Fieldnotes · Mohamed Amir Gomaa

How the environment fits together

Fri, 16 Oct 2026 00:00:00 +0000

Start with the source

NixOS configuration repository · Published Neovim configuration

The public NixOS repository is the published baseline. The local security maintenance verified on 16 October has not yet been committed to that repository. The interactive guide uses short, explanatory examples; it is not a copy of private machine state.

Two configuration layers

The flake pins Nixpkgs, Home Manager and Fenix. NixOS declares system packages, hardware integration and services. Home Manager declares user packages and configuration links. The system build integrates both layers, so package/configuration changes can be reviewed together.

Using Batchlark

Fri, 16 Oct 2026 00:00:00 +0000

What you are observing

Batchlark is the retained concurrency study formerly called Queue Lab/Hermes. It is not a differentiated service; a practical science-oriented replacement is being evaluated. Its dashboard represents simulated 50 ms work, not real email, webhook or language-model deliveries. It retains state in memory and resets on restart.

Run it locally

Follow the tested source setup in the implementation notes . The served application uses Go and opens a dashboard on port 8080 by default. No API key or database is needed.

Using Noteweld

Fri, 16 Oct 2026 00:00:00 +0000

Try it without an account

Open the public editor to try Markdown, preview and optional Vim. The scratchpad lives in the tab; this example does not save it to a private account. Download anything you want to keep before leaving.

Explore the public graph for an original six-note sample. It does not read private notes or start a processing job.

Create a private workspace

Open the hosted workspace , register using the displayed requirements and keep the private recovery key securely. Recovery is based on your passphrase/key; no email reset service is connected. Keep the recovery key out of messages and screenshots.

Using Pageweft

Fri, 16 Oct 2026 00:00:00 +0000

See a completed result

Open Pageweft or explore the original PDF/Markdown/ZIP example linked from its website and the portfolio. Viewing the completed example does not run a GPU conversion.

Inspect before processing

Upload a PDF you have permission to process. Each file can be at most 64 MiB and 1,000 PDF pages; up to five files can be inspected, with each conversion confirmed separately. Choose the original PDF page numbers or ranges and review the text-layer inspection and quote.

Using Sonaweft

Fri, 16 Oct 2026 00:00:00 +0000

Hear a completed example

Open the original example to hear source audio, inspect the actual recognition/review and download a hashed bundle. This precomputed example does not start another transcription job.

Upload and inspect

The hosted workflow accepts authorized English audio/video up to 64 MiB and 60 minutes. Upload a file and inspect its duration/quote before confirming work. The hosted application does not download arbitrary platform URLs or use your browser cookies.

Using the vanilla demo

Fri, 16 Oct 2026 00:00:00 +0000

Open the portfolio demo

Open the bundled demo or launch it from the portfolio’s vanilla JavaScript card. This is a separate implementation from Noteweld. It demonstrates native browser APIs rather than the professional Next.js application.

Work in one browser

Try notes and checklists, pin/color/label controls, archive and trash. Storage belongs to this browser and website origin. It does not create a cloud account, synchronize devices or inherit Noteweld’s editor/graph features.

Why these previews run on Modal

Fri, 16 Oct 2026 00:00:00 +0000

More than a GPU rental

Modal runs application containers on CPU as well as GPU resources. It supports HTTP applications, background functions, schedules, persistent volumes and secrets. An HTTP server can be Python/FastAPI or a Node process; the deployment wrapper does not require every application to be written in Python.

Official references: HTTP applications , autoscaling , persistent volumes and secrets . Checked on 16 October 2026; provider policies and prices can change.

The current deployment

Component	Hosting	Persistent state
Portfolio, vanilla demo and this Hugo library	GitHub Pages	Static versioned files; vanilla demo uses browser storage
Pageweft web/API and extraction jobs	Modal CPU containers and A10G workers	Separate document and serialized SQLite ledger volumes
Sonaweft web/API, speech and review coordination	Modal CPU containers and L4 workers	Separate media, model-cache and serialized SQLite ledger volumes
Noteweld web/API	Modal Node/Next.js CPU container	Isolated Neon PostgreSQL; encrypted backups in Modal
Queue Lab	Local Go demonstration	In-memory state; no separately verified public production backend

Paddle handles commercial payments rather than hosting these websites. DeepSeek is an optional text-inference provider for Sonaweft. No shared customer wallet or login is implemented. Linguavell is a separate startup project.

Batchlark: the retained study and replacement decision

Fri, 16 Oct 2026 00:00:00 +0000

The actual starting point

The old repository mixed early RAG scaffolding with a later webhook/concurrency demonstration. Surviving original history is:

Commit	Actual historical purpose
`431b89e`	Initial project
`01e0a92`	Render binding/missing-key repair
`acb9d80`	Rewrite toward a webhook dispatcher UI
`52b6258`	SSE and unit-test work

The later reviewed working tree had already repaired bounded admission, worker accounting, shutdown and slow-stream behavior, while making clear that task delivery was simulated. It was preserved as checkpoint ce80475 before this cleanup. That checkpoint records pre-existing work, not newly invented development.

Building Noteweld: accounts to connected notes

Fri, 16 Oct 2026 00:00:00 +0000

Start from the evidence

The current source is connected-notes handoff f6901ff . Application code for that collection is e1b8740; earlier foundations are not descriptions of every current feature. Read the usage guide before following the implementation record.

1. Repair account and note persistence

The early registration work validated identity/password input, seeded accounts and preserved failed form input. The then-selected Neon HTTP adapter supported a batch operation rather than interactive transaction callbacks. Its mocked-transport checks were explicitly not evidence of database rollback.

Building Pageweft: extraction to a hosted workbench

Fri, 16 Oct 2026 00:00:00 +0000

1. Establish actual extraction

The initial requirement was private PDF-to-Markdown preparation using MinerU on the owner’s Modal account. The design moved from customer-provided Modal credentials to owner-hosted processing; customers do not provide cloud tokens. MinerU 2.7.6 runs in a cloud GPU image and retains original engine output. The laptop’s 2 GB GPU is not required.

A bounded one-page public trial was verified before describing full-book processing. The initial deployed web startup failed because modal_fn.py was missing from its uploaded source. Explicit inclusion fixed the image, followed by a fresh deployment and public workflow checks. Original verification record .

Building Sonaweft: speech, source and recovery

Fri, 16 Oct 2026 00:00:00 +0000

Source chronology

The original instascript command and import paths remain compatible. Hosted workbench 7fde01a added the upload/review/export workflow. 5ac1d7c added the original completed audio example; handoff 68b8472 normalized artifacts and corrected the provider reference. The old optional Claude/vault documents are archives, not required setup.

1. Preserve source and recover work

The standalone CPU CLI uses FFmpeg/FFprobe and faster-whisper. Immutable original recognition, separate corrections, generated notes and sourced questions became distinct artifacts. Private checkpoints, exclusive locks, completion hashes and atomic publication support deliberate recovery; cross-path perceptual deduplication is not promised.

Building the vanilla Keep demo

Fri, 16 Oct 2026 00:00:00 +0000

Keep the two implementations separate

This is the vanilla JavaScript/localStorage application, not Noteweld. Native DOM modules, Webpack, checklists, labels, colors, pin/archive/trash and browser storage form its implementation. Existing source history includes f6bc24ce for labels/search/filter/demo notes and 64471b8d for an unused import cleanup.

The 16 October reliability release incorporates the preserved local changes and intentional tracked-dependency cleanup. Source commit a03b0ed1 publishes the browser reliability work; the portfolio refresh copies its production bundle. Earlier statements below describe the baseline they reviewed.

How to review these projects

Fri, 16 Oct 2026 00:00:00 +0000

Start with behavior

Open a project’s usage guide and its live or completed example. Identify the intended action, what should persist, the expected failure behavior and the current limits. A polished cover illustration is not a running backend.

Follow the implementation

Read the technical overview, then the specific architecture/editor/connections manual. Imported documents link an exact committed source revision. A local documentation edition is marked when it includes uncommitted documentation changes; the link then identifies the committed baseline.

Neovim: find, edit, inspect

Fri, 16 Oct 2026 00:00:00 +0000

Follow a small loop

Find a file, make a focused change, inspect diagnostics, format it, and run the relevant check. The editor shortens that loop; it does not establish that a change is correct.

The configured leader is Space. These mappings come from the published Neovim source , inspected alongside the local checkout.

Action	Configured mapping	Tool
Find project files	Space, s, f	Telescope
Search file contents	Space, s, g	Telescope live grep
Remember a working file	Space, a	Harpoon
Open the remembered-file menu	Ctrl + e	Harpoon
Format a buffer	Space, f	Conform
Move between editor splits	Ctrl + h / j / k / l	Neovim window navigation

Language support through Nix

The Neovim wrapper supplies language-server and formatter executables: gopls, TypeScript language server, nil, Pyright, clangd, Lua language server, Marksman and rust-analyzer among them. Formatting tools include Stylua, nixfmt, gofumpt, Black and Ruff.

Build, activate, verify, recover

Fri, 16 Oct 2026 00:00:00 +0000

Review before changing the system

Inspect Git state and the relevant diff. Back up files being changed privately. Preserve unrelated work, the current lock file and a known working boot generation. New files referenced by a Git-backed flake must be deliberately included in its source.

Use a Git-filtered flake source. An explicit unfiltered path: source may include ignored private files before a build-time check can reject them.

Match the check to the change

Change	Relevant evidence
Documentation	Generated pages, links, keyboard and browser checks
User-profile application	Package build, executable, launch and a real file operation
NixOS service	Option evaluation, system build, activation and actual service behavior
Graphics/kernel update	Supported hardware branch, complete build, rollback route and controlled boot acceptance
Backup procedure	Retained-data fixtures, successful real copy and a separate restore test

For this machine, the nr shortcut runs sudo nixos-rebuild switch against the NixOS flake. It does not decrypt secrets, stage files or commit changes. Administrative authentication remains an operating-system boundary.

Maintaining this documentation library

Fri, 16 Oct 2026 00:00:00 +0000

Source and publication

The actual portfolio repository contains documentation/ for Hugo source and manuals/ for generated static output. The existing GitHub Pages root serves both the portfolio and /manuals/. No application server, visitor login, external font, analytics service or search API is needed.

Hugo 0.156.0 is the recorded build version. Imported project README/docs files are explicitly listed in documentation/manifest.json; this is not a recursive copy of the workspace. Current usage guides and journal entries are authored directly in Hugo content.

Building the portfolio and Fieldnotes

Fri, 16 Oct 2026 00:00:00 +0000

Portfolio publication

The active site source is the standalone m-amir-gomaa.github.io repository. Its static HTML/CSS/JavaScript redesign was published before adding the source-aware Pageweft sample viewer, corrected contact address, Sonaweft audio example and Noteweld links. Covers that illustrate a workflow remain labeled illustrations.

Selected published revisions give useful audit anchors:

Revision	Change
`52154f9`	Document Studio sample and contact correction
`946645e`	Pageweft branding
`5ca6dc8`, `9bb3adb`	Sonaweft workbench and completed example
`5ef281e`	Noteweld reliability preview
`394b6c3`	Public Markdown/Vim editor
`ca00da5`	Connected notes and graph example
`bc141c3`	Hugo Fieldnotes library

Each source/application release has separate evidence; a passing portfolio check does not establish private persistence, model accuracy or payment delivery.

Neon, Modal costs and opening Paddle

Fri, 16 Oct 2026 00:00:00 +0000

Neon in this application

Neon hosts PostgreSQL. Noteweld connects through standard pg/Drizzle and stores accounts, notes, labels, revisions and recovery/session state there. Modal runs the Node application; the two providers have separate roles. Moving the database later needs migration/connection configuration, not a replacement for SQL.

Neon supports autoscaling, idle suspension and branching. Its advertised Free tier currently includes 100 CU-hours and 1 GB per project. That is a published allowance, not verification of an individual account plan. Neon overview , pricing , checked 16 October 2026.

Running the queue simulation

Sun, 18 Oct 2026 00:00:00 +0000

Status: retained engineering exercise; a practical replacement is being evaluated. This is the former Queue Lab/Hermes implementation. A worker pool alone is not a differentiated service, and this project is not presented as a paid SaaS or production queue.

What is implemented

Fifty fixed Go workers and a bounded 10,000-task queue.
Whole-batch admission with HTTP 429/Retry-After under pressure.
Strict JSON/body/count validation, bounded retained batches and event subscribers.
JSON metrics, batch status and server-sent progress events; slow consumers disconnect.
Graceful shutdown drains accepted simulated work.
A browser dashboard with bounded recent events and polling recovery.

Tasks simulate a 50 ms delay. No webhook, LLM, scientific computation or paid customer work is performed. State is in memory and disappears on restart. The simulation is useful for studying concurrency mechanics, not evidence of durable delivery or production throughput.

Source and implementation notes

Sun, 18 Oct 2026 00:00:00 +0000

A free, browser-only notes and checklist demo built with native DOM APIs, CSS, localStorage and Webpack. It shares Noteweld’s product family and deliberately has no account, server or paid features. The existing Keep repository and public URL are preserved.

Live demo · Source · Portfolio and manuals

Features

Create and edit notes and checklists; pin, color, label, archive, trash and restore them.
Search titles, bodies, labels and checklist text from either search control.
Use keyboard navigation, named controls and a focus-contained edit dialog; Escape saves and returns focus.
Export versioned JSON including archived/trashed notes and labels; validate and merge imports without replacing existing IDs.
Preserve drafts when a storage write fails, display an error and allow retry.
Recover raw stored JSON if existing notes are malformed, without silently resetting user data.
Use a responsive layout with readable colored notes and reduced-motion support.

Delete from Archive moves to recoverable Trash. Delete forever and Empty Trash require confirmation. There is no automatic seven-day purge. Header Refresh reloads this browser view; the layout control switches compact display. Placeholder reminders/settings are omitted from the available controls.

Technical overview

Sun, 18 Oct 2026 00:00:00 +0000

Public address and existing-account access

Account enhancement work on feat/account-signin is local and gated; setup, behavior and remaining acceptance . Existing public hosting is unchanged. Google/GitHub login is not publicly verified by these source changes. A private notes/checklist preview built with Next.js 16.3.8, React 19, Auth.js and PostgreSQL/Drizzle. This repository retains its original keep-nextjs GitHub identity. The old Vercel deployment is a historical version; it is not updated by this release.

Technical overview

Sun, 18 Oct 2026 00:00:00 +0000

Formerly Document Studio; renamed with Amir’s approval on 15 October 2026. Canonical local path: ~/Projects/freelance/pageweft, with the old directory kept as a compatibility symlink. Existing public GitHub/Modal URLs and permanent storage/session/secret identifiers are retained so bookmarks, accounts and credits keep working.

Hosted PDF-to-Markdown conversion for books and technical material, using MinerU on Amir’s Modal account. Inspect selected pages, review issues, correct text and export portable files for your own editor or LLM workflow.

Technical overview

Sun, 18 Oct 2026 00:00:00 +0000

Open Sonaweft · Hear completed example · Product contract · Operations · Release evidence

Turn authorized recordings into timestamped transcripts, editable drafts, subtitles and portable Markdown. Hear each segment while correcting it. Original speech recognition, your corrections and optional AI notes remain separate. Formerly instaScript / Transcript Studio; existing commands, Python imports and GitHub URLs remain compatible.

Hosted workflow

Upload English audio/video, up to 64 MiB / 60 minutes; inspection returns a quote without starting transcription.
Confirm media rights, then request a bounded 30-second sample or paid processing when commercial access opens.
Optionally consent to sending transcript text to DeepSeek for standard/deeper review. Audio is not sent to DeepSeek.
Follow the job, hear timestamped segments, search text and save separate corrections with revision checks.
Export original Markdown/text/JSON/SRT/VTT, corrected Markdown/JSON, optional summary/research questions and a hashed instascript-export/v1 manifest.

AI suggestions never silently replace the original. Source questions quote actual segments; explicit Crossref lookup returns scholarly metadata candidates, not factual verdicts. Hosted URL downloads, diarization, speaker identification and certified transcription are not implemented; use an authorized upload or the local CLI.

Conversion product design

Sun, 18 Oct 2026 00:00:00 +0000

Implementation update

The approved core features are now implemented: full-book/range inspection (bounded at 64 MiB/1,000 pages), batch inspection, quotes, bounded section processing, selective retries/force OCR, source-linked review, corrections/history, portable outputs and prepaid-credit accounting. Read BOOK-RELEASE.md for tested evidence and limits. Earlier candidate language below describes the design rationale. Actual merchant activation and comparative difficult-document fidelity remain unverified; these features do not certify every source symbol.

Pageweft: conversion utility

User clarification, 14 October 2026: PDF-to-Markdown for serious work with books, including Linguavell preparation. Users control the downstream reading/LLM workflow. No study platform, AI chat, automatically generated teaching materials or decorative feature expansion.

Markdown and Vim editing

Sun, 18 Oct 2026 00:00:00 +0000

Open a text note or choose Take a note…. Checklists keep their separate checklist editor.

Write, Preview and Split change the view without discarding source or enhanced-editor undo. Split stacks vertically on narrow screens.
Standard keys work by default. Vim enables normal/insert/visual modes, motions, selection and Vim undo. Use i to insert, Escape to return to normal, v to select and u to undo. Escape in Vim stays inside the note.
Simple text provides a native textarea for accessibility, device compatibility or enhanced-editor loading failure. Switching to/from it starts fresh enhanced-editor undo; source is retained. Settings apply to this editing session.
Undo, Redo, Find and the formatting buttons work in the enhanced editor. Find includes replacement. Tab leaves the editor instead of trapping keyboard focus.
Save or Ctrl/Cmd-S writes through the normal API. Vim :w uses the same path. Existing notes stay open after these saves; Save & close also closes. Saving a new note completes its creation. Escape/backdrop closure asks before discarding an edited note; it does not silently save it.
A failed save leaves the draft and undo available. If another tab changed the note, copy or download the draft before Reload saved version. Reloading is a deliberate replacement and clears the old draft’s undo history. There is no durable server revision archive yet.
Download .md exports the exact current body, including unsaved changes, with the title as a portable filename. It does not insert a heading or modify stored data. Account JSON export still includes all notes and labels.

Privacy and limits

Markdown rendering, syntax and Vim run in the browser. No AI request, external font or paid processing provider is added. Preview supports CommonMark and GFM tables/task lists. Raw HTML is ignored. Only explicit HTTP(S), mail and fragment links are clickable; application-relative and script/data/file links are disabled. External links open separately without a referrer. Images display their description and an optional external link, never an automatic image request.

Product contract

Sun, 18 Oct 2026 00:00:00 +0000

Amir explicitly made the transcript product active ahead of Keep, delegated a stronger brand and asked for hosted ASR/LLM processing, Paddle and a Pageweft-like usage model. Sonaweft is the selected brand (bounded exact-name searches found no obvious matching software listing; this is not trademark clearance). Preserve instascript imports/command and existing remote URLs. Noteweld is the proposed next Keep brand; its implementation remains deferred.

Source decisions: CLI-001/002/003/005/009/010/013/022/023/033/037/041 and the current chat. Broad audiences and good defaults; English first; user-authorized media; Linux CLI/self-hosting plus hosted UI. Success: ingest media, hear/read timestamped ASR, keep original text, save separate corrections, optionally derive a faithful summary/research questions, export Markdown/JSON/SRT/VTT and a versioned bundle. AI review must not silently change the source or pretend to verify facts. A selected all-or-none hosted package is charged only after all its deliverables complete; retained private ASR checkpoints support explicit retry.

Dependencies and licensing

Sun, 18 Oct 2026 00:00:00 +0000

Verified 14 October 2026. The extraction worker installs the official PyPI release mineru[pipeline]==2.7.6 and invokes its CLI without patching it. Official source release: mineru-2.7.6-released , commit e31c0ec34da89f6d7db78866ffec13eb0aaa3a3c. That release’s LICENSE.md is GNU Affero General Public License version 3. Current master licensing must not be substituted for the pinned release’s terms.

MinerU attribution and the exact upstream source link are visible in the UI and README. Pageweft’s application source is published in its existing public repository. This note does not relicense the owner’s wrapper or certify that subprocess separation resolves every license boundary. Any engine/model changes need their own exact-release source and license review. A complete transitive dependency/model license inventory has not been produced; package version pinning alone does not establish it.

Links, backlinks and graphs

Sun, 18 Oct 2026 00:00:00 +0000

Choose Link note while writing to insert a stable reference such as [[note:UUID|Readable caption]]. Search titles/labels, select a note and insert; the editor keeps your cursor position and undo. Escape closes the picker and returns focus. A stable ID survives a rename; its optional caption remains the text you wrote, while the link tooltip shows the current title. Copy a stable reference from any note’s detail page too.

You can also type [[Title]] or [[Title|caption]]. Matching uses trimmed, case-folded NFKC titles across your own active and archived notes. Exactly one match resolves. Duplicate titles show ambiguous; missing titles show unavailable. These are name lookups, not permanent identity: use the picker for links that must survive renames or replacement by another note with the same title. Empty titles cannot be looked up by name; link them by ID. No note is created or rewritten automatically.

Processing architecture

Sun, 18 Oct 2026 00:00:00 +0000

Browser --> secure cookie + CSRF --> FastAPI / Modal
 |--> serialized SQLite ledger writer
 |--> private media volume (48-hour TTL)
 |--> confirmed CPU coordinator
 |--> FFmpeg normalization
 |--> L4 / faster-whisper large-v3
 |--> optional DeepSeek text review
 |--> committed result --> settle hold
Review player / separate corrections / portable exports <--|
Explicit selected claim --> Crossref metadata candidates

Isolated checkout --> Paddle --> signed durable inbox
 |--> fresh transaction + all adjustment pages
 |--> intent/owner/product-bound credit or revocation

instascript/contract.py validates original ASR and exact source-quoted review and creates versioned exports. pipeline.py provides local private checkpoints, exclusive locks, completion hashes and atomic publication. review.py bounds text/output, validates JSON and has no autonomous tools. Local source.py downloads are never invoked for public uploads.

Deployment and recovery

Sun, 18 Oct 2026 00:00:00 +0000

Scope and least privilege

The hosted preview uses a new database and login role on the existing Neon project, separate from the old Keep database. The original credential files and original two accounts are preserved. Migration credentials remain private on the owner’s machine; only the new runtime role is sent to Modal. It has connection/schema-usage/table CRUD privileges, no database/schema creation or role-management privileges. All app queries enforce ownership; PostgreSQL RLS is not claimed.

Owner operations

Sun, 18 Oct 2026 00:00:00 +0000

Run from the Pageweft repository with the existing authenticated Modal profile. There is no publicly accessible admin endpoint or new monitoring subscription.

.venv/bin/python -m scripts.operations report
.venv/bin/python -m scripts.operations pause
.venv/bin/python -m scripts.operations resume
.venv/bin/python -m scripts.operations budgets --daily 10 --monthly 25
.venv/bin/python -m scripts.operations backup
.venv/bin/python -m scripts.operations verify-backup .local/operations/backups/ACTUAL-FILENAME.fernet

report writes private .local/operations/dashboard.html and JSON. Open the HTML locally. Refreshing means rerunning the command; this is a generated snapshot, not a live stream. It retrieves the actual current-cycle Modal workspace billing summary through the installed SDK and displays it separately from internal estimates. Missing provider data is reported as unavailable, never assumed zero. Aggregate activity includes bots/reloads; it does not prove unique users, conversion rates or causal attribution. Recognized landing tags: ?utm_source=portfolio, github, obsidian or research; arbitrary values collapse to other.

Owner operations

Sun, 18 Oct 2026 00:00:00 +0000

Live: https://sonaweft.amirgomaa.com (custom domain / recovery ); original Modal endpoint remains available.

Observed account state

The existing owner Modal account hosts sonaweft. Volumes: sonaweft-media-v1, sonaweft-ledger-v1, sonaweft-models-v1. Secrets: sonaweft-session, sonaweft-payments, sonaweft-llm. Large model weights are cached only in the cloud.

The existing verified-email Paddle seller has a separate Sonaweft SaaS product, one-time USD $5/$10 prices, client token and paused notification destination at the live regional hostname’s /api/payment-event. The account-scoped runtime key is currently reused from Pageweft; product/prices/intents are checked independently. Private setup values are ignored under .local/integrations/.

Connected notes release

Sun, 18 Oct 2026 00:00:00 +0000

Original record: Connected notes release — 16 October 2026.

Milestone 005 adds stable ID wiki-link insertion, safe code-aware resolution, explicit missing/ambiguous titles, owned metadata, linked-note detail pages, outgoing/backlink panels, a bounded/filterable graph and an account-free original graph example. See the guide for exact syntax, limits and architecture.

Local evidence

Forty unit/React checks and 14 actual PostgreSQL checks passed, including parser escapes/code/GFM URL boundaries, rename/ambiguity/foreign-ID semantics, maximum graph bounds/layout and actual archive/trash/foreign metadata isolation. Lint, TypeScript and the isolated production build passed. All 14 production Chromium checks passed after the final graph fitting/picker refinements, including two actual accounts, stable links across rename/trash/restore/permanent deletion, no rebinding to a replacement title, ambiguous names, picker Escape/focus, dirty link navigation, graph filters/camera and simulated metadata failure/retry. The original authenticated account/recovery, editor/Vim/export and failed-editor-chunk journeys also passed. The account-free graph was verified to make no private API requests. A further public-graph-only capture passed for visual review, without registering accounts. Runtime audit is clean at this date, with previously documented dev-only findings retained separately. Graph, editor, forms and public examples fit 320/390/768/1440. Phone/desktop graph screenshots were visually inspected; fit now uses visible bounds and SVG labels/dots retain a constant screen size. Seven supporting portfolio DOM checks passed; its publication remains pending hosted acceptance.

Hosted release evidence

Sun, 18 Oct 2026 00:00:00 +0000

Original record: Release evidence — 15 October 2026.

Sonaweft is deployed at https://m-amir-gomaa--sonaweft-web-app.us-east.modal.run . /health reports hosted processing enabled and billing disabled. Existing Pageweft cloud identities, balances and files were not migrated.

Check	Actual evidence	Scope
Automated suite	52 tests passed, Python safety lint and both JS syntax checks passed	Real temporary SQLite/files/FFprobe; model dispatch and Paddle HTTP are mocked
Local speech	Existing cached small model transcribed the original generated recording in about 5.4 seconds	CPU/int8 on this laptop; not an accuracy benchmark
Cloud speech	Pinned large-v3 revision on Modal L4 returned the expected two English sentences	Real inference, cloud-cached weights; not local MX350 inference
Optional review	Real configured DeepSeek Flash returned schema-valid notes; combined cloud processing reached ready	Actual text provider, original preserved; advanced model listed by API but not inference-benchmarked
Public workflow	Chromium uploaded/inspected the generated recording, confirmed a trial, waited for actual cloud ASR, opened source audio, saved a separate correction and downloaded a ZIP	Public HTTPS and real workers/storage; no customer media or purchase
Responsive browser	320/390/768/1440 workspace/review widths and 20 policy/checkout cases passed with zero page errors/overflow	System Chromium; desktop/phone screenshots visually reviewed, not physical iPhone/Safari certification
Source lookup	Real Crossref request returned DOI candidates; tests exercise malformed results/outage and route consent/ownership/rate limits	Bibliographic matches only; no truth verdict or general search
Recovery/accounting	Once-only reservation/settlement, partial/full refund, chargeback, outage, replay, ownership and revision cases passed	Provider responses mocked; SQLite durable and concurrent checks real
Owner backup	Actual remote financial snapshot encrypted/downloaded/decrypted into a disposable SQLite clone; integrity `ok`, zero balance violations, zero media rows	Financial recovery snapshot, not media restore or production rollback
Nix	`nix build .#default --no-link` succeeded; built CLI actually transcribed the generated audio and published a complete hashed export	Nixpkgs Python 3.14.6 / faster-whisper 1.2.1 / PyAV 17.0.1; no system rebuild
Portfolio	Seven DOM tests and four Chromium viewport checks passed for the updated card/link/details	Sonaweft linked explicitly; Noteweld remains a proposal
Completed example	Original audio, actual ASR/review and downloadable bundle; four browser widths, decoded audio playback/seek and ZIP hash validation passed with no processing/API job calls	Public precomputed evidence, separate from the bounded live sample
Paddle setup	Separate real product, fixed $5/$10 prices, client token and paused notification created; Sonaweft hostname submitted and observed In review	Seller identity approval, real checkout/webhook/refund/payout remain unverified

The initial cloud probe exposed cross-environment exception/NumPy serialization failures; native JSON values and a plain error envelope fixed them before public enablement. Local testing found PyAV 19 incompatible with the pip speech dependency combination, so pip/cloud use the tested 15.1.0 pin. Nix supplies a separately tested compatible pair. The launcher also handled a Nix library-discovery SIGPIPE before release.

Payment activation checklist

Sun, 18 Oct 2026 00:00:00 +0000

Updated 16 October 2026. Paddle Billing is the selected provider. The integration is deployed; customer checkout remains disabled. This supersedes the former Lemon Squeezy activation instructions. Existing balances, session keys and permanent Modal identifiers were preserved. Email verification is complete according to Amir; merchant identity/domain approval and actual payment delivery are separate.

Latest session observation — 16 October 2026

The initial dashboard navigation displayed Paddle Login. After Amir confirmed signing in, the existing authenticated tabs were inspected and refreshed: Pageweft and Sonaweft both show In review under Website approval. The existing Tell us more about yourself verification form has empty date-of-birth and current-trading-name fields; it was left open for the owner to complete truthful personal details. No new seller account/signup, identity submission, payout change, billing activation, real purchase or email was performed. This supersedes the initial login-only observation; approval is still not established.

Historical agent setup

Sun, 18 Oct 2026 00:00:00 +0000

Why this record is separate

This optional discovery document predates the hosted Sonaweft pipeline. Its provider/model/agent setup was not validated by the current release. No agent installation is needed to upload a recording, use the supported CLI or export files. Jev is retired and is not part of this workflow.

Read the current usage guide and project index first. The historical Markdown remains available at the source revision linked above; its commands, provider assumptions and completion claims are not current instructions.

Launch readiness register

Sun, 18 Oct 2026 00:00:00 +0000

Updated 15 October 2026. Pageweft is a working, bounded PDF-to-Markdown utility. Commercial demand, representative extraction quality and real payment delivery remain separate evidence. Initial spending and feature scope should stay small until those are measured.

Milestone	Current evidence	Remaining action / completion evidence
Functional conversion	Synthetic whole 1,000-page CPU conversion; selected-page real MinerU; public sample/edit/export and ownership checks	Representative technical/scanned PDFs: compare output and manual correction time; no perfect-fidelity claim
Owner visibility and safety	Private owner report, admission caps/pause, credit identities, aggregate events, consistent encrypted snapshots; recorded actual Modal billing is retrievable	Verify provider-level out-of-pocket limit in owner dashboard; independently store backup/key; review actual job costs
Payments	Transactional grants/refunds/holds, signatures/scope verification and local mocked cases	Actual seller identity approval and processing-pack acceptance; real isolated Paddle checkout/webhooks/refunds and recovery; actual live purchase/settlement/payout
Customer readiness	Limits/retention/credit rules visible; billing closed; Amir supplied mo.gomaa.formal@gmail.com for support	Actual seller details and final refund/privacy terms; dependency/model license inventory before paid opening; inbox delivery unverified
Discovery	Original public conversion guide, actual downloadable PDF/Markdown/ZIP, canonical pages, robots/sitemap, aggregate source tags	Owner Search Console verification and indexing request; permission-based first-user recruitment and measured purchases/repeat use
Viability	Conditional arithmetic and per-attempt cost controls; no invented sales	Measured contribution after fees, failures, support and acquisition; willing returning buyers; recalibrate tariff/caps from evidence

Next steps, in order

Read operations and check the actual dashboard. Select a workspace-wide net spending ceiling, recommended $0 out of pocket during validation while available credits permit work. Saving this limit is an owner account operation, still unverified; it affects the other work using the same Modal account. Internal admission controls are already independent.
Read payment activation . Complete the store’s real identity/business questionnaire using truthful details. Paddle is now selected: catalog/runtime/webhook configuration is present, the actual app domain is submitted for review, and the existing verification flow is at personal details. Do not collect money until that decision and the actual payment journey are verified.
Support email is confirmed as mo.gomaa.formal@gmail.com . Decide the seller’s actual published details/refund policy. Use one product scope and one processor; adding more providers is unnecessary until a specific failure requires an alternative.
Run the quality comparison on 10 permission-cleared PDFs containing scanned pages, columns, tables, equations, diacritics/IPA and images. Record source-page ground truth, missing content, correction minutes and each competitor’s total price. Existing private books are not public examples; seek rights/permission before publishing excerpts. Decide using measured usefulness, not fabricated accuracy percentages.
Ask 10–20 people who already need Markdown to try their own documents. Begin with the remaining bounded sample allowance, not unlimited free full books. Pricing and discovery contains the script and milestones. Owner decides/executes outreach; no emails or posts were sent by this release.
After payments work, target five genuine buyers and three returning for another document/top-up. These are pragmatic decision thresholds, not statistical proof. Record refund/support time and attributable costs. Expand only if customers get a useful export and paid work contributes positively before fixed costs.

Markdown editor release

Sun, 18 Oct 2026 00:00:00 +0000

Original record: Markdown/Vim release — 16 October 2026.

This collection follows the reliability foundation . It adds Markdown source/preview/split, optional Vim, in-session undo/redo, find/replace, formatting, exact Markdown downloads, guarded dialog closure and a public browser-only editor example. Checklists, account recovery, private API boundaries and revision conflicts remain intact. No database schema, original Keep data, billing configuration or NixOS settings are changed.

Evidence

Local validation passed: 33 unit/React boundary checks, 13 actual PostgreSQL checks, lint, TypeScript and the isolated production build. All 11 production Chromium checks passed, including the existing real-cookie/database/recovery journey and the new editor journeys. All three editor journeys passed again after the final mobile split-height adjustment.

Account reliability release

Sun, 18 Oct 2026 00:00:00 +0000

Original record: Noteweld foundation release — 15 October 2026.

This milestone repairs account/note/label reliability before the connected Markdown/Vim workspace. It is a free bounded preview. Paid access, editor/graph and source-bundle imports are separate future collections, not released features.

Local evidence

npm test: 27 unit/React boundary tests passed, covering auth validation, driver contracts, protected fields, pending/failed UI and idempotent draft retries. Network/auth substitutions in these tests are explicit.
npm run test:database: 13 actual PostgreSQL checks passed. They cover account/seed commit and rollback, duplicate signup races, ownership, concurrent note revisions, trash constraints, idempotent creation, quotas, rate limits, rotating recovery/session versions, label races, migration collision rollback and logical snapshot restoration.
npm run lint, npm run typecheck, isolated production build: passed. nix eval .#devShells.x86_64-linux.default.name evaluated the pinned Nix shell without a system rebuild.
Chromium full cookie/database journey passed locally, including registration/recovery acknowledgment, failed-create retry, exact whitespace persistence, stale-edit recovery, checklist mutation/trash/restore, label rename identity, phone navigation, JSON export, password recovery invalidating an existing session and permanent account deletion. All 8 combined browser checks passed after the standalone build.
Auth forms and workspace widths are checked at 320/390/768/1440; 390/1440 screenshots were visually inspected. The native note dialog focuses the title; the mobile drawer supports Escape and focus return. No authenticated journey is permanently skipped.
Compatibility validation entrypoints import maintained project tests and pass. Private source backup precedes this work; pre-existing changes were retained.

Isolation and deployment

A new Neon database and restricted application role were created. Migrations were applied only to that empty target and the isolated local database. The original database still has two accounts and no notes/labels; its credential files were not edited. The runtime role is verified unable to create database/schema/roles or read the original accounts.

Book conversion release

Sun, 18 Oct 2026 00:00:00 +0000

Original record: Book conversion release — 14 October 2026.

Implemented in the hosted Pageweft application; supersedes the one-page-only product and $1/month access proposal. Amir delegated a better pricing model and approved the conversion utility improvements. The public free sample remains one selected page, but inspection accepts books and paid jobs use separately funded credits.

Features and implementation

Capability	Implementation
Books / ranges / batch inspection	Up to five PDFs, each <=64 MiB and 1,000 source PDF pages; strict PDF validation and page-range parser; retained inspected PDFs can be resumed. Each conversion is separately confirmed.
Preview and price before extraction	Source page count, selected pages and text-layer samples are inspected without GPU work; fixed tariff quote must match the confirmation.
Automatic / force OCR / cheaper text-only	Pinned unmodified MinerU 2.7.6 uses its documented mode/language options; pypdf text-only worker uses no GPU and explicitly excludes OCR, image extraction and layout reconstruction.
Bounded sections / partial recovery	Selected pages split into groups of 20; at most two GPU workers. Successful sections survive failures; explicit retries select missing or chosen original pages and can force OCR. Duplicate requests cannot reserve or dispatch twice.
Traceability / fidelity review	Original page indices are mapped from the subset PDF back to the full source. Renderer preserves recognized headings, equations, HTML tables, captions, footnotes and images. Unhandled blocks, unmapped content, empty/short output, missing images and replacement glyphs are reported. Untouched engine output is retained separately; flags do not certify correctness.
Corrections / history	Plain Markdown editor with original-PDF page link, optimistic edit versions and saved history. Reprocessing retains user corrections. Engine text can be loaded explicitly into the editor draft.
Portable outputs	Single Markdown, individual pages, bounded page groups, or sections based on original PDF bookmarks. Local image paths, source hash, unavailable-page list, raw section extraction and review metadata. Source PDF inclusion is optional.
Cancel / discard / delete / expiry	Queued work releases reserved allowance. A running cancelled section may finish and be charged. Job deletion removes edits/output and releases an unshared source. Unused inspected sources can be discarded. Hourly cleanup removes 48-hour-expired files; balances/recovery persist separately.
Account / purchase recovery	Encrypted Secure/HttpOnly session; high-entropy recovery key stored only as keyed hash. One-time purchase keys can restore the owning account. No customer Modal credentials.
Payments / refunds	Separate one-time pack integration verifies scoped provider orders and signed events; duplicate orders grant once, full/partial refunds revoke pack credits. Atomic reserve/settle/release accounting in integer 1/10,000 USD units. Public test-mode grants disabled. Live seller configuration remains absent.

The active HTTP factory is web/studio.py. web/main.py and web/billing.py preserve the retired interface/verifier for regression/compatibility work; they are not the public paid implementation. Books use /ledger/books.sqlite, separate from the original /ledger/state.sqlite. Each is owned by one serialized function. Do not run a second Modal app with a separate book_store function on the same volume. Owner verification scripts invoke deployed functions instead.

Historical demo script

Sun, 18 Oct 2026 00:00:00 +0000

Why this record is separate

The old recording script includes illustrative output, legacy paths, a third-party media example and cleanup instructions. It is not current acceptance evidence or a recommended command sequence. Use the original generated hosted example and the release evidence instead; no third-party recording is republished here.

Accessibility review — 16 October

Sun, 18 Oct 2026 00:00:00 +0000

Original record: Editor accessibility review — 16 October 2026.

The public scratchpad’s CodeMirror scroll region was flagged by axe-core 4.13.0 as lacking focusable content. The editable textbox now has an explicit tabindex=0 through CodeMirror’s contentAttributes facet, including its read-only states. Tab still exits the editor normally. This does not add a second focus target around the editor or change data, save, account or payment behavior.

Local verification: lint/types, all 40 existing unit/UI checks, isolated production build, and the existing account-free editor browser journey passed. The journey checks real Markdown download, preview/split, guarded reset/navigation, four viewport widths, Tab exit, strict production CSP and absence of POST requests. A separate local axe WCAG A/AA/2.1 AA run returned zero violations on the public example. No database/schema change was needed or made; this patch does not claim a new run of the full authenticated PostgreSQL suite.

Historical vault design

Sun, 18 Oct 2026 00:00:00 +0000

Why this record is separate

This is an earlier optional Obsidian/agent design, not a shipped hosted feature. Research matches do not establish truth and language-model notes do not replace the immutable recognition source. The current architecture documents the supported pipeline. Noteweld imports and shared login remain future work.

Paddle integration evidence

Sun, 18 Oct 2026 00:00:00 +0000

Original record: Paddle integration evidence — 15 October 2026.

Delivered

Paddle Billing replaces Lemon Squeezy as the deployed selected provider. Customer billing remains closed. The existing signed-in live account was used; no new seller/sandbox signup was completed. Runtime credentials are separated from setup credentials. Permanent app, volumes, session secret and customer credit identities were preserved.

Official API observations: one SaaS product, two one-time USD prices ($5/$10), a client token, and a platform-only notification destination were created successfully. Notification delivery is paused until actual activation. Runtime reads of transactions/adjustments returned 200, catalog read returned 403 as intended, and a customer-free unpaid checkout validation returned 400 transaction_checkout_not_enabled. No charge was made. The actual Modal application hostname was submitted through the dashboard; the API reports pending_review.

Custom domain and account access

Sun, 18 Oct 2026 00:00:00 +0000

Public address: https://noteweld.amirgomaa.com/ . The account-free editor and graph are /example and /example/graph; private writing starts at /auth/signin. A dedicated Cloudflare Workers Free gateway fronts the existing Modal deployment. The same isolated Neon database, restricted role, AUTH_SECRET, BACKUP_KEY and backup jobs remain in use. Sign in again with existing credentials at the new host; browser cookies do not transfer between hostnames. No database migration or account merge is required.

This release branches from production 5ac512f, adding authenticated client-IP forwarding and the independent noteweld-edge Modal secret. It deliberately excludes the unfinished account/social-provider branch. Shared source/config/tests are maintained in freelance/workflow/cloudflare-edge/ in the workspace. Each app has its own fixed upstream, secret and host-only secure cookies. Direct Modal access preserves its previous proxy fallback. CSRF/origin checks and application quotas remain enforced: foreign unsafe requests fail at the gateway, and same-origin requests are mapped to the current internal Auth.js origin. Bodies and private responses stream without caching.

Custom domain and account access

Sun, 18 Oct 2026 00:00:00 +0000

Public address: https://sonaweft.amirgomaa.com/ . A dedicated Cloudflare Workers Free gateway fronts the preserved Modal origin; processing, files and ledger remain on Modal. No paid provider plan, live checkout or new customer credentials are activated by this route. Reproducible shared code/config/tests are in the workspace at freelance/workflow/cloudflare-edge/.

The independent sonaweft-edge Modal secret authenticates short-lived client-IP assertions for existing quotas. Direct origin requests retain their previous proxy fallback. The gateway rejects unsafe foreign Origins before mapping its own Origin for upstream CSRF checks; CSRF tokens and host-only Secure/HttpOnly cookies remain enforced. Request/response bodies stream without document/audio buffering at the gateway. Cache mode is explicitly no-store; never use TTL zero as a cache-disable switch, because Cloudflare can strip session cookies when forcing caching.

Extraction verification record

Sun, 18 Oct 2026 00:00:00 +0000

Original record: Hosted trial evidence — 14 October 2026.

Scope: owner-hosted PDF-to-Markdown trial, not live commerce, academic-document fidelity certification or full-book acceptance.

Automated: .venv/bin/python -m unittest discover -s tests -v — 22 passed. Includes real SQLite persistence and 30 concurrent quota attempts; PDF/session/ownership/archive checks; fake CLI and mocked payment-provider boundaries explicitly remain simulated.
Static: node --check static/app.js, Git whitespace check passed after removing one trailing space.
Real private compute: original one-page PDF processed on A10G by pinned MinerU 2.7.6; readable expected text, original PDF, source mapping and review report in a 2,295-byte ZIP. No copyrighted book was submitted for this probe.
Real hosted HTTP: /health and /api/session returned 200, billing_enabled=false, old /upload returned 404. Upload → job polling → completed ZIP download passed. Duplicate request returned the same job. Separate session received 404 for both status and download.
Browser: Playwright 1.63.0 with system Chromium and chromiumSandbox=true, widths 320/390/768/1440. No horizontal overflow or page errors. Oversize file rejected in browser; invalid PDF rejected by real HTTP service without dispatch; retry reenabled; privacy disclosure keyboard action worked. Phone/desktop screenshots inspected. No physical iPhone or Safari certification.
Deployment: .venv/bin/python -m modal deploy app.py succeeded; public URL https://m-amir-gomaa--mineru-pdf-extractor-web-app.modal.run . First startup failed due to missing modal_fn.py; explicit inclusion fixed it and checks above were run after redeployment.

Private artifacts live in ignored artifacts/: verification.pdf, verification-result.zip, public-verification-result.zip, public-verification.json, browser-verification.json and browser screenshots. No session cookies, CSRF values or provider credentials are in this evidence. One bounded real public trial was consumed; unit and browser-invalid-input checks did not submit additional GPU work.

Pricing research archive

Sun, 18 Oct 2026 00:00:00 +0000

Why this record is separate

This record contains historical fee comparisons, owner income scenarios and a discovery plan. It predates the Paddle selection. Its numerical assumptions are not a current quotation or a forecast. Current service tariffs are in the technical overview; live billing remains disabled.

Unreleased account work

Sun, 18 Oct 2026 00:00:00 +0000

Why this record is separate

The account foundation lives on a separate development branch. Production domain release 550be9c branches from 5ac512f and excludes this account work. This source record does not establish public Google/GitHub login or a completed provider consent flow.

Product decisions archive

Sun, 18 Oct 2026 00:00:00 +0000

Why this record is separate

The resulting choices are owner-hosted extraction, no customer Modal accounts, one-time processing packs, a small bounded sample and separate operational controls. Personal conversation, account questions and private preparation context are not reproduced here. The technical overview and payment checklist carry the actionable public information.

Accessibility review — 16 October

Sun, 18 Oct 2026 00:00:00 +0000

Original record: Retained-document panel accessibility — 16 October 2026.

The inspected-PDF list had an aria-label on an otherwise generic div. Its explicit region role now gives that existing label valid semantics. No job, file, quote, credit, payment or storage behavior changes.

All 78 existing tests passed. The existing Modal deployment succeeded in 4.110 seconds at its unchanged hostname. Runtime secrets, volumes, balances and disabled billing were preserved; no extraction, funded fixture or real purchase was run. Hosted narrow/desktop and automated WCAG checks are recorded in the supporting portfolio release milestone. Unrelated bytecode/book-workflow changes remain outside this commit.

Custom domain and account access

Sun, 18 Oct 2026 00:00:00 +0000

Public address: https://pageweft.amirgomaa.com/ . A fixed-origin Cloudflare Workers Free gateway connects to the preserved Modal endpoint; Modal still runs the app and processing. No Team upgrade or live payment activation was made. The shared reproducible gateway source/config/tests and rollout guide are in the workspace at freelance/workflow/cloudflare-edge/.

Session/payment secrets, volumes, existing balances and processing limits remain unchanged. A separate pageweft-edge Modal secret authenticates time-bounded client-IP assertions from this Worker for rate quotas. Direct Modal requests retain the existing last-proxy-hop fallback. The web boundary still verifies CSRF tokens; the gateway rejects unsafe cross-origin requests before mapping its own trusted Origin to the upstream origin. Uploads/responses stream, cache mode is explicitly no-store, and cookies stay host-only Secure/HttpOnly. Initial live verification caught and corrected a TTL-zero cache override that stripped cookies; do not reintroduce cf.cacheTtl: 0 as a cache-disable switch.