What changed — without the tech fog
The latest MACRODOSE work adds a dedicated Video Intelligence workspace while continuing to improve private transcription, transcript review, and Book Flop’s large photo batches. This page separates what is implemented from what still needs live setup and device acceptance.
Explore a recording one moment at a time
The new workspace accepts MP4 and MOV files, creates a timestamped transcript, adds up to six sampled-frame notes, and builds a concise summary tied to the available evidence. Transcript and observation timestamps can jump to that point in the player. An interrupted upload now has an in-page retry action.
New in R26 — Transcript Intelligence II
A finished transcript is no longer just a wall of text. In the guided Review step, Daniel can search for a remembered phrase, jump to the matching words, Play Here from real transcription timing, follow timed words while audio plays, mark useful passages and send a trustworthy timed selection into Studio.
MACRODOSE does not invent clip boundaries. Studio handoff is enabled only when Vault has a real start and end from provider timing. Untimed text can still be searched and marked normally.
Open VaultBook Flop II — how to find it and use it
The easiest way to think about Book Flop is: choose pages → let MACRODOSE read them → start learning. Photos can come straight from the iPhone gallery in a multi-select batch, while PDF, EPUB and text use the document picker.
Tap Book Flop
Open Book Flop from the bottom navigation. You do not need to start in Knowledge.
Pick photos or a document
For photos, tap Open Photo Gallery and mass-select screenshots/pages. One Book Flop batch can hold up to 200 images, and you can add more gallery groups before reading. For a PDF, EPUB or text file, use the document button.
Check the photo order
If you chose photos, make sure Page 1 really comes first. Move pages with the arrows or remove anything you do not want.
Tap Read
MACRODOSE saves the source privately, reads it and builds one summary. Wait until the source says Ready.
Start Book Flop
Tap Start / resume Book Flop. Use Got it when an idea makes sense or Not yet for another explanation. Your place is saved automatically.
Photos: Book Flop → Open Photo Gallery → select several → Done → check order → Read → Start / resume Book Flop. Documents: Book Flop → PDF / EPUB / text → Read → Start / resume Book Flop.
Book Flop — visual batch intake
Daniel's requested multimodal intake is now implemented in the current build: Book Flop can accept PDF, EPUB/e-book and text sources, plus mass batches of up to 200 ordered pictures selected from the phone gallery. Large batches are processed in resumable passes, successful pages are cached, the full ordered batch contributes to the source summary, and the result feeds the existing adaptive learner.
Built ahead of paid acceptance. The additional $450 remains the quoted scope increase because gallery/EPUB ingestion required a new multimodal intake pipeline, private image handling, visual analysis, retry/recovery logic and iPhone-specific QA rather than a simple UI change.
Prior real-device QA milestone
During the recent client QA cycle, the full long-recording production chain completed on the real iPhone flow: private storage → the cloud transcription service processing → secure result return → finished transcript in Vault. This guide records that QA milestone; viewing it does not re-run the phone test.
Vault stays simple
The main path remains Record/Import → Review Transcript → Listen → Export. Transcript Intelligence II now lives inside Review instead of forcing the client into the dense advanced workspace.
Timing stays truthful
Karaoke highlighting and clip handoff use real provider word/segment timing. If timing is incomplete, MACRODOSE says so instead of guessing an audio range.
Book Flop II is source-first
It begins from an imported book rather than from the generic Learning Engine, so concept selection, re-teaching and recall stay connected to the actual document.
Read the Book Flop tutorialPremium read-back stays protected
Premium voice credentials remain server-side. Device voice stays available, and cached premium replays avoid unnecessary new generation.
Community work remains intact
Community posting still respects approved membership and visibility rules, moderators can post announcements and pin threads, and Home can surface private Learning Pulse information from explicit learning interactions.
What the QA push hardened
Earlier estimated effort breakdown
A simple view of where the prior roughly 80 hours of engineering and QA effort went. R26 Transcript Intelligence II is the next paid roadmap chunk and is separate from this earlier estimate.