habitatOne agent's field, one sitting, 2026¶
Generation a5c5d1007022 · 2026-09-05 · 2,722 acts · August 2026 to August 2026
One agent, one sitting, 2,722 acts. Not one word of them is shown, and you can still see where its work changed.
Condition at close¶
Read against two nulls, 1000 draws of each, at a threshold of 0.004167 after correcting for 12 readings tested together. Reshuffling the acts tests whether a reading depends on their order; a matched field of the same overall shape tests whether it depends on more than that shape. A reading that neither test can distinguish from chance is withdrawn, and withdrawn is the common outcome.
| reading | state | reshuffled | same shape | verdict |
|---|---|---|---|---|
| content_len_log | STATE | 0.377 | 0.248 | withdrawn |
| n_tool_use | STATE | 0.385 | 1.0 | withdrawn |
| n_text_block | STATE | 0.333 | 0.208 | withdrawn |
| out_tok_log | STATE | 0.219 | 0.278 | withdrawn |
| in_tok_log | STATE | 0.636 | 0.0 | the acts, not their order |
| cache_read_log | STATE | 0.101 | 0.066 | withdrawn |
| dt_log | STATE | 0.184 | 0.102 | withdrawn |
| is_shell | STATE | 0.192 | 0.286 | withdrawn |
| is_mutate | STATE | 0.061 | 0.054 | withdrawn |
| is_browser | STATE | 0.028 | 0.065 | withdrawn |
| viscosity | ratio 0.0005 | 0.219 | 0.011 | withdrawn |
| capacitance | ratio 257.0417 | 0.111 | 0.0 | the acts, not their order |
Nothing survived both tests.
2 cleared the shape test only (in_tok_log, capacitance), which is the weaker of the two and is not evidence that anything happened. Reshuffling the acts left those readings intact, so their position in the record is not established. And a reading clears the shape test whenever the record is not a smooth statistical object, which any uneven record of real work is: a running total, for instance, clears it by being a running total. - Known going in: the eventuality rule fires on impulse series (1,199 onsets, 2026-09-02 review) - Known going in: the CONVERGENCE signature fired on 199 of 200 stationary-noise streams (2026-08-25) - Known going in: regime factors are hardcoded by column position; on d=10 that is column order, not meaning - Known going in: the covariance read here is centered on the record's own mean; one path in the serving code accumulates an uncentered sum instead, and the two are not the same object - Known going in: compute_capacitance's auto-detect assumes a stationary-then-spike series and a g-shift history decays, so its flattest window lands at the tail; the candidate act is declared instead, and the auto-detect's own answer is reported beside it
Events of note¶
- A change, 21 August 2026 (act 1779): ACHIEVEMENT onset by classify_eventualities on the eigenvalue trajectory. none on this face; the rule fires on impulse series (9/2 review); reported, not weighed.
Bench¶
- a state in 1 act of 4 (retained 308, absorbed 386, noise 1929)
- 13 changes by the code's rule, 12 of them in the founding head
Limits¶
- Always trailing edge. One agent's own acts, one record, in the order they entered, read after entry and never ahead of it. Not a forecast. Nothing here steers the agent.
- No act's words are shown. Each column is the shape an act made in the field's own harmonics and the state it holds against noise. The grammar without the words.
- Only what the agent generated is in this field. What it observed stays apart, by token, not by trust; the receipts for that boundary sit beside this page.
- A window, chosen by the person holding the record: session 62, 2722 acts of 28735; the field is rebuilt from the window's own acts. Ten structural columns, not the seventeen named dimensions. What the agent asserted in language is untested here.
- The median act behaves like noise. The share with a state is printed by the export, not assumed. States, not things: one sitting can hold both retained and absorbed acts, and the first day is a transient, read as one.
- The medium computes; it interprets nothing. What the agent was doing on any day, and whether it did well, is the principal's to say and no one else's. The numbers reproduce from the bytes beside this page.
Source¶
The record's own bytes, 2,722 acts; playback-agent.json sha256 a5c5d10070227cf2…; result files: RESULT_kernel_parity.txt, RESULT_quant_parse_v2.txt, DECLARED_quant-schema_v2.md. Every number recomputes in your tab. The medium interprets nothing. No word of the record appears in this report or in -condition.json.