Farm Proceedings, Schedules & Watchdogs

Nitro end-to-end proceed document, version one, 4 Oct 2026

Proceed signal. When the user says "proceed" or "end to end", that one word authorises every stage below, in order, through to done. Don't re-ask between stages. Only "hold" or "stop" pauses the lane. A result that lands starts the next stage automatically unless the user has said hold.

Stages

1. Scrape. The trader's top 20 long videos and top 20 Shorts by views, at full depth: captions (or whisper when there are none), frames, contact sheet, card, live views, upload date. Done when all 40 are carded or the failures are logged in FAILED.md. Report the count, the failures and the A+ examples.
2. Extract. Build the master scope document in Berg format, following Part 2. Done when the deduplicated master list, master playbook, contradictions table and totals are complete. Report the category counts and the top gaps.
3. A+ definition. Lock the trader's A+ setup as the "<setup name> locked rules template one", with a checklist and NinjaTrader-style parameters. It feeds the Roman and ZZ entry contracts: the entry setup only, not the SMT body. Done when the template is on the trader's Dev page.
4. Code. Code the missing rules first, then the partly coded ones, in one universal codebase with the trader as a settings template. Back up before every change. New settings default to off. Each rule gets a unit test plus a smoke run on clean data. No tuning and no performance claims. Done when every codeable rule is coded. Report what's newly coded, what's blocked and why, and the before and after counts.
5. Contradictions. Each one becomes a switchable setting, flagged in the code and the document. Never pick a side silently. A mismatch with no clear answer stays flagged as unresolved.
6. Publish. Dev site only, through the Dev publishing agent: the trader page sections, the top-ten video page, the scrape badges and the rule-extraction tags. Done when the URLs are verified live. Report the links.
7. Test hand-off. The locked rules go to the backtest lane on the farm map: in-sample tune, then out-of-sample once, then holdout once. Commission comes out of everything, with the standard label and end metrics. Failures stay visible. Done when the results are reported.

Every hand-off reports: what finished, the counts, what's blocked and why, the links or file paths, and the next stage that's already started.

Never without asking: production pushes, money or sends, inventing a stop, target or other unstated rule, or presenting in-sample results as final.

Top Traders playbook building procedure, version one, 4 Oct 2026

1. Inputs. The stage 1 deep cards: timed transcript, frames, contact sheet and card for each video or Short. List videos with no rules (vlogs, psychology talks) as no-rules and skip them.
2. Extract per video. One row per rule: rule ID, section, rule text, model, and source as video ID at a timestamp. A stop, target or time the trader never says is written "not stated", never guessed. Log the trader's profit, win-rate and RR figures as claims, not facts.
3. Deduplicate. Merge the same rule from different videos into one row and add each source to it. Rule IDs never change once issued. A new rule gets the next free ID in its section. Count new rules and confirmations separately.
4. Label against the real code, not old runners:
   • Coded: it exists and works.
   • Partly coded: a module exists but isn't wired in, or is wired in wrong.
   • Missing: no code at all.
   • Judgement call: discretion that can't be made mechanical.
   • Not stated: something the trader never says.

Note whether each one is Python only or also NinjaTrader.

1. Master playbook, at the top of the document:
   • The model in one paragraph, and a session clock in ET with the user's local time.
   • The totals tables: overall, by section, and by model.
   • The master rules list, grouped by label, with every source cited.
   • Per model: the check sequence, then the if-then decision tree, the stand-aside list and the grading.
   • A table of where the per-video playbooks differ.
   • The new code needed, listed as NinjaTrader-style parameters in Roman's Essentials groups plus an entry-plugin group.

1. Contradictions table. One row per conflict, with both sides and the video and timestamp for each. Flag a mismatch with no clear answer as unresolved. Never pick a side.
2. A+ examples table. Each clip that matches the A+ template, with its video ID, timestamp and the rule IDs it shows. List partial fits separately, with what's missing. Note any disagreement with an earlier keyword card.
3. Appendix. A rule table and short playbook for each video and each Short that teaches a setup, using that video's timestamps only. Indicator and tip clips are listed only. Add a CSV with one row per rule.
4. Feeds the next stages.
   • Stage 3: the A+ examples, grading and check sequence become the locked rules template.
   • Stage 4: the missing and partly coded rows, in that order, become the coding queue, using the parameter names from step 5. Contradictions go in as switchable settings.

1. Done means: every source is processed or listed as no-rules, and every rule has a source, a timestamp and a label. The totals add up, the contradictions and A+ tables are filled in, the CSV matches the document, and timestamps are spot-checked against the cards.
2. Report at hand-off:
   • The totals by label and by model.
   • New rules versus confirmations.
   • The top gaps in priority order.
   • New contradictions.
   • The A+ examples.
   • Caveats, such as Python only or unverified claims.
   • The file paths.
   • That stages 3 and 4 have started.

Built from Nitro's 20 long videos and 20 Shorts (126 rules); the reference document is NITRO_SCOPE_DOCUMENT.md in the same folder.

Proceedings

Append new rows to tbody#proceedings-rows, newest entry last, one row per stage hand-off.