# The self-learning loop

Source: https://tactiq.pro/docs/learning-loop · Trading ideas · updated 2026-10-31

Every day Tactiq reviews closed trades, writes lessons for the analyst and scores each lesson. You stay in control: any lesson can be paused, edited or removed.

Tactiq learns the way a trading desk does. Every day it reviews what happened, writes down what it should do differently, and reads that note before the next trade. The people stay in control.

1. **Rules propose.** A strategy finds an idea.
2. **The analyst reviews** it, with the playbook's relevant lessons in the dossier (at most 6, most specific first).
3. **Followed to the end,** taken or passed, on 1-minute bars.
4. **Post-mortem.** Tactiq *measures* the trade: best and worst excursion in R, how far TP1 was missed by, time held, and where price went in the 4 hours after the exit. The writer then explains what decided it, or says it was noise.
5. **The daily review, 00:15 UTC.** The writer plays the desk manager and reads the day's closed trades, even one, against the playbook. It may add a lesson, refine one, or pause one. A second call plays the risk manager and checks each proposal.
6. **Lessons are live at once.** The owner of Tactiq gets a daily Telegram message that lists the changes, each with Pause and Remove buttons.

### What a lesson can and cannot do

- A lesson changes **what the analyst is told**. It never changes a strategy's rule, a stop, a target or a size.
- A lesson carries no prices. A price in its text is dropped.
- A loss is not a lesson. "No change" is the usual answer, and the review is built to say it.

### Strength of a lesson

| Strength | Needs | Wording | For the analyst |
| --- | --- | --- | --- |
| Observation | Anything, even one trade | "check", "watch for" | Where to look. Never a reason to pass on its own. |
| Tendency | 3 or more trades, outside the range that chance alone gives (2.0 standard errors) | "weigh", "prefer" | May tip a borderline call. |
| Rule | 8 or more trades, far outside the range (3.0 standard errors) | Firm | May decide a borderline call. |

Tactiq lowers a strength that the numbers do not support. These caps are about wording. A lesson is live either way.

### The scorecard

Every lesson shows how the ideas it matches went before and since it went live, on how many days, and how often the analyst was shown it. It is information, not a gate. The writer and the analyst read it too, and are told that small counts are noise.

### Your controls

- **Telegram:** Pause (or Resume) and Remove under each change in the daily message.
- **Learning page (owner only):** the playbook with scorecards. The owner can pause, resume, edit (saved as a new version, live at once) or remove. A removed lesson is never re-created by the machine. A paused one is not touched.
- **History:** every version is kept and every change is logged.

### What the loop does not do yet

- It does not prove that lessons make money. That can only be judged forward, from their scorecards.
- A lesson drawn from one trade can be noise, and the analyst will weigh it. The strength labels and the scorecard exist for this.
- Stops, targets and sizing never adapt. A rule change is a new version that you approve.
- The review was tested on invented desks where the truth was known. There, on pure noise, 7 of 8 runs wrote only observations, and a planted pattern was found in 5 of 8 runs. It is a guide to behaviour, not evidence of profit.

### Learning about the market itself

The Learning page also holds Tactiq's market knowledge: the readings it makes, each scored against random times. A reading that comes true 66 % of the time means little if a random moment comes true 66 % of the time. Tactiq proposes a change only with at least 30 resolved readings and 30 controls, a lift over random times that clears a strict threshold for the number of tests, and a stable result in older and newer halves. Major changes wait for the owner, and every version can be rolled back.