Multi-Table Tournament Bots: Methodological Notes on a Mid-Stakes Cohort

An operational record covering the 2022–2026 window in online MTT play.

Document v. 2026.06 · Compiled for internal distribution · Buy-in scope: $5–$100.

This document outlines observations on bot behaviour in multi-table tournaments (MTTs) compiled over the 2022–2026 operational window. Results are drawn from a non-random sample of mid-stakes events, and conclusions should be read in that scope. Three areas receive particular attention: the structural difference between MTT and cash policies, the failure modes of ICM-naive agents, and the empirical position of late registration as a meta-game decision.

1. Scope and methodology

The cohort consists of approximately 41,000 entries logged across 2,300 events on three online networks between January 2022 and April 2026. Events are filtered to scheduled MTTs with payout ladders of nine or more positions; satellites, knockout variants, and sit-and-go formats are excluded from the headline figures and treated separately in appendix material not reproduced here.

Agents in the sample share a common core but differ in tournament-specific overlays. Cash-derived policies were used in early cohorts and progressively replaced after 2023.Q3 once equity-loss patterns in late stages had been quantified [1]. Where comparisons are drawn between cash and MTT performance, only paired periods are used.

2. What differs from cash

Three structural features of MTTs are emphasised, since each one degrades a cash-derived policy in a distinct way.

1. Variable stack depth. Effective stacks span roughly 8 to 200 big blinds within a single tournament, often within a single orbit at the same table. A policy tuned for 100bb cash play is, by construction, suboptimal across most of this range.
2. Non-linear chip value. The marginal value of a chip declines as stack increases, which the chip-EV objective ignores. This is the domain of ICM (see ICM considerations).
3. Registration as a degree of freedom. Unlike cash, MTT participation is a temporal choice: enter at the start, late-register, or skip the event entirely. This is treated separately in the late-reg strategy note.

3. Findings

The bubble row reproduces the most consistent finding in the literature and in this cohort: an ICM-naive policy that approximates GTO under chip-EV produces measurable equity loss in the late stages of MTTs. The magnitude is sensitive to payout structure and remaining-player count, with the steepest gradient near the money bubble.

4. Illustrative bracket

The diagram is illustrative only. Real MTT progression is asynchronous; "rounds" here stand in for blind-level transitions rather than discrete eliminations.

5. Caveats

Sample selection is non-random across both networks and stakes. Mid-stakes pools differ in skill density from micro and high-stakes pools, and the findings should not be extrapolated outside the $5–$100 band without re-sampling. Variance at the final-table stage remains substantial even at the cohort size described, and the bb/100 figures in Table 1 carry intervals wide enough to overlap across the lower three rows.