This page is Chiron · Intuition · Tarot — Last 20 Work Products. It is one of the projects Brandon built and put online. The grown-up details are below, tap Grown-up mode to read them the normal way.
CEO-Mind Evaluation · Work-Product Portfolio
Chiron · Intuition · Tarot — Last 20 Work Products
Two real assets buried under seventeen versions of themselves: a 31MB verbatim reader-voice meaning engine and a Matilda/Lester reading-video factory. Right now it is a tooling hobby that ships to Vercel and sells nothing. Stop forking, package twice, list this week.
Generated 2026-06-11 13:35 · grounded in the Work Product Command Center inventory · evaluate → criticize → streamline → monetize
20Products
0Scale
0Ship
13Fix
7Kill
$650–$2,40030-day if executed
The CEO read
This lane is producing two things over and over with different filenames. Engine #1 is the "reader-voice combo-meaning" stack: tarot-voice, tarot-voice-v2, kate-engine-v3, eyephim-combos, kate-combo-fraud-eval, kate-meaning-fidelity, and tarot-v3-gate-results are all the SAME capability at different commits, mostly verdict-JSON and gate dashboards. The genuine asset underneath them is real: a 31MB verbatim meaning_db (kate + chiron) mined from 201 transcripts, 23,370 Kate combo windows, 97.7% pair coverage. Engine #2 is the Chiron reading-VIDEO factory: chiron-lester-2min-test, chiron-matilda-rws-video, chiron-matilda-cards, chiron-chalkboard-matilda, chiron-tarot, and tarot-padbed are all the same chalkboard/Matilda-or-Lester-voiced reading video at different render stages. The pattern is build-tool, evaluate-tool, fork-tool, ship-an-HTML-dashboard-to-Vercel, repeat — and never attach a price. Twelve of these scored 75 only because they are "deployed," not because they earn. The 97.7% number is co-occurrence proximity, not teaching capability, and the strict judge FAILS the synthesized reading at 35%, so the moat is the corpus, not the generator. Verdict: this is an R&D hobby with two productizable cores. It becomes a business the moment the meaning_db ships as a Gumroad reference deck and the video factory renders three saleable card-of-the-day shorts instead of a seventh test render.
Where to put the money — top 3 scale plays
SCALE PLAY 1
Ship the verbatim meaning_db as a paid 'Tarot Combination Codex' — kate+chiron, ~5,300 real combo passages — as a Gumroad PDF/searchable web deck, using the existing tarot-combo-index as the UI.
It is the only thing in this lane a competitor cannot clone — 201 transcripts of real reader language already mined, indexed, and coverage-scored. Generated readings fail the strict judge at 35%, but the RAW corpus windows are gold for any reader, student, or app. You already built the front-end (tarot-combo-index/data.js).
First moveExport kate (97.7% / 2,933 pairs) + chiron (80.5%) from tarot-voice-v2/meaning_db into a deduped, deck-formatted PDF and a gated web build; wire tarot-combo-index as the live preview.
Revenue pathGumroad core deck at $19, 'pro' searchable web version at $39, plus an Etsy digital-download cut at $12 → 30–90 day path $400–$1,500 from a near-zero-marginal-cost asset.
SCALE PLAY 2
Turn the Chiron video factory into a daily/weekly 'Card of the Day' short series — one engine (chiron-matilda-rws-video + tarot-padbed bed), three renders a week, posted to YouTube/TikTok with a Gumroad reading link in bio.
You have a working, shipped video pipeline (Matilda + Lester voices, chalkboard visuals, worship-pad bed) and proof it lands (chiron-tarot, chiron-matilda-cards are live). The bottleneck has been re-testing the renderer, not audience. Shorts are the cheapest top-of-funnel for a reader brand.
First moveLock chiron-matilda-rws-video as the ONLY render path, batch-produce 5 card-of-the-day shorts this week from the meaning_db text, attach padbed, publish.
Revenue pathFree shorts drive traffic → personalized-reading-video SKU at $25–$45 on Etsy/Fiverr + channel monetization → $300–$1,200 in 30–90 days, scaling with cadence.
SCALE PLAY 3
Productize the personalized reading-video as the premium SKU: buyer picks a spread, you render a Matilda/Lester chalkboard reading from the meaning_db in under an hour of human time.
This is the highest price point you can charge and it is fully assembled — video engine + voice clones + worship bed + 5,300 combo passages to source language from. It converts the R&D into a service with margin, and the avoidant-video-pipeline proves the same engine extends to a second niche (attachment-style readings).
First moveWrite a one-page Fiverr/Etsy listing ('Custom Tarot Reading Video, chalkboard + narrated'), set a 48–72h delivery SLA, render the first as the sample using chiron-tarot as the trailer.
Revenue pathEtsy/Fiverr custom video at $35–$60 each; 8–15 orders over 90 days → $300–$900, with the avoidant variant adding a second listing on the same engine.
The scorecard — all 20, ranked
Click any column header to sort. Each row carries its next-best action: open the live page, copy its streamline+monetize plan, or check approve to greenlight it.
#
Product
Verdict
Quality
Critique
Streamline move
Monetize
30d $
Next action
1
tarot-voice-v2produced local
FIX
58
• This folder is internal tooling, not a deliverable. The only HTML (index.html) is a #111 dark coverage dashboard of card-pair heatmaps — a developer artifact, unsellable and unship-worthy as a public page. • Its headline metric is a known-false signal. The '97.7% kate coverage' in _coverage_summary.json measures 400-char name co-occurrence, NOT combination-teaching capability — MEMORY explicitly logs 'don't re-trust it.' The engine's own verdict (39% real fidelity) contradicts the dashboard the folder ships. • It is half-finished and duplicated. 33.9MB / 103 files of overlapping dev passes (combo_extractor.py, kate_combo_extractor.py, kate_16combo_reading.py x3 revisions, _kr_round / _kr_round_v2 / _k16_round, a delete/ dir, pre-eyephim-fix backups). The work is mirrored across _also at ~/Projects/_outputs/tarot-voice-v2 — two copies of the same churning experiment.
Stop treating this folder as a thing to ship. Extract ONLY the kate + chiron meaning_db (the two voices above 80% with real combo windows) into the canonical tarot-voice-v2 reading-delivery path the user already uses, archive the rest of the dev churn into delete/, and retire the misleading coverage index.html entirely. The engine's job is done — its single durable output is the two meaning_db folders.
None · $0
high
Do not list this. Point its meaning_db at the already-shipped chiron-tarot / tarot-combo-index pages and let THOSE be the monetized surface (YouTube chalkboard reading videos funneling to brandonpag3 Substack).
$0
2
chiron-matilda-rws-videoshipped live
FIX
62
• This is a pipeline TEST, not a product. The shipped page literally calls itself a '2-minute evaluation cut' and says the four full ~20-min renders are 'held until this test is approved.' One 2-min clip of one of four videos. • The narration is copied VERBATIM from someone else's copyrighted YouTube reading (Chiron Intuition, 6hTEvbKZUC8) with bleeps restored and filler swapped. That is a re-upload of another creator's reading, not original Brandon IP. It is unsellable and a takedown/strike risk if monetized as-is. • Known unfixed bug shipped: the closing 'tower' card was lowercase so card-detection missed it, caption shows but no card image (index.html admits this). Minor, but it shipped flagged.
Do not treat this as a standalone project. Fold it into the chiron-teaching-gen / chiron-matilda-rws-video skill as the validated render preset, then make ONE original full-length reading (Brandon's own card pull and interpretation, not a re-voiced transcript) as the first actual YouTube upload. Kill the verbatim-transcript narration path for anything public.
YouTube · $0 (ad/affiliate; funnel to $5-8/mo Substack)
low
Replace the copied transcript with one original Brandon-authored reading script (own card pull), render a full 15-min cut on this exact proven pipeline, and upload it as the channel's first real video to prove watch-time before batching the other three.
• This duplicates the canonical engine. Memory feedback_kate_combos_db_pull_not_rebuild explicitly says combos live at ~/New Desktop/tarot-voice-v2/meaning_db/ and must be PULLED, never rebuilt, and that folder already has chiron + kate built. This _outputs/tarot-voice is a parallel rebuild of the same data, the exact macro-consolidation/redundancy the vault warns against. • It is dev tooling and raw corpus, not a product. The .jsonl combos are noisy transcript fragments with char offsets and ASR garbage ('ell]\n>>', cut-off sentences), not cleaned buyer-facing meanings. The verdict JSONs are internal scaffolding. Nothing here is listable as-is. • 5 of 7 reader voices are near-empty: risingsoultarot 0.0% coverage (1 combo), lightassefa 0.3% (12), quanyin 15.1%, eyeaphim 18.3%, m3ksmith 39.8%. Only kate (99%) and chiron (49.7%) cleared a usable bar, so the '7 readers' framing is mostly empty shells.
Fold this into the canonical ~/New Desktop/tarot-voice-v2 engine and delete the parallel build: keep only the chiron + kate JSONL (the two voices with real coverage), move the 4 sample readings into tarot-voice-v2 as the demo set, and retire the 5 empty-shell readers plus the duplicate build_combo_meanings.py rather than maintaining two combo extractors.
Gumroad · $15
low
Take chiron_patched_reading_2026-06-10.md + kate_patched_reading_2026-06-10.md, generate 4 more spreads in that exact voice, run anti_ai_voice_gate, export one PDF, list on Gumroad under the brandonpag3 Occult/Tarot lane.
$30
4
chiron-tarotshipped live
FIX
42
• This is a demo render, not a product. The 540-byte index.html is a single naked <video> tag with one caption line, zero packaging, no title card list, no card index, no CTA, no funnel, nothing to buy. • Provenance is the dealbreaker: the filename and source (wpUCkqhE8v4.txt) trace to a third-party reader's YouTube reading. The teaching/audio is derived from someone else's video, so this specific render is not cleanly licensable or sellable as-is, only usable as an internal engine demo. • It is one of two near-identical engines (chiron-matilda-rws-video and tarot-chapter-voice-video both carry the same transcript_chalkboard.py) producing the same chalkboard format, so this lane has duplicated tooling and only one stray output bolted onto the Vercel site.
Stop shipping one-off MP4 pages to Vercel. Re-render the chalkboard videos from Brandon's OWN sanctioned reader voice (Lester/Brandon XTTS) on ORIGINAL combo-meaning scripts from chiron_combo_meanings.jsonl, upload those to a YouTube channel as the real distribution surface, and retire this single derivative file into a delete/ folder once the originals exist.
Re-render ONE chalkboard video from an original chiron_combo_meanings.jsonl script in Brandon's own XTTS voice, post it to YouTube with a Substack/Etsy link in the description; do not monetize the existing wpUCkqhE8v4 derivative.
• This is a dev showcase page, not a product. It is one demo episode sitting next to four other chiron-* pages (chiron-tarot, chiron-lester-2min-test, chiron-matilda-inventory, chiron-intuition-ceo-eval). Nobody buys this index.html; it has no title, no channel, no CTA, no listing. • The page literally publishes its own three defects: wrong narration voice (Lester XTTS, because ElevenLabs Matilda hit 0 credits, despite the page being named 'matilda'), runtime 7:00 over the 5:00 target, and 3 of 101 caption beats showing a broken '[ __ ]' word gap. A live deliverable with a visible blank in the captions is not shippable to YouTube as-is. • It derives from an intuition-reader's YouTube transcript (WgWy35I643I), so a verbatim narrated rebuild carries copyright/originality exposure if monetized directly rather than used as a format template with original readings.
Do not treat this page as the deliverable. Fix the three named defects in the chiron-matilda-rws-video engine once (re-mux Matilda audio, run the caption-fill pass, trim script to 5:00), then batch-render 8-12 ORIGINAL Chiron-style readings and publish them as a YouTube channel. Kill the standalone showcase page or demote it to an internal engine-QA artifact.
YouTube · $0 per video (ad + affiliate); funnel sells $12 Etsy printable spread + $6/mo Substack
low
Fix the caption '[ __ ]' gaps and re-mux the Matilda voice in the engine, render 4 original readings (not the transcript rebuild), and upload the first to a YouTube channel with an Etsy spread-pack link in the description.
• PROVENANCE / COPYRIGHT RISK is the headline: 3 of 4 files are named after external YouTube video IDs. If the narration is the original intuition-readers' audio (or a transcript re-voice of it), uploading these to monetize invites Content ID claims or strikes. The chalkboard+cards+subtitle treatment is transformative, but nothing in the folder documents where the voice/reading text came from. • Zero packaging. No README, no JSON, no titles, descriptions, thumbnails, tags, or chapter markers. It is a raw render dump named tarot-padbed, four loose mp4s and nothing else. • Bitrate is low (~340 kbps total, ~163 kbps video) for 1080p. Fine for talking-card YouTube, soft if anyone wanted a premium download.
Resolve provenance first (confirm the narration is Brandon's own re-voice of the transcript, not lifted creator audio), then run the four files through one small package step: auto-generate title + description + 1280x720 thumbnail per video from the on-screen cards/keywords, and publish to a single YouTube channel as the lane's top-of-funnel. Do not invent new renders; just title, thumbnail, upload.
YouTube · $0 to view; funnel into Substack $5/mo and Etsy printable spread pack $8
low
Confirm narration provenance, then upload all four to one YouTube channel with generated titles/descriptions/thumbnails and a pinned Substack + Etsy link in every description.
$30
7
kate-engine-v3produced local
KILL
38
• This is internal dev tooling, not a product. Buyers buy readings or reading-voice packs, not a stacked-improvements benchmark report with v1-through-v6 of a single test reading. • The flagship outputs are visibly broken. The v3 sample reading (report line 74) leaks dozens of unscrubbed '[this energy]' placeholder tokens and repeats the same modifier sentence 4-5 times. The report itself flags scrub-marker leakage as 'the single biggest visible quality bug' (line 90). • v3-v6 .txt files are all 0 bytes (empty) on disk; only the highlighted .md mirrors survived. The actual deliverable text is half-gone.
Harvest then retire: lift kate_gap_fillers_vs_rws.md (the per-card verbatim-modifier mappings) and kate_clean_final.txt into tarot-voice-v2's meaning_db, then move the whole kate-engine-v3 folder to a delete/ subdir. Do not maintain two Kate engines.
None · $0
high
Copy the gap-fillers modifier mappings into tarot-voice-v2, then archive this folder to delete/ so it stops competing with the live engine.
$0
8
chiron-lester-2min-testshipped live
FIX
78
• This is explicitly an internal QA gate dressed as a deliverable, not a product. index.html calls it a 'test', '2-minute evaluation cut', flags 'One beat flagged', and says the full ~20-min renders for all four videos are 'held until this test is approved'. Nobody pays for a pipeline self-check. • Source is scraped from another creator's YouTube reading (6hTEvbKZUC8, 'THEM ATTACKING UR RELATIONSHIP BACKFIRED...', auto-captions cleaned). The script is a near-verbatim re-voicing of someone else's reading, which is a copyright/originality problem the moment it is monetized on YouTube or sold. • Known unfixed defect shown on the page: closing 'the tower' card image fails to render (lowercase missed by card detection), so a beat shows a caption with no card. Minor, but it ships broken.
Stop shipping 2-min QA cuts as standalone pages. Fold this into the chiron-matilda-rws-video / chiron-teaching-gen pipeline as a hidden gate, then render the four queued full readings (BvU_3hFOcvo, FaBYZiD7GwU, FaBYZiD7GwU, y4fV_c3kEUE) into ORIGINAL-script readings (rewrite from the cards, not the scraped transcript) and publish those as the actual YouTube channel content.
YouTube · $0 to view (ad-supported); funnels to $5-8/mo Substack + $5-20 Etsy packs
low
Render ONE of the four queued readings end-to-end with an original (non-scraped) script and the 'the tower' card-detection fix applied, then upload it as the first video on a fresh Chiron Intuition YouTube channel with an Etsy/Substack link in the description.
• Fatal content problem: the narration script IS the verbatim transcript of someone else's video. transcript_raw.txt opens 'Hello, this is Chiron Intuition and I'm here to do a reading for the collective' — the source creator names herself. The 5min script is just that transcript with the intro stripped, not original teaching. • It is unedited stream-of-consciousness speech read aloud: grep finds 4x 'damn', 7x 'cuz', 3x "ain't", and the script still contains [ __ ] profanity-censor markers. Lines like 'touch that [ __ ] and see what happen' get narrated by the Lester clone. This is not a finished, brand-safe product. • Copyright/identity collision: a clone voice reading another reader's exact words over RWS cards is republishing a third party's intellectual work. That is unmonetizable on YouTube (Content ID / reused-content strikes) and not legally Brandon's to sell.
Keep the render engine, throw away this episode's content. Re-point the pipeline at an ORIGINAL Brandon-written reading script (combos-first, real card meanings from the tarot-canon corpus, profanity stripped) and lock the voice to one consistent clone. The chalkboard renderer is the reusable asset; the scraped transcript is disposable. Do not ship this specific MP4.
YouTube · $0 for this file; channel ad+affiliate funnel later
low
Replace the transcript with one original Brandon-authored reading (clean meanings, no [ __ ], no 'cuz/damn'), re-render via the same engine, then test-upload ONE episode to gauge retention before building a catalog.
$0
10
avoidant-video-pipelineproduced local
FIX
68
• This is a project PLAN dressed as a deliverable, not a product. The page itself sells nothing and the revenue asset it describes (live Shorts channels) does not exist yet: it is Phase 1 of 3, dispatch/upload is not wired, the secular lane is gated, and zero videos have posted. Source is a 37KB Brandon-Vault planning .md. • Two test renders is the entire output after a full build pass. A buyer cannot purchase a pipeline, and the niche-validation case (does this actually grow channels?) is completely unproven. • Heavy overlap with the broader chiron/tarot video stack (chiron-matilda-rws-video, tarot-chapter-voice-video, book-to-video, avoidant-video-pipeline skill). The occult lane is a thin tarot-card-per-video veneer over a generic attachment-advice script, easy to confuse with sibling engines.
Stop treating the page as the deliverable and execute Phase 3 on the OCCULT lane only: fix the one v3 trim that clears both occult fails, wire the YouTube dispatch plist for brandonpage, and post 10 tarot-energy Shorts to validate the funnel into the Tarot & Voice Readings Etsy section. Park the Christian and secular lanes until the occult lane proves a single click-through.
Render the occult v3 (trim ~3s to clear the length + dead-air fails), then manually upload 5 occult Shorts to brandonpage with the Etsy tarot-section link in the description and watch for the first click-through before automating.
• This is internal dev tooling, not a product. It has zero buyer-facing value: no reading, no teaching, no card meanings, nothing a customer would pay for. It is a manifest ABOUT products, not a product. • It is already stale and partly a record of waste: ElevenLabs credits are fully depleted (0 left), the headline batch of 8 longform readings FAILED to render, and the most recent work fell back to free XTTS. The page mostly documents a dead-end spend night. • It duplicates information that belongs in the Vault/MEMORY, not on a public Vercel domain. Publishing internal file sizes, API receipts, and skill paths to drwu-htmls.vercel.app is link-clutter on the money domain and competes for attention with the actual sellable siblings (chiron-matilda-cards, chiron-tarot).
Demote it: move the manifest content into the Brandon-Vault Chiron Map of Content as an internal note and remove the public Vercel deployment so it stops cluttering the money domain. Fold its one durable lesson (gate paid TTS behind .matilda_approved, log a matilda_run.json receipt per run) into the chiron-matilda-rws-video SKILL.md, then let the siblings (chiron-matilda-cards / chiron-tarot) carry the customer story.
None · $0
high
Do not list this. Instead open the sibling it points to, chiron-matilda-cards/, decide if that RWS chalkboard reading video is YouTube-ready, and route THAT into the YouTube -> Etsy/Substack funnel. This page just gets archived to the Vault.
• This is internal dev tooling masquerading as a deliverable — a Phase 1 STATUS page with a deliverables table, 'user actions required,' and literally 'Deploy this HTML' as numbered action item #6. No buyer would ever see or pay for this. • It belongs to the video-PIPELINE lane, not the tarot product lane; only one of three lanes (Occult/Tarot, brandonpage 2nd YT) is even occult. The tarot value is one-third of an unfinished tooling page. • The actual product (videos) is blocked on every axis: format spec 'AWAITING APPROVAL,' secular lane gated on a spend token, occult render 67.9s vs a 65s gate (over), backgrounds 'PENDING user download,' and the 16-video audit is a 'USER TASK.' Nothing is shippable downstream of this page.
Kill the public page — demote it to a Brandon-Vault session note (it is a build log, not a Vercel deliverable) and fold the underlying skill's tarot lane into the existing chiron-matilda-rws-video / chalkboard reading-video stack instead of standing up a 3rd parallel video engine. The MP4 renderer is duplicative of the tarot video pipeline already in the lane.
None · $0
low
Stop treating the status HTML as output. Unblock ONE lane end-to-end: trim the occult narration ~3s to pass the gate, render a finished 60s tarot short, and post it to the brandonpage 2nd YouTube channel as the first published unit — that is the first thing that can ever earn a dollar.
• The core promise is false. This indexes card-name PROXIMITY (the same 400-char co-occurrence method memory flags as a relabel, not a capability — see project_kate_combo_coverage_is_co_occurrence.md, CEO honesty score 34/100), not combination MEANING. It never reads the text between two cards. • Even the highest 'Teaching score: 9' passages are raw pile-reading windows, not pair meanings: e.g. 'why is the four of Swords here / the world yeah there's some chapter that they're ending / permit'. The score is a keyword-density heuristic, not a meaning filter. • The passages are unedited auto-caption text: lowercase mid-word starts ('k this person is'), line-break garbage every 4-5 words, filler ('um', 'okay okay'), and ASR noise ('permit', 'page of volunteer'). Unreadable as a product; embarrassing if a buyer pays for it.
Demote it from "product" to its true role: an internal retrieval layer. Keep the data.js + lookup as the candidate-finder, then run a DeepSeek pass that READS each window and writes one clean 2-3 sentence pair meaning per card pair (label co-mention vs real combo teaching, drop the noise). The distilled, human-readable pair-meaning table is the only thing with a buyer — this raw index is the upstream tool that builds it, not the thing you sell.
None · $0
high
Stop treating it as a deliverable. Point it at the meaning_db distillation job: DeepSeek-summarize each pair's windows into a clean meaning, dedupe against IP-safe rewording, and ship THAT as a Gumroad PDF ($15-25) or Etsy combo-cheatsheet pack later.
• This is internal dev/QA tooling dressed as a deliverable. The page IS a bug-fix changelog -- coverage tables, judge variance, 'to clear the 30% bar' next-steps. No buyer pays to read 'we fixed an attribution bug and coverage rose to 24.4%.' • The headline metric is misleading per the user's own memory (project_kate_combo_coverage_is_co_occurrence.md / feedback_kate_combo_coverage): the 24.4% is 400-char co-occurrence proximity, not actual combination-TEACHING capability. So the single number the whole page is built around overstates the real product. • Thin corpus, self-admitted: 10 Rank readings + Verbatim = 13 docs. One sample reading covering only 4 pairs is not a catalog. There is no pack, no deck of combos, no printable -- nothing a customer could buy.
Stop treating this as a standalone page. Fold the eyephim recount/heatmap into a single internal tarot-voice-v2 'coverage dashboard' alongside Kate and Chiron (one engine, three tabs), and redirect the buyer-facing energy into actually generating eyephim-voice combo readings on demand via the RAG path the page itself recommends -- that is the only output anyone would pay for.
None · $0
low
Decide whether eyephim becomes a real product line; if yes, wire the eyephim corpus into the RAG generation path and ship a Gumroad pack of 30-50 eyephim-voice card-combo readings ($12-18) -- this page only proves the voice works, it sells nothing.
• This is internal dev QA tooling dressed as a deliverable, not a product. No buyer pays for a fidelity audit of someone else's tarot voice. It scores the asset; it is not the asset. • The headline is a self-indictment: 39% card-meaning fidelity and 3/16 combo claims sourced means the lane it measures is still mostly unverified. The page documents a half-finished engine more than a finished one. • The three NBA buttons are dead onclick=alert() stubs that just print a shell command, not real actions. The page cannot regenerate or rescore anything on its own; it is a static snapshot of one 8-card sample.
Stop treating this as a deliverable. Fold meaning_fidelity.py into the tarot-voice-v2 generate step as a silent pre-ship gate (reading must clear an X% card-context bar or it regenerates from meaning_db windows), and demote this HTML to an internal build report under the tarot-voice-v2/ folder. The public artifact should be the de-identified READING product, never the audit of the reader.
None · $0
high
Pull the fidelity score into the generator as a hard gate, then take the gated, de-identified reading template (NOT Kate's name or transcript windows) to Gumroad as a separate listing. This page itself earns nothing and should not be listed.
• Zero buyer in the tarot content lane. This is an internal forensic-QA artifact about a metric dispute, not a deck, reading, guide, or teaching product. No tarot customer will ever search for or pay for a fraud audit of an internal coverage number. • Audience is exactly one person (Brandon) plus future Claude sessions. It is meta-tooling -- it evaluates a prior session (2c180964), references sfg/MASSIVE TAROT internals, and is unintelligible/uninteresting to anyone outside this pipeline. • It is a one-off, hard-coded to the Kate corpus and one claim. Nothing here is templatized or reusable as-is for another reader or claim without a rewrite, despite containing a reusable prompt buried at the bottom.
Do not productize it. Extract the one durable asset -- the 5-step reusable audit prompt at lines 187-189 -- into a shared skill (e.g. /session-fraud-check or a 'claim-audit' template) so the methodology is invocable, then leave this page as the frozen reference instance. No further investment in this specific page.
None · $0
high
None for revenue. The only productive move is to lift the reusable audit prompt into a shared skill; the page itself stays as an internal reference and is never listed anywhere.
• This is internal dev tooling dressed as a deliverable, not a product. There is no reading, deck, spread, or guide a buyer could ever purchase here, only a QA scorecard for a pipeline. • The news it ships is failure: BOTH tested files (567w pre-fix, 792w fixed) score below the 0.65 Kate gate (0.580 / 0.600). It documents that the engine it is grading is not good enough to sell yet. • It is one of a sprawling family of near-identical internal eval pages (kate-combo-fraud-eval, kate-meaning-fidelity, chiron-intuition-ceo-eval, tarot-combo-index). The lane is drowning in self-evaluation dashboards instead of finished readings.
Kill it as a standalone Vercel page and fold its one real output (the gate verdict + variance table) into a single rolling "Tarot Voice Engine - Gate Log" dashboard that reads live from ~/Desktop/tarot-voice-v2/_gate_results.jsonl, so every n-run gate appends a row instead of spawning a new orphan URL per task.
None · $0
high
Do not list it. Instead use its own finding: the Kate v3 pipeline fails the 0.65 gate, so fix RAG grounding/temperature to push the median past 0.70, THEN the engine can produce sellable readings worth packaging.
• This is internal dev tooling masquerading as a shipped deliverable. It has zero buyer value as a page: no teaching, no embedded video, no narrative, just a list of build artifacts. • It is broken for any public visitor. All 21 references are file:/// local paths (hero_frame.jpg images and the MP4 links all point to /Users/drwu/Projects/...), so on the live Vercel URL every thumbnail is a dead image and every 'open mp4' link 404s. • It advertises its own failure: the summary line reads '7 rendered videos · 0 fully passing hard gates', every card shows 'no self-check', and all timing stats render as broken placeholders ('avg ?s · total ?s'). It reports a non-passing pipeline state.
Kill this page as a public deliverable and demote it to a local-only build monitor; the renderer it tracks should instead feed finished chalkboard videos straight to a YouTube channel, with the dashboard's job replaced by the existing session-fraud-check / hard-gate self-check output (no public hosting).
None · $0 (the dashboard); renderer output would price as YouTube + Etsy/$5-15 packs downstream
high
Stop deploying this page; point the chalkboard renderer at a real reading and publish ONE finished 2-min video to YouTube as the first monetizable asset, using the existing 10 hard gates as the ship gate.
• This is NOT a product and it is off-lane. It is Phish jam-review narration wiring, not Chiron/Tarot/Intuition spiritual content. There is no buyer for a shell script. • Zero output exists. There is no renders/ directory, no .wav, no .mp4 (verified with find) -- render_fish_thoughts_lester.sh has never been run. It is a TODO list dressed as a bundle. • The README itself admits the engine is broken: the most recent Lester render (avoidant-video-pipeline/queue/review/2026-06-07-occult-silent-treatment.mp4) failed QA on a 24.05s audio dropout and 67.93s length. There is also an unresolved 22s-vs-30s speaker-ref conflict in config.json and no Phish background footage staged. Three open blockers, all unfixed.
Fold the entire chain into phishthoughts-writer as a "--narrate" output step (call xtts_render.py with the long Lester ref, then hand off to avoidant-video-pipeline), and move this folder to _outputs/fish-thoughts-lester/delete/. There is no reason for a 3-file glue bundle to live as a separate deliverable.
None · $0
high
Do not list anything. Fold the wiring into phishthoughts-writer as a narrate step, fix the Lester QA dropout via xtts-glitch-qa, and only then consider whether a Phish-review YouTube channel (ad + affiliate) is worth standing up separately.
$0
20
light-assefa-readersshipped live
FIX
58
• This is internal triage tooling, not a product. The whole bottom half is a 'review each video / approve and continue' gate with cite/excerpt/skip/archive buttons. No buyer ever sees or wants that, there is no price, no download, no funnel • It is a single-creator, last-5-videos snapshot. It goes stale the instant @light_assefa uploads video #6, and it only covers ONE reader, so it is not a library or a reference anyone would pay for • Content is hard religious-occult material (demonic attachment, 400-year lineage strongholds, 'this attack would have killed you,' cancer-shaped illness) that is brand-risky and effectively unsellable as a packaged product
Stop treating this as a shipped page. Fold it into a single reusable reader-intake template (1 generator script that takes any reader's channel -> transcripts -> the same 3-block card format) and route its OUTPUT into the brandonpag3 Substack / chalkboard-video pipeline. The page itself becomes a private staging view, not a public URL. Demote from public/ to an internal _staging path.
None · $0
low
Pick the 2-3 strongest 'Signature lines' (the bridge/return-to-sender material) and draft ONE paid brandonpag3 Substack post from them, citing the reader. Measure if that single derived post earns before investing any more in the indexer itself.
$0 direct; ~$15-40 indirect if one derived Substack post is written and converts a few brandonpag3 subs
3-Card Reading Video (pre-made, evergreen popular spreads) — digital download
Etsy
$12
S
$150-$450
Card of the Day shorts series (free funnel) with reading-video link in bio
YouTube + TikTok
$0 (ad/funnel)
M
$100-$400 (channel + funnel conversions)
Avoidant / Attachment-Style Reading Video — same engine, attachment-theory niche
Etsy + Fiverr
$45
S
$120-$400
Reader-Voice Combo Lexicon for creators (raw deduped jsonl + license) — for other tarot YouTubers/app devs
Gumroad
$49
S
$100-$300
Chiron Tarot mini-ebook (the chalkboard teachings written up, sourced from the combo corpus)
Gumroad + KDP
$9
L
$80-$300
Streamline — merge the overlap
tarot-voice, kate-engine-v3, eyephim-combos, kate-combo-fraud-eval, kate-meaning-fidelity, tarot-v3-gate-results (six engines + eval pages)→tarot-voice-v2 as the single canonical 'reader-voice / combo-meaning' engine, with the verbatim meaning_db as its core and ONE internal scoreboard absorbing all the gate/fidelity/fraud dashboardsKills 6 redundant trees and 4 public Vercel 'work products' that are really QA pages; one source of truth, one coverage matrix, one place to add a reader. Stops the fork-evaluate-fork loop that produced 7 near-identical stacks.
chiron-lester-2min-test, chiron-matilda-cards, chiron-chalkboard-matilda, chiron-tarot, tarot-padbed, chalkboard-pipeline, avoidant-pipeline-phase1 (seven render-stage folders)→chiron-matilda-rws-video as the single canonical reading-video pipeline (padbed becomes a step, chalkboard-pipeline becomes its README, chiron-tarot/cards become published samples, avoidant-pipeline-phase1 collapses into avoidant-video-pipeline)One render path instead of seven half-finished ones, ~250MB of intermediate frames archived, and every future video ships from one tested engine instead of a fresh 'test' folder. Ends the re-test-the-renderer treadmill.
Kill list
fish-thoughts-lester (off-lane Phish/music artifact — remove from the lane)
light-assefa-readers (wrong family + 0.4% dead coverage, 12 pairs)
eyephim-combos (eval heatmap, reader is below 30% threshold)
kate-combo-fraud-eval (QA instrumentation, not a product)
kate-meaning-fidelity (QA instrumentation, duplicate of fraud-eval)
tarot-v3-gate-results (CI gate report dressed as a deliverable)
chiron-chalkboard-matilda frame dump (225MB of intermediate frames — archive to delete/, keep only the final video)
chiron-lester-2min-test as a listable item (it is a render test; demote to a sample clip)