YC application workstream (Marton + Xavier)

Started 2026-10-19. Separate venture. Marton's public brand stays v8.0 (coaching).

Start here (any new agent or session)

This folder is the one home for all YC work. Point a new agent at ~/code/xmarts/yc/README.md and have it read, in order:

This README: the rule, the video slots, who does what, the plan.
../brief/xmarts-brief.md: the venture, the founders, the wedges (W1 to W9), what the system can do. It is also everything the @xmarts Slack bot knows.
angles/index.html: Marton's idea angles (A to G), ranked and judged by Claude and GPT, with pros and cons.
research/: the sources. Landscape and YC facts (yc-landscape-*), Within (within-study-*), the system (system-inventory-*), calls (left out of this shared copy), Xavier (xavier-background-*), the angle research (angles-*).
rubric/: how wedges are scored. Do not open sealed score folder before the blind reveal.
teardowns/: YC companies torn down against our system (skill yc-teardown, source in skills/yc-teardown/). One file per company in teardowns/companies/; the landscape in teardowns/landscape/.

Rules for new agents: words only (no build), never invent numbers or quotes, never touch LinkedIn, commit only your own files. After changing the brief, run ../slack/sync-brief.sh so @xmarts learns it.

The one rule

Everything ladders up to the 60-second founder video. If a piece of work does not fill a slot in the video (or the form answer behind it), it waits.

The video spine (the slots we must fill)

Who we are: two names, how we met, who builds. (~10s)
The problem: one buyer, one pain, in their words. (~15s)
What we built: the system, and the one number it moved. (~20s)
Why us, why now: the insight funded competitors miss. (~15s)

Who does what

Marton: builder. Owns the system demo and the proof numbers. Final say with Xavier.
Xavier: business side. Runs buyer conversations in the consultancy and services world.
Claude: research, synthesis, rubric upkeep, scoring as judge 3, drafts. Never self-reviews.
GPT (Codex, gpt-5.6-sol, effort high): adversarial judge 4. Plays the YC partner. Attacks each wedge and the video script.

Folders

angles/ Marton's idea angles A to G, ranked (page: angles/index.html)
research/ landscape, YC facts, system inventory, call evidence, angle research
rubric/ the scoring system (rubric-v1.yaml) and every score sheet
skills/yc-teardown/ the teardown skill (linked into ~/.claude/skills/)
teardowns/ screen of all YC companies, deep teardowns, landscape
discovery/ one file per buyer conversation, exact quotes only
pitch/ 50-char line, form answers, video script drafts, takes log

Deadline and plan (14 days)

On-time deadline: Mon 2 Nov 2026, 8pm PT = 04:00 UK Tue 3 Nov. We submit Sun 1 Nov (one day spare).

Days	What	Done when
Mon 19 - Wed 21 Oct	Back-test the rubric on 6 anchors. Longlist wedges. 4 judges score blind. Set up Paxel.	Top 3 wedges named
Thu 22 - Mon 26 Oct	Discovery: 10+ buyer conversations on the top 3. Xavier: consultancies. Marton: startup engineering leads.	Quotes logged in `discovery/`; one pilot or LOI asked for
Tue 27 - Wed 28 Oct	Re-score with evidence. Pick ONE. Test the 50-char line on 5 strangers.	One wedge, one line
Thu 29 - Fri 30 Oct	Write the form. Record the video from bullets, 3 takes. Record the demo. GPT plays the partner.	Draft application
Sat 31 Oct - Sun 1 Nov	Fix, re-record, submit.	Submitted

Form trap: the revenue question asks for a split by source. Coaching revenue must be shown apart from XMarts.

Loop (repeat until the pick)

Longlist wedges (from the inventory + calls).
Back-test the rubric on funded anchors. Fix weights if it ranks them wrong.
Four judges score blind. Reveal. Talk only where the spread is over 2.
Top 3 go to discovery. Xavier and Marton each run calls. Log quotes.
Re-score with the new evidence. Kill, keep or merge.
Test the 50-char line and a 60s voice-memo version on 5 strangers.
Pick one. Write the form. Record the video. GPT attacks it. Re-record.