives with its own citation attached, and an answer that says which of its two calls found it.

Then the skeleton, and its hardest honest paragraph. The examiner-applied art is the natural starting material, but it is by definition art the Office already considered, so every combination built on it must clear the "new light" bar: a materially new argument or presentation, not a rerun. Expect a good draft to say that some of the record art probably cannot clear that bar alone, and that a real request would go hunting for art the examiner never saw. The draft should read like a skeleton with citations, carry its educational-mock label prominently, and claim no view on whether the request would actually be granted.

Step 11: the finale, part two: the attack that CANNOT go in the request

Now try to put the Step 8 material into the reexamination request, and explain why you can't. Where CAN a 112 invalidity theory against this patent be heard? Walk the forums: ex parte reexamination, post-grant review, inter partes review, district court. For THIS patent (granted June 2011, pre-AIA), which doors are open, which are closed, and why? Use the MPEP law servers for the routing, with as-of dates.

Then draft the 112 theory in the only form left standing: a short MOCK litigation-style invalidity contentions section (educational exercise, never to be filed, not a legal opinion) asserting indefiniteness and written-description theories against the granted claims, built strictly from the Step 8 prosecution record, with a citation for every assertion and a closing paragraph on what the contentions do NOT establish.

What you should get: the routing lesson, which is the real content of the step, and the place the three-clocks framing pays off hardest. The 112 material bounces off every Office door: ex parte reexamination cannot hear it (patents and printed publications only, the same ground limit Step 10 retrieved); and post-grant review, the one Office proceeding that does hear 112, is not merely time-barred for this patent but regime-barred: PGR exists only for patents with an effective filing date on or after 16 March 2013 (AIA first-inventor-to-file patents), and a 2007-filed patent is categorically outside the regime. The nine-months-from-grant window (which would have closed in early 2012 anyway) never even opened for it, because when this patent granted in June 2011 PGR did not yet exist. A good answer gets the order of those two reasons right: regime first, clock second. Inter partes review is open in principle but carries the same patents-and-printed-publications limit as reexam, and is never anonymous. What is left is litigation: 112 invalidity against this patent is a defense or counterclaim in district court (the one forum that hears all grounds), raised by someone who has been accused of infringing, which also means the anonymity that Step 10's request enjoyed is structurally unavailable to a 112 challenger. The routing table the answer should land on: ex parte reexam (prior-art grounds only, anonymous, available); PGR (112-capable but regime-unavailable for a pre-AIA patent); district court (all grounds, never anonymous). The forum choice and the ground choice are one decision.

Then the mock contentions: short, record-bound, and labeled. Expect the indefiniteness theory to lean on the claim language the examiner himself once called indefinite (and the awkward fact that it was withdrawn; a good draft addresses the withdrawal rather than hiding it), and the written-description theory to lean on the specification passages the examiner found insufficient for the claims that died. And expect the boundary section to be blunt: a mock contention built from a closed prosecution record, by a model, on the patent owner's own walkthrough page, establishes nothing about validity. It demonstrates what the record supports arguing, which is a different thing.

Step 12: write it up

Write a short memo titled "US 7,971,071: what the record shows," covering: the prosecution arc and what claim 1 traded for allowance; the clean PTAB record and what it does and does not mean; the obviousness record (applied art vs cited art, and the KSR-era framework it was examined under); the 112 signals; the petitions check; and the forum-routing conclusion from Steps 10 and 11. Re-pull anything you rely on; do not trust earlier answers from memory, and use the slash-form application number for every file-wrapper re-pull. Put a citation on every factual claim: office action date and claim numbers, or the MPEP/guidance source with its URL.

Finish with two sections: what this analysis does NOT establish, and anything you inferred rather than read directly in the record.

What you should get: a sourced memo whose limitations section does real work. This walkthrough produced two mock adversarial drafts against a real patent; the memo is where the frame is set permanently: record analysis, educational exercise, no legal opinion, no filing. The re-pull instruction matters here more than on the EDGAR page: twelve steps of context is enough for an assistant to start trusting its own summaries, and the citation requirement is how you catch it.

Optional: the threads this run left unopened

This one is not part of the twelve and is not needed to finish them. It exists because the session that produced the worked result was asked afterwards what it had noticed and left out, and its answer named three documents it had seen in an inventory it had already pulled and never opened. Every one of them was a single filtered call away. Run this if you want to see what a second pass buys.

Optional follow-up. Go back to the document inventory for 11/752,072 and open three things you have not read yet, by document code.

1. The applicant's own Remarks (document code REM). Everything you told me about what we argued came through the examiner's characterization of it inside his own office actions. Read our side in our own words, and tell me where the examiner's summary of our argument differs from the argument.

2. The advisory actions (document code CTAV, two of them). An advisory action is what the examiner issues after a response to a final rejection and before a continued examination, and it says whether that response put the case in condition for allowance. Read the one that sits between the 2009-09-21 final rejection and the 2010-03-22 continued examination.

3. The petition decision (document code PETDEC). Earlier you established that the petitions corpus holds no decision for this application and that a 2007 petition document is nevertheless in the file wrapper. There is also a PETDEC document in the same inventory. Find out whether it is the decision on that petition, and what it says.

For each, say what it changes about the story you told me, and what it does not.

What you should get: the applicant's side of the argument in the applicant's own words rather than in the examiner's summary of it, which is the asymmetry this walkthrough otherwise leaves in place; a procedural document that says what the examiner made of our response to the final rejection; and an answer to a question Step 9 raised and never closed. Some of these are scanned images, so expect the assistant to say when it is working from extracted text rather than from a text document, or to offer you the download links instead.

What you just used

That workflow crossed all six connectors without you naming most of them: the file wrapper's search, family, and term lanes; the document bag with its code filters; the XML claims for both the grant and the published application; the office-action text lane with section targeting; both citation indexes; the PTAB trials and appeals indexes; the petitions corpus; and the two law servers answering the same domain keys two different ways. Here is the map, step by step:

About the USPTO Bundle · Setup instructions · The worked result · The EDGAR walkthrough