v61: preserve benchmark cases, execute grounded income research on V100 and stabilize live activity

Author: Mantiee

SYNTA_RECOVERY.md

@@ -71,3 +71,47 @@ execution and measured progress must be checked after installation on Debian.
 No payment ledger is connected. Nothing here guarantees an undiscovered edge,
 fast large income, validated strategy, automatic sale or universal continual
 learning. Prepared useful work, actual outcomes and payment remain distinct.
+
+## v61: demonstrated benchmark, research and dashboard faults
+
+- Official benchmark finalization previously overwrote `cases` with a count,
+  causing `'int' object is not iterable`. Counts now use `case_count`; all case
+  rows, including zero scores, remain intact. A restart resumes the newest
+  matching original baseline prefix after checking model, generation, immutable
+  snapshot and row identities. It never chooses a run by score or skips gates.
+- The operator explicitly stopped Windows RTX after a black-screen hang. This
+  update records that preference, cancels its periodic benchmark and sends
+  resident model research to the protected accepted V100 already serving. GUI
+  observations and dashboard probes do not contact the disabled helper. Windows
+  CPU jobs retain their existing two-thread / four-GiB child configuration.
+  No Windows GPU process, board clock, power limit or task is started here.
+- Bounded primary-terms collection runs independently of the master training
+  phase every ten minutes, rotating starting sources across human research,
+  usability and explicitly authorized software-bounty programs. It archives an
+  actual source even when model analysis is unavailable. A protected V100 then
+  prepares a source-linked feasibility dossier with eligibility, labor, costs,
+  delay, blockers and a mechanism-specific test. Sources are starting points,
+  not recommendations or a closed discovery list. Human participation is not
+  automated; no account, application, scan, submission or payment occurs.
+- Known microtask/affiliate hypotheses tested with BTC prices are rejected before
+  review. Exact repeated notes without a new recorded operation are marked as
+  repeated advisory. These conservative lexical/receipt checks are not semantic
+  proofs. Old findings are annotated with a backup; original texts are retained.
+- Token deltas no longer displace tool calls/results/errors from the action
+  timeline. Historical actor cards are collapsed; queued/running job cards put
+  old attempts in a separate disclosure, never the current result. New requests
+  clear their actor's old error, evidence and token usage. An unchanged journal
+  filename is not treated as shared source evidence.
+- Updates preserve inner reader scroll and anchor the visible expanded detail
+  when new content is inserted above it. Tests execute the renderer in a Node
+  DOM fixture; actual browser acceptance remains a hardware check.
+- Existing crypto prices are refreshed before fee parsing. A fee-layout failure
+  archives the exact source and blocker instead of suppressing price observation.
+  Product boundaries accept case changes. Unknown or ambiguous fees still block
+  paper fills, and published fee assumptions remain distinct from account rights.
+
+Inspect `research/income-work/status.json`, its source-linked dossiers,
+`research/research-quality/latest.json`, public benchmark resume receipts and
+`research/paper/fee-source-status.json` for measured execution. Feasibility
+estimates are neither real income nor training labels. Actual master training
+and promotion still need independently verified outcomes and all quality gates.

SYNTA_REQUIREMENTS_AUDIT.md

@@ -1,4 +1,4 @@
-# Synta: requirement and verification audit, v60
+# Synta: requirement and verification audit, v61
 
 This audits the visible conversation. "Implemented" means code exists and the
 listed checks pass, not that it has run successfully on the operator's computers.
@@ -49,9 +49,9 @@ The financial goal is not replaced by arithmetic, dashboard maintenance or a
 small forecast task. Arithmetic is an auxiliary experiment. Failed trading tests,
 losses and rejected updates remain evidence. No result here establishes income.
 
-Source-size snapshot: 46130 physical lines in 190 tracked/new Python, PowerShell and shell files under `rlm/v100`, `tools` and `tests`. Includes comments/tests, excludes temporary files.
+Historical v60 source-size snapshot: 46130 physical lines in 190 tracked/new Python, PowerShell and shell files under `rlm/v100`, `tools` and `tests`. Includes comments/tests, excludes temporary files.
 
-Verification result: full test suite 1028 passed, 65 skipped, no deselections;
+Verification result (v61): full test suite 1086 passed, 65 skipped, no deselections;
 ruff and upgrade shell checks pass. Windows startup policy has static contract
 checks, not a Windows runtime execution. Real CPU child-training/tensor-validation
 integration is included with simulated resource readings.
@@ -84,3 +84,19 @@ Validation includes a real child HTTP server standing in for the native binary,
 actual SQLite/host source archives with mocked network, concurrent fetch-counter
 updates, invalid/stale evidence rejection and explicit unknown payment. It does
 not test V100 CUDA, profitability, actual demand or the operator's Windows run.
+
+## v61 acceptance boundary
+
+The uploaded int/benchmark fault, mismatched income tests, advisory repeats,
+previous-attempt errors displayed as current, token-dominated timeline and
+visible-reader anchoring have regression tests. Native serving was observed by
+the operator in v60; v61 protected V100 routing is tested with simulated transport.
+Windows RTX remains disabled by explicit operator choice. The CPU mailbox is
+retained; a fresh worker heartbeat and executed job are still required for an
+actual remote-training claim. No income has been independently verified.
+
+Concrete execution is a public source receipt plus a prepared feasibility dossier
+and registered proposal when analysis succeeds. A blocked source/analysis has
+its own receipt. This is not autonomous financial execution, a profitable strategy,
+a security scan, a sale or universal task learning. See SYNTA_RECOVERY.md for
+exact modules and runtime artifact paths.

pyproject.toml

@@ -3,7 +3,7 @@ name = "rlms"
 authors = [
     {name = "Alex Zhang", email = "altzhang@mit.edu"},
 ]
-version = "0.1.3+v100.60"
+version = "0.1.3+v100.61"
 description = "Recursive Language Models."
 readme = "README.md"
 license = "MIT"

rlm/v100/competition.py

@@ -237,6 +237,15 @@ def waiting_researcher(
         with managed_server(profile_path, root, log_path) as actual:
             yield actual
         return
+    from rlm.v100.mission_chat import preferences
+
+    if preferences(root).get("remote_helper_enabled", True) is False:
+        if fallback is None:
+            raise RuntimeError(
+                "Remote GPU disabled by operator; no accepted serving master supplied"
+            )
+        yield fallback
+        return
     active_path = root / "research/mission/active.json"
     status_path, original_state = None, None
     if active_path.exists():
@@ -328,6 +337,13 @@ def helper_client(
     from rlm.v100.remote_helper import OllamaResearchClient, remote_profile
 
     remote = remote_profile(profile)
+    if remote and root is not None:
+        from rlm.v100.mission_chat import preferences
+
+        if preferences(root).get("remote_helper_enabled", True) is False:
+            raise RuntimeError(
+                "Remote GPU disabled by operator; use the accepted V100 master or CPU jobs"
+            )
     from rlm.v100.scratch_master import ScratchClient, is_scratch
 
     scratch = is_scratch(profile)