A vocabulary is only worth having if it is measured
ZEME labels every covered stock with one of nine Campaign Phases™, a description of where an operator appears to be in a campaign, not a prediction. The phases are only worth learning because each one carries a measured next-day upper-lock rate beside it. A taxonomy without numbers is decoration.
| # | Phase | Plain meaning | Next-day lock rate (base 3.81%) |
|---|---|---|---|
| 1 | WATCHING | on the radar, nothing committed | 3.89% (1.02x) |
| 2 | WATCH_ONLY | tracked, below the action floor | not measured |
| 3 | EARLY LOAD | first accumulation appearing | 2.54% (0.67x) |
| 4 | LOADING | accumulation underway | 0.87% (0.23x) |
| 5 | FULLY LOADED | position held, not yet moving | 1.40% (0.37x) |
| 6 | LAUNCHING | the move is starting | 6.87% (1.81x, the best) |
| 7 | RUNNING | the move is in progress | 5.19% (1.36x) |
| 8 | DISTRIBUTING | selling into strength | 4.41% (1.16x) |
| 9 | EXITED | the operator has left | 2.82% (0.74x) |
[LEDGER: 20 months, liquid universe, project_zeme_lexicon]
The counter-intuitive part
Look at LOADING: 0.87%, or 0.23x base. A stock being quietly accumulated is less likely to lock tomorrow than a random stock. That is not a flaw in the label, it is the label working. Accumulation is, by definition, the phase where nothing visible happens. The operator's entire objective during LOADING is that you do not notice.
Which produces the actual trade-off nobody says out loud:
- LAUNCHING has the highest hit rate (1.81x) and the worst price, the move has already started.
- LOADING has the worst hit rate (0.23x) and the best price, and an unknown wait.
There is no phase that is both cheap and imminent. Anyone selling you one is selling you a story.
UNKNOWN is not a phase
The data contains UNKNOWN. That is a gap in measurement, not a state of the world, and it must never be shown to a customer as though it were a phase. This is the same discipline as the null rule: a missing value is "not measured", never zero and never a label.
Evidence · History · Confidence · Counterargument
Evidence. The table, computed over 20 months on the liquid universe. [LEDGER: project_zeme_lexicon]
History, 20 months is roughly 400 sessions. It spans one broad market character, not several.
Confidence, medium. The ordering (LAUNCHING best, LOADING worst) is a large and consistent effect. The exact percentages are one measurement window, and one band of the related confidence model has already been observed to drift 8.4 points within 60 days [LEDGER: project_brain_tuning]. Treat the rank order as the finding and the decimals as provisional.
Counterargument, phase labels are re-derived frequently and an internal audit found the phase behaves partly as a relabelled confidence scalar that can flip every couple of days [LEDGER: project_operator_engine_audit]. So a phase is a description of the current reading, not a stable property of the company. If you build a rule that requires the phase to hold still, it will break.