Skip to content

M54–M59: voices by lead over price; money after costs; every department daily; capital follows the record; methods judged by mechanisms; the path to real money - #1

Merged
martex-dev merged 9 commits into
mainfrom
claude/youthful-sagan-zn3k0s
Sep 26, 2026
Merged

M54–M59: voices by lead over price; money after costs; every department daily; capital follows the record; methods judged by mechanisms; the path to real money#1
martex-dev merged 9 commits into
mainfrom
claude/youthful-sagan-zn3k0s

Conversation

@martex-dev

@martex-dev martex-dev commented Sep 25, 2026 •
edited
Loading

Copy link
Copy Markdown
Owner

Six milestones, numbered after main's M53, which this branch has merged in. Each milestone is one commit.

M54: the company chooses whom it follows (ADR-0055)

  • Each voice's record. Every X author, Telegram channel and Discord channel is measured by the 24h move after its posts against desk peers, and the same over the 24h before, by episode. A move in the post's own bar counts as before.
  • The bar. A voice leads or trails price by an exact sign test at 0.05 ÷ voices measured.
  • A daily curation seat follows and drops voices from a closed set. aurelis social voices|curate, /voices.

M55: money is judged after costs, by episode (ADR-0056)

  • mechanism/earnings.py: paper round trips after fees, grouped by independent episode, with a sign test and the drawdown.
  • Two bars. Earning needs the family bar; losing is judged at 0.05 undivided.
  • A losing scheme is suspended by Risk. The mandate gains a fourteenth condition, earning.
  • tests/test_world.py no longer reaches Coinbase.

M56: every department works, every day (ADR-0057)

  • judge and discover are standing duties, so research no longer stops at its first success.
  • autonomy/duties.py: a daily duty per idle department, run by its own agent:
    • audit, integrity and health checks, in software, each raising alerts;
    • a lesson and a research director's memo written into the shared brain, withheld if they cite unshown figures.
  • aurelis duty list|run.

M57: capital follows the record (ADR-0058)

  • A declared ladder in mechanism/sizing.py: 5% gathering or luck, 10% earning after costs, 0 losing. Schemes are capped at half the book.
  • trade_firings sizes positions from the live allocation, instead of a constant that ignored it.
  • Book.withdraw. A daily Portfolio Manager duty withdraws and re-allocates where the record moved. Risk still assesses every order.

M58: a method is judged by the mechanisms too (ADR-0059)

  • Discovery fitness: the mechanisms decided under each method, schemes versus retired.
  • Evolution rewrites a method when its views or its mechanisms fail. A colleague whose mechanisms thrive teaches it.

M59: the station shows the path to real money (ADR-0060)

  • /mandate ("money") shows the latest assessment condition by condition, every scheme's paper record after costs, and the verdict history. It states plainly that no live adapter exists.

Tests

  • New or extended: test_voices.py, test_earnings.py (sizing and the money page), test_duties.py and test_evolution.py (discovery fitness).
  • ruff check . and mypy are clean. The full suite passes: 1104 passed and 14 skipped, run with bare pytest as CI does.

🤖 Generated with Claude Code

https://claude.ai/code/session_01Vxtju5i7j2GwPr9hkbQaGT

@claude
M53: the company chooses whom it follows, by each voice's lead over p…
e1c5022 
…rice

A handle was followed because a token linked it or the operator typed it,
and nobody measured whether it was worth reading. ADR-0052 left the
curation seat as next.

- A voice's record, in software: every X author, Telegram channel and
  Discord channel the company has read, each post priced as the
  instrument's 24h move after it less the median of its desk's peers, and
  the same over the 24h before. Counted by episode (ADR-0040), read from ten.
- A move inside the bar a post was made in counts as before, never as a
  lead; a bar still open when fetched is not a close; a gap is unknown, not
  the last price; only what was recorded by the moment measured is read.
- Leads or trails price by an exact sign test at 0.05 divided by the voices
  measured, because the agents are shown the best of many records.
- Once a day under the news grant a market-intelligence agent follows voices
  ahead of their peers and drops followed ones that do not lead, from a
  closed set, with a reason; the record is on each row and the sitting cites
  the full record as an artifact. No call when nothing is eligible.
- Episodes since a follow are counted apart: the only ones nobody chose on.
- `aurelis social voices`, `aurelis social curate`, and the station's /voices.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vxtju5i7j2GwPr9hkbQaGT
@claude
M54: money is judged after costs, by episode
8741ba0 
A candidate scheme was chosen on calibration and traded on paper, and its
P&L was "reported, never judged". A scheme can beat the drift and still lose
to its fees and to the slippage of an order placed an hour after its trigger;
it kept trading, and nothing told the operator whether the company would
have made money.

- mechanism/earnings.py: round trips closed at their horizon, after fees,
  grouped by the episodes of their predictions (ADR-0040), counted by an
  exact sign test so one tripled memecoin cannot carry a record; drawdown
  alongside; gathering below ten episodes.
- Earning needs the bar for the family of schemes measured; losing is judged
  at 0.05 undivided, because a stop protects the book.
- A scheme losing after costs is suspended by Risk, once, with the numbers:
  it opens nothing more, closes what it holds at the horizon, and its
  predictions are still scored.
- The mandate gains a fourteenth condition, `earning`, right after `scheme`.
- The mechanisms page, `aurelis mechanism trades`, the brain and the feed
  show each traded scheme's record after costs.
- sign_test moves to core/stats.py, shared with the voices record.
- test_world's service test stubs the book and trades feeds: it was reaching
  Coinbase for real, and failed wherever that call is refused.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vxtju5i7j2GwPr9hkbQaGT
@martex-dev martex-dev changed the title M53: the company chooses whom it follows, by each voice's lead over price M53–M54: curate voices by their lead over price; judge money after costs Sep 25, 2026
@claude
test_earnings: build its own scheme instead of importing another test…
e8d0faf 
… module

CI runs bare `pytest`, which does not put the repository root on sys.path,
so `from tests.test_schemes_trade import ...` failed at collection. The
repository's tests do not import each other; this one now carries the edge
market, the stated mechanism and the scoring it needs, as the M31 tests do.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vxtju5i7j2GwPr9hkbQaGT
@claude
Merge main: its M53 comes first; voices become M54, money M55
c9a033e 
main took M53 and ADR-0054 for "a changed catalogue is a new question".
This branch's milestones follow it: the voice curation is M54 (ADR-0055)
and money judged after costs is M55 (ADR-0056), renumbered in the docs,
the ADR files and the code comments that cite them.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vxtju5i7j2GwPr9hkbQaGT
@martex-dev martex-dev changed the title M53–M54: curate voices by their lead over price; judge money after costs M54–M55: curate voices by their lead over price; judge money after costs Sep 25, 2026
@claude
Merge main: the Telegram no-preview follow-up
5f2247d 
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vxtju5i7j2GwPr9hkbQaGT
@claude
M56: every department works, every day
6995cc0 
Seven of the seventeen launch agents never acted in the service: the CEO,
the research director, Operations, the auditor, governance, knowledge and
infrastructure had no duty in a wake. And judging and discovery ran only
while `calibrated` and `scheme` were unmet, so the forward record stopped
growing at the first success and the hunt stopped once MEC-0001 became a
candidate scheme.

- judge and discover are standing duties (after source, so a changed
  catalogue is still asked about first); their no-repeat rules still hold.
- autonomy/duties.py: a daily duty per idle department, run by its own
  agent and recorded as org.duty_done with that agent as actor:
  audit (ledger chain, round trips held late, agents mostly refused),
  integrity (mechanism and view seals, the day's recordings),
  health (wakes, longest gap, model calls, open alerts) -- software, with
  each problem an alert raised by the agent holding the scope;
  lessons (a Lesson row and a brain note from what the day closed) and
  memo (the research director's note every seat reads) -- one model call
  each, none on a day with nothing in it, withheld if they cite a figure
  the agent was not shown.
- The wake runs them once a day before the brain is written;
  `aurelis duty list|run`; the feed has a line for each.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vxtju5i7j2GwPr9hkbQaGT
@martex-dev martex-dev changed the title M54–M55: curate voices by their lead over price; judge money after costs M54–M56: curate voices by lead over price; judge money after costs; every department works daily Sep 25, 2026
@claude
M57: capital follows the record
736f3a1 
Every scheme was sized at equity x 5% from a constant; the allocation row
trade_firings wrote was never read again and nothing ever withdrew one, so a
scheme earning after costs and one indistinguishable from luck were the same
size, and the Portfolio Manager decided nothing after the first day.

- mechanism/sizing.py: a declared ladder from the after-costs record --
  5% gathering or luck, 10% earning, 0 losing -- with schemes together
  capped at half the book.
- trade_firings sizes positions from the scheme's live allocation; a zero
  share opens nothing and says why.
- Book.withdraw keeps the row with its reason (portfolio.allocation_withdrawn).
- A daily Portfolio Manager duty re-reads each scheme's record and withdraws
  and re-allocates where the ladder says otherwise; a matching share is left
  alone. Risk still assesses every order.
- The mechanisms page shows each scheme's share of the book.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vxtju5i7j2GwPr9hkbQaGT
@claude
M58: a method is judged by the mechanisms too
5ccfd1c 
A method is carried to every seat, but evolution judged it only by the
agent's views (ADR-0049's open item): an agent whose every tested mechanism
was retired kept its method while its views were no worse than chance.

- DiscoveryFitness: mechanisms stated since the method was adopted that the
  company decided on -- candidate schemes or retired. Unproven below five
  decided; failing at five or more with none a scheme; thriving at half.
- Evolution rewrites a method whose views or whose mechanisms fail. A
  colleague whose mechanisms thrive teaches a discovery failure, with the
  retired mechanisms and why they were retired in the material.
- org.evolution_ran records every agent's discovery verdict.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vxtju5i7j2GwPr9hkbQaGT
@claude
M59: the station shows the path to real money
d55b5e6 
The company's conditions for asking to trade real money were assessed every
autonomy cycle and shown only in the terminal.

- /mandate ("money" in the navigation): the latest stored assessment,
  condition by condition with what the record says; every scheme's paper
  record after costs by episode; the last twenty verdicts; and a plain
  statement that no live adapter exists.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vxtju5i7j2GwPr9hkbQaGT
@martex-dev martex-dev changed the title M54–M56: curate voices by lead over price; judge money after costs; every department works daily M54–M59: voices by lead over price; money after costs; every department daily; capital follows the record; methods judged by mechanisms; the path to real money Sep 26, 2026
@martex-dev
martex-dev marked this pull request as ready for review September 26, 2026 08:19
@martex-dev
martex-dev merged commit e3ba5df into main Sep 26, 2026
4 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

No reviews

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

2 participants

@martex-dev @claude