Stage AU — the Australian researcher

One signed number per company, so the day's ASX names can be ranked. No call, no
threshold, no direction label. Identical question and identical output contract to the
US, Japanese and European stages, which is the point: scored the same way, the
Australian result means something next to the others.

This stage places no orders and reads no broker. There is no alpaca_trade.py step
here and there must not be one. Alpaca does not carry the ASX. It is research.

What is different, and all of it matters

	US (stage E)	Australia (stage AU)
Calendar	Nasdaq vendor feed	TradingView vendor feed, dates shifted to Sydney time
Event window	after US close → before next open	close(D−1) → close(D), because 91% report pre-open
Sealed	on the day	the evening before, after the Sydney close
Option anchor	implied move + 25d skew	none; substituted by ASIC's short register
Short register	—	daily, untruncated, back to 2010 — the best in the repo
Reaction history	real dates	real dates, from the ASX archive
Hunter passes	one	one — no local-language pass, deliberately
Tail	uncapped	uncapped; no daily price limit, unlike Tokyo

The seal happens the evening before the print. Measured 2026-09-22 against the ASX
announcement record, 67 of 74 results announcements (91%) landed before the 10:00 Sydney
open. So au_universe.py --date defaults to the next ASX session and the Routine
fires at 06:30 UTC, after the 16:00 Sydney close. Sealing on the morning of the print
would seal after the event.

The vendor's date is wrong by one day for 85% of Australian rows, and the code fixes
it. The vendor stamps the UTC instant and Sydney is ten or eleven hours ahead of it, so
BHP's 08:31 Sydney lodgement on 18 August reads as 17 August. au_market.sydney_event_date()
converts the instant rather than adding a constant, so it survives the daylight-saving
change on 2026-10-04. Every row carries event_date_basis. Do not "correct" it back.

The missing option anchor is substituted, and the substitute is better than Tokyo's or
Europe's. No ASX single-stock chain is retrievable here (AAPL 22 expiries, BHP.AX
zero, measured 2026-09-22), so options is all null. In its place the baseline carries
ASIC's aggregated daily short position — and unlike every other register in this repo it
is not a 0.5% disclosure threshold: 430 of the 755 rows on 2026-09-16 were below
0.5%, minimum 0.000000%. So the level is a real number at every size and there is no
anchor_covered arm to split out, because there are no truncated zeros. What it costs is
the lag: ASIC publishes about four business days in arrears, carried as
positioning.lag_sessions and never to be treated as zero.

Half the ASX lodges a cash-flow report, not a profit result. history.filer_type is
results (Appendix 4D/4E), quarterly_report_only (Appendix 4C/5B under Listing Rule
4.7B) or none_found. These are different events with different bars and the note must
report the mix. none_found on an issuer whose three years of archive read cleanly is a
reason to doubt the print exists at all.

The hunters run one English pass and there is no second language to search. There is
no Australian-language press the wires do not read, and the entire regulated disclosure
channel is one English feed everybody reads. unpriced-hunter-au has no pre_local
field and no language_note. Do not add either. The pre_lessons freeze still runs.

Stage EU and stage CA dropped their own pre_local freeze on 2026-09-22 and now run one
bilingual pass each, so this is no longer the exception it was. The difference that
remains is that they have a second language to search and Australia does not: they carry
a prose language_note and this stage carries none. A field that would be structurally
empty on every Australian name is worse than absent — somebody would eventually pool it.

Steps

Resolve paths with python3 scripts/run_paths.py --json. Re-read the clock with
date -u; do not trust the date you were told at startup.

0. Heartbeat, before you spend anything.

python3 scripts/run_log.py --heading "Stage AU — Australia researcher — STARTED" --line "<the plan>"
scripts/publish.sh "stage AU: started for <YYYY-MM-DD>"

One cheap commit, and the only thing that distinguishes a Routine that never fired from
a session killed on its first subagent.

1. Universe. <RUN> is research/<YYYY>/<MM>/<DATE>/australia/, where <DATE> is
the event date, which is the next Sydney session, not today.

python3 researcher_australia/scripts/au_universe.py -o <RUN>/universe.json

It scans the vendor calendar, shifts every date to Sydney time, keeps the rows landing
on the target session, drops names below $200k/day of turnover, folds second share
classes into their issuer, and if more than cap survive takes a random sample seeded
by the date. Report selection.method, eligible and hunted in the note. The draw
is random on purpose: any other cut is a second ranking the scorer cannot see, and the US
run has already paid for that once.

Read the reason for an empty universe rather than guessing it. market_open false
means the ASX is shut and there is nothing to wait for. market_open null with the note
about the index tape means the fetch failed, which is a container fault and not a quiet
market — do not hunt on that file. scheduled_today 0 on an open day is a normal thin
Australian session, most likely outside the February and August reporting seasons.
Publish the empty universe, say which case it is in one line, and stop.

2. Seal the baselines. Before any hunter.

python3 researcher_australia/scripts/au_priced_in.py --universe <RUN>/universe.json --out-dir <RUN>/baselines

Sealed means sealed. Nothing downstream may revise a baseline. If the ASIC register
could not be read the script says so loudly: priced_lean_pct then falls back to the
run-up alone, which is the free control, and the baseline has no independent
directional content. That belongs in the note, not in a log line.

3. One hunter per name. Spawn unpriced-hunter-au, in waves of
australia_hunt.wave_size, publishing after each wave. Give each hunter only its own ASX
code, the window, and the path to its own baseline. Never another name's baseline, never
another hunter's findings, never your own view.

A wave that completes is banked. If a whole wave returns empty, stop and write the
— HALTED section rather than starting another.

4. Score. The scorer is the US one, unchanged and deliberately so:

python3 researcher_us/scripts/edge_score.py --run <RUN>

The ranking key is whatever edge-scores.json reports in ranking_key. Read it; do not
carry a remembered contract into the run.

5. Note. Write <RUN>/australia-note.md, answer first: the ranked table, then the
finding and URL driving the top and bottom name, then the names that could not be ranked
and why.

The note must also say, every time:

• selection.method, and that the draw was random among the eligible names
• the filer-type mix: how many names lodge a 4D/4E and how many a 4C/5B, because
  those are different events and a pooled ranking spans both
• the register lag in sessions, and that a "shorts are building" reading may be
  describing history on a name that has already moved
• lean_vs_free_control_rho from the previous resolved run if there is one, because it
  is the check that the lean has not collapsed back into the free control
• that the lean's weights are priors carried over from stage J with no Australian
  measurement behind them
• how many names carried session_unresolved
• which names sit above conviction_floor, and that over the whole US sample the sign
  was a coin flip below it
• that nothing has resolved in Australia, that this stage runs anchor-less in the
  regime archive/backtest/FINDINGS.md §33 priced at ρ=+0.073 over 104 events, and that
  one day is not a result

6. Publish.

python3 scripts/update_index.py
scripts/publish.sh "stage AU: Australia ranking for <YYYY-MM-DD>"

Resolving, after the print

python3 researcher_australia/scripts/au_resolve.py --run <RUN> -o <RUN>/resolved.json

Confirms each print against the ASX per-issuer archive, measures the move over the sealed
window off daily closes, and reports Spearman against the realised move with a
permutation p, the two free controls, every lean component separately, and a split by
filer type. It refuses to settle event_occurred for a date that has not passed —
stage EU shipped a resolver that killed names which simply had not reported yet.

Yahoo's non-US daily closes can lag, so every row carries move_pending and the summary
warns when all of them are. That is a data lag, never a day on which nothing moved.

Unlike Tokyo, the confirmation source does not expire: the ASX archive is queryable by
year and goes back years, so a late resolve loses nothing.

Budget

config/pipeline.yaml, australia_hunt. When the day would exceed the cap, shed by
australia_hunt.degrade_order and record what you shed in the run log. Half a run that
finishes beats a full one that gets cut off.

The rule that outranks the rest

Never fabricate a number. Every company-specific figure carries a source URL or is marked
unavailable/null. A missing anchor correctly lowers confidence downstream; an
invented one corrupts everything after it.

This is research, not investment advice. Keep the disclaimer from config/pipeline.yaml
on the note.