# TREDERCOPIS — CANONICAL PROJECT CONTEXT

Updated: 2026-09-12 UTC

READ THIS FIRST in a new chat. This document is the human-facing source of truth for project direction. Verify volatile runtime state from server files before acting. Never infer current LIVE state from old notes.

## 1. System map

`datasetis.info` — DATASET/data plane. Owns Parser0/1/2, Coin Passport, market/orderbook history, Wave/regime runtime, normalized data contracts, profile generation/build/library and persistent DATASET SHADOW league.

`tredersolo.online` — SOLO research/LAB. Owns historical causal replay, profile research, robustness, survival/damage experiments. Research does not promote directly to production.

`combatsolo.info` — production strategy consumer. DEMO strategy brain, SHADOW/Champion selection, operator controls and LIVE execution boundary.

`lab.combatsolo.info` — isolated Combat SHADOW/ranking LAB. No Bybit execution/orders and no production mutations.

Flow: `DATASET -> SOLO/LAB research + profile Library -> Combat SHADOW -> selector -> DEMO -> LIVE executor -> Bybit`.

## 2. Current production concept

Generator builds aligned LONG/SHORT HSCAN profile families for 1/2/3/4/6/12/24h. Generated Library profiles are immutable research/selection inputs.

DEMO is the strategy brain. LIVE should mirror durable DEMO OPEN/CLOSE events; LIVE must not independently choose/recalculate profiles. Profile IDs in LIVE are audit metadata.

DEMO profile mode should remain AUTO unless explicitly changed by operator. Anti-stuck operator rotation is a one-shot AUTO rotation, not MANUAL mode. It changes DEMO only and must not directly mutate LIVE.

Champion/profile quality is not WR alone. Relevant dimensions: traffic, PnL/expectancy, losses, WR, evidence size, drawdown/damage, causal forward behavior and robustness.

## 3. Risk / execution assumptions

Current intended common trade assumptions for new profile-generation evidence: leverage 5x, margin budget 6 USDT, hard stop -10% ROI, profit floor +5% ROI, trailing arm +10% ROI, trailing gap 2%.

LONG LIVE has previously been explicitly authorized for real execution. SHORT LIVE must remain OFF unless separately authorized. Therefore never run a normal production Engine Tick merely to inspect state: it can have execution side effects.

Before claiming current positions/orders, read current runtime and, when possible, current exchange-backed state. Local files can be stale.

## 4. Selector / anti-stuck

Production anti-stuck action `rotateDemoAuto()` selects another eligible SHADOW candidate while preserving AUTO, clears manual profile selection, journals the switch, and temporarily blocks immediate return to the profile the operator explicitly left. The block is an anti-reversal cooldown, not a global minimum profile age.

Automatic degradation logic is still a research/design area. Do not use hair-trigger conditions such as one negative interval. The desired direction is broad evidence of damage/degradation plus a demonstrably better eligible alternative.

## 5. LAB — canonical research direction

Implemented modules / existing feature map: `docs/MODULES_IMPLEMENTED.md`. Read it before proposing a new subsystem or duplicate Passport field.
Master LAB index: `docs/LAB_INDEX.md`.
History Fingerprint canonical document: `docs/LAB_HISTORY_FINGERPRINT.md`.

Latest completed LONG 24H decay experiment:
- tool: `tredersolo.online/modules/history_fingerprint_lab/tools/long_24h_predictive_decay_v1.php`
- report: `tredersolo.online/modules/history_fingerprint_lab/storage/rolling_profile_v1/long_24h_predictive_decay_v1.json`
- 22 frozen LONG HSCAN 24H profiles were causally replayed forward on 1/2/3/4/6/12/24h cumulative windows.
- pooled expectancy stayed positive through 24h; peak pooled expectancy was around 6h. This did NOT support simply deleting 24H.

Current research priority is Survival/Damage, not wall-clock TTL. Primary axis: `RESOLVED_EVENTS_AND_PNL`; time is observation only. Goal: determine the smallest defensible damage/evidence point at which a current profile is mismatched, then compare contemporaneous 12/6/4/3/2/1H alternatives and measure avoided damage versus false-switch cost.

## 6. Documentation discipline

This file answers: WHAT IS THE SYSTEM and WHAT DIRECTION ARE WE FOLLOWING.
`LAB_INDEX.md` answers: WHAT LABS/EXPERIMENTS EXIST.
Each LAB has one canonical human-facing document; machine results remain JSON/NDJSON beside the experiment.

Old install/stage notes are historical only and must not override canonical docs or current code. When a new experiment materially changes a conclusion, update the canonical docs in the same change.

## 7. Rules for ChatGPT / operator work

1. Start from this file, then read only the relevant canonical subsystem document.
2. For current runtime/LIVE questions, inspect current server state; documentation is not runtime truth.
3. Preserve DATASET / SOLO LAB / Combat DEMO / Combat LIVE separation.
4. Never promote a research result directly to DEMO/LIVE.
5. Do not change generator formulas, selector policy, execution or LIVE state merely because a LAB result looks promising; require explicit implementation decision.
6. Report what was measured separately from interpretation/hypothesis.
7. Avoid reviving superseded architecture from old README/install/stage files.