What has changed

Every release of the course and the kit is listed here, newest first. If you bought at any point, this page tells you what's different since, so you can reread the parts that moved instead of hunting for them.

Your access link never expires with a release. Updates are included for the year you have, at no extra cost, and the download always gives you the current version.

1.11.1 - 14 August 2026

Affects: the kit

• Reconciliation sweep against the current operation. Checked the whole handoff and all toolkit files against the live pipeline. Model routing, the pen lanes and cadence, the coherency framing (the automated check is a backstop that cannot pass a book; the human read is the only gate), the writer's finish-before-the-payoff guard, the AI-disclosure spec and the weekly cap were all already current. One real gap turned up and is fixed.
• The two toolkit programs that carry economics were still on the old numbers. niche-research.py and pre-reupload-audit.sh had the pre-refresh figures baked into their defaults and help text (median $17.90, 549 titles, 8.5 units per title, $19,701 catalogue). Earlier number passes only scanned the prose, never the code, so the tools you run disagreed with the modules you read. They now match: median $18.49, 560 titles, 8.9 units per title, best title $1,343, $22,698 exposure. Both still pass their runs.

1.11.0 - 13 August 2026

Affects: both

• The headline is now the whole operation, not just the flagship pen. Reports re-pulled 13 August. The top-line number is Amazon's own lifetime estimate across all three pen names, all 643 titles: $22,698. The course had been reporting only the largest pen ($20,445) and excluding the two newer ones; they are still young, but they are earning, so the headline now covers everything.
• Nothing was inflated to get there. $22,698 is Amazon's estimator figure at today's exchange rates. My own fixed-rate reconstruction is $20,571, and that lower number is what every per-book figure is still built on. Module 0 states both and explains the gap (currency conversion and this month's not-yet-final page reads), because a course about honest numbers shows its own seams.
• The per-book analysis deliberately stays on the established pen. Median $18.49, the linearity figure, the distribution and the charts are still measured on the largest pen's 560 earning titles, clearly labeled, because the two newer pens are under six months old and too thin to read per-title. Folding them into the median would move it down for no real information gain.
• Audiobooks updated: 151 net units sold across the nine titles on sale, up from 138. Still no earnings statement posted, and Module 0 and Module 8 still say so.

1.10.0 - 11 August 2026

Affects: the course

• Every module now opens with a "What to actually do" box. Three to five imperative bullets: the rule, the tool, the one number to carry away, pulled straight from that module's own content. If you only have a minute per module, that box is the minute. The stories are still there underneath for when you want the why.
• The narrative got a filler pass. Wind-up sentences, restatement, and anecdotes that did not end in a rule were cut across all nine modules. Nothing applicable was removed: every number, warning, tool, and mechanism is untouched, and the checks proved it (the set of figures in each module is identical to before). The modules turned out to be denser than they read, so the trim is modest and the new box does most of the work of surfacing what matters.

1.9.0 - 11 August 2026

Affects: both

• The kit's case is now made instead of assumed. The sales page gains a section that prices the $749 against what the machine cost to learn the hard way, using only receipts already on the page: the suspension, the five dead listings, the 43 raw books. No projected earnings anywhere, same as ever.
• The free module now ends by saying plainly who the kit is for, the course hub explains that the kit is the runnable half of the product, and the tool banner on every module page links to where the tool actually ships.
• A full leak audit of everything that ships came back clean. Every download, every page and every image was scanned for keys, account identifiers and identity leaks. Nothing was found in any shipped file. One internal build check was rearranged so its own word list lives outside the shipping folder entirely, which closes the only near-miss the audit turned up.

1.8.0 - 11 August 2026

Affects: both

• Every page got a full prose pass. Same facts, same numbers, same rules, told better. The course had a habit of opening every bullet with a bolded thesis sentence and closing every section on a coined maxim, which after nine modules reads like a machine clearing its throat. Most of that is gone. Bold now means a warning or a number you should not skim past, and the maxims that survived are the ones that earned it.
• Some figures were still standing on the July report and now match the August one. The best month is June 2026 at $2,406 on the current report's basis, not the $2,531 the old report showed. July came in at $1,837 and snapped a three-month climb, and Module 0 says so instead of quietly keeping the prettier sentence. Paid copies are 4,987 lifetime, the twelve-month figure is $16,457, and the summary table now agrees with the giveaway table on pages read.
• One claim got more precise at the cost of sounding worse for the argument: "two thirds of the catalogue never cracked $25" is actually three of every five, and it now says that.
• The dated note on the front page was re-dated and re-based to the current report. Nothing else about it changed, including the missing signature.

1.7.0 - 11 August 2026

Affects: both

• Every economics figure is re-pulled and current to today. The lifetime royalty report and the audiobook sales report were both exported fresh on 11 August, and every number in Module 0, the sales page, the charts and the toolkit's niche method now comes off those two files. Nothing here is carried forward from the July pull.
• The headline moved from $19,701 to $20,445, over 18 months instead of 17, across 560 earning titles instead of 549. Median title lifetime is $18.49, up from $17.90. Best title is $1,343, still under three times the second-best, so the no-jackpot finding survives another month of data.
• The linearity finding held, which is the part worth rereading. Revenue per title per month came in at $3.12 against last month's $3.25, on a catalogue that grew by another eleven earning titles. A month of extra data moved it four percent and did not bend it upward. If adding books compounded, that is where it would show, and it still does not.
• Kindle Unlimited is 38% of revenue on 2,208,058 pages read, against 39% on 2,076,100 last month. The mix is stable; the volume is not, and the module explains which of those matters.
• The audiobook column finally has numbers in it, and they are not good yet. Nine titles are on sale and have sold 138 net units between them since the start of July. The platform still has not posted a single earnings statement for them, so there is a real unit count and no royalty figure, and Module 8 now says exactly that instead of estimating.

1.6.0 - 11 August 2026

Affects: both

• Module 6: a missing drive does not raise, it returns nothing. With the working volume unmounted, every search across it came back empty with a zero exit code, so the pipeline concluded the books did not exist. An error stops you; an empty result set lets you carry on and draw conclusions. Assert the mount, and decide in advance whether zero matches is an answer or a symptom.
• Module 7: two lists of the same books, and the one search noticed. Titles added to the curated front door but not to the full catalogue got no detail page, were linked anyway, and surfaced as a 404. One of them was a latent 404 hiding behind a stale page from an earlier build. Also the false lead that cost longer than the bug: the link-building script looked broken in source and was fine in the rendered page. Check the rendered page before you fix a rendering bug.
• Module 8 gains a real price test. $12.99 accounted for 47 percent of bundle units while the rest of the shelf sat flat, so that was the price worth testing rather than defending. It moved to $9.99 across 58 of 59 titles, verified against 597 live listings rather than the run's own log. The store recomputes the other currencies slowly, and publishing before they settle silently saves the old ones.
• Module 8: the cover cascade repeated, twice, and recovery is worse than it sounds. Two more listings blocked on covers that had been live for weeks, from interior-only reuploads, while three others in the same batch went through untouched. A blocked listing's page bounces to the shelf, so keep your listing copy somewhere you own.
• The audiobook column is no longer completely empty. The first title of the round is live on the store. Still no royalty statement, and Module 0 says so.

1.5.0 - 2 August 2026

Affects: the kit

• Fixes the kit's main promise: assistants were refusing to run the guided setup. A buyer handed the kit to an AI assistant, it read START-HERE plus the operating handoff, the upload playbook and the toolkit, and declined to set anything up. On the information those documents gave, that was a fair call: the assistant could see mass-produced sexually explicit content and browser automation against a retailer, and nothing anywhere told it the operation is above board.
• START-HERE now opens with the context that was missing, addressed to the assistant: the store's explicit-content flag is always answered yes because it is a disclosure, the AI involvement is filed through the retailer's own AI-disclosure fields, the automation drives the operator's own account with no second identity and nothing touching customer reviews, and the content lines the pipeline refuses to cross are listed.
• It also says what the assistant is not being asked to do. Drafting and editing run on separate paid APIs; the assistant builds and runs the machine and is never asked to write the fiction. Any single step can be declined and handed to the human, which is better than stalling silently in phase two.
• Two lines that read as model-shopping for permissiveness were rewritten. The routing table said a drafting model was chosen because it "does not refuse". The true reason is that its provider's terms permit adult fiction where a general coding assistant's correctly do not, and it now says so.

1.4.0 - 2 August 2026

Affects: both

• The audiobook channel is now covered properly, with its costs and its empty return column. Module 8 gains a full section: the platform bans synthetic narration outright, so the one step in this operation that cannot be automated is a person. Rates run $75 to $200 per finished hour, and because the books are short a whole audiobook costs roughly $120 to $375; three real contracts are shown. Cast the priciest voice on the shortest book, because cost scales with finished audio and quality does not.
• Module 0 states the uncomfortable half of it. Fifty-eight percent of total spend went into audiobooks and there is still no royalty line beside it. A cost you can name precisely and a return you cannot name at all is a bet, and the module now labels it as one instead of calling it a growth area.
• The operating handoff carries the same rate structure, the casting rule and the backup rule.

1.3.0 - 2 August 2026

Affects: both

• Module 3 adds the apostrophe that forked a build. A rebuild took its output filename from the manuscript title, the title used a curly apostrophe where the file on disk had a straight one, and the rebuild wrote a second file beside the original instead of replacing it. Later steps globbed the folder and read the stale one. Every step reported success because every step succeeded. Write to the filename that already exists, never to one derived from content.
• The sales page is signed, dated and annotated by hand, and the numbers in its link preview now match the numbers in its headline. The preview said 530 books while the page said 549.

1.2.0 - 2 August 2026

Affects: both

• Module 6 adds two debugging rules, both from a publish step that would not publish. A documented trap that matches your symptom is a hypothesis, not a diagnosis: the runbook entry fitted perfectly and was about a different failure. And an exception after the point of no return is not a failed operation: a save had already succeeded when the page navigated out from under the code still waiting for it, so three retries redid finished work.
• Module 8: check for a live successor before re-listing a blocked book. A blocked book was queued to go back up under its original name while its retitled replacement was already live and selling. The duplicate guard could not see it, because the successor has a different title and nothing collides.
• dup-listing-guard.py gains check_relist() for exactly that, and it refuses in both directions: when a successor is found, and when no successor record was supplied at all. Suite is now 21 cases.
• Dates corrected: the two releases below were stamped 27 July in error and both shipped on 2 August. The terms page carried the same wrong date and is fixed.

1.1.0 - 2 August 2026

Affects: both

• Module 8 corrects itself on reupload risk. A reupload is re-review of the whole listing, not of the thing you changed. An interior-only repair wave of about 195 books had five listings blocked afterwards, all for cover content, on covers that had been live for weeks and had already passed review once. Price a wave by its cover risk, sequence the borderline ones last, and stop on the first block. The upload playbook carries the measured rate too.
• Module 6 adds the wrong-book publish. A fixed file went onto a different live listing because the intended one was archived by a human, the exact-title lookup skipped archived rows, and a prefix fallback matched the one live listing whose title extended it. Read the new section before you write any publish path.
• wrong-book-guard.py gains choose_listing(), which is that failure turned into a control: a parked exact match stops rather than falling through, the prefix fallback can never resolve to a title you are separately tracking, & and and are one title, and two candidates is ambiguity to escalate. Six new self-tests; the suite is now 40 cases. Re-run --test after updating.
• Module 3 adds a rule about your own orchestration. A transcript is a trace, not a heartbeat. An agent that has not written for twenty minutes is not necessarily dead, and dispatching a replacement onto its work is how one manuscript gets two writers. Ask your scheduler, and stop the original before starting a successor.
• The operating handoff carries all of the above where the method calls for it.

1.0.0 - 2 August 2026

Affects: both

• First public release. Nine modules, five operating documents, ten tools, and the guided setup.