docs/plans A3+A6+A1: surfacing economy — all 30 sensates, top-2 priors, top-X convictions

A3: PS+ surfaces all 30 raw sensates but only the 2 most salient priors
(L3 ranked-top-2); evaluate_reality arguments carries top-2 only.
A6: top-X convictions rationale — the conviction list can be massive.
A1 step 2: all 30 sensates + 2 most salient priors; note the asymmetry
(fixed-30 surfaces whole; unbounded prior/conviction lists are ranked).

https://claude.ai/code/session_015hmgREHNsxYCuim33yUF2c

docs/plans/A1-drivebox-hub.md

@@ -62,10 +62,11 @@ Sequence per cycle, consistent with A2 (per-tool cost), A3/A4 (raw sensates, age
 
 1. **Assemble (hub, concurrent).** All drivers + tarot + SOUL write into the slot at once (L1).
    Nothing is subordinated; nothing decides here.
-2. **Surface sensates raw — *before* tool listing** (A3 §4, A4). The 30 channels go to the agent
-   unsummed (incl. zeros — silence is data), with their arguments + any prior-derived argument.
-   They **describe & convince**; illogical by design (A3-L1), so there's nothing to refute — they
-   steer beneath cognition. No load scalar, no friction.
+2. **Surface sensates raw — *before* tool listing** (A3 §4, A4). **All 30** channels go to the agent
+   unsummed (incl. zeros — silence is data), plus the **2 most salient priors** (ranked, not the whole
+   prior set — A3-L3). They **describe & convince**; illogical by design (A3-L1), so there's nothing to
+   refute — they steer beneath cognition. No load scalar, no friction. (Asymmetry vs step 3: sensates are
+   a fixed 30 so all surface; priors & convictions are unbounded lists, so they're ranked & truncated.)
 3. **Then list tools.** **Valid (unlocked) tools show their per-tool cost** (Anja) so the agent can
    weigh price before choosing. A **locked** tool is **not hidden** but its *name is replaced in-band*
    by the lock token — see L4 — because the agent is a text model and **cannot perceive colour/greying**

docs/plans/A3-psplus.md

@@ -12,15 +12,16 @@ Structure WORKING (`driver_ps_plus.R`, 63 L); the old aggregate-sum + friction a
 R · `src/endocrine/driver_ps_plus.R` (sources `endocrine_array.R` + `priors.R`; + `test_ps_plus.R`).
 
 ## 4. Does / does-not
-- **Does:** read the active endomotiv vectors (A4) — **no friction**; read priors (A5); **pass the raw
-  sensates + their arguments to the agent before tool listing** (they describe & convince — the agent
-  then decides, with the per-tool costs from A2).
+- **Does:** read the active endomotiv vectors (A4) — **no friction**; read priors (A5); **pass all 30 raw
+  sensates + the 2 most salient priors to the agent before tool listing** (Anja — sensates in full, priors
+  ranked to the top 2). They describe & convince — the agent then decides, with the per-tool costs from A2.
 - **Does-not:** sum, reason, gate, or decide. It asserts "X is happening"; it never approves or routes.
 
 ## 5. Interface contract
 - `init_ps_plus_state() -> { endocrines:A4, priors:A5 }`.
 - `evaluate_reality(state) -> { sensates:[raw per-channel, 30], arguments:[str], is_logical:FALSE }`
-  — **raw, not summed; no friction term; no load scalar.** Surfaced to the agent **before tool listing**.
+  — **all 30 sensates raw, not summed; no friction term; no load scalar.** `arguments` carries the
+  **2 most salient priors only** (ranked, not the whole prior set). Surfaced **before tool listing**.
 
 ## 6. Dependencies & stubs
 A4 endomotiv array — *stub:* `init_endocrine_state()` with canned channel magnitudes.
@@ -34,7 +35,8 @@ themes, **tied to descriptors of taste / smell / sound + location** (not hand-ad
   (sensation, not proposition — no argument to refute, so they steer beneath cognition).
 - **L2 (C5):** sensates are sent **raw, never summed** — qualia (the field) and any economy (priced
   elsewhere, A2) ride separate rails; PS+ aggregates nothing.
-- **L3 (C4):** a relevant prior injects a large, specific argument (priors derived per §6).
+- **L3 (C4):** the **2 most salient** priors inject large, specific arguments (priors derived per §6);
+  only the top 2 surface, ranked by salience — the rest stay silent.
 
 ## 8. Build steps
 1. **[test approach was wrong — Anja]** rewrite `test_ps_plus.R` for the corrected model: raw/unsummed

docs/plans/A6-ethint.md

@@ -15,7 +15,8 @@ R · `src/endocrine/driver_ethical_integrity.R` (+ `test_ethical_integrity.R`).
 ## 4. Does / does-not
 - **Does:** evaluate a trajectory's E-cost (alignment discount × antithesis penalty × compromise);
   harden convictions upheld under load; treat middle-ground as its own trajectory; **surface the top X
-  convictions** (by strength) into the agent's slot — *not* the whole array (Anja; A1 step 3 / input-slot).
+  convictions** (by strength) into the agent's slot — *not* the whole array (Anja: **the conviction list
+  can be massive**, so it must be ranked & truncated; A1 step 3 / input-slot).
 - **Does-not:** assert moral truth, or persist itself as a stratum (it derives from memory; see A7/E2).
 
 ## 5. Interface contract