Mover research refresh 2026-09-28: extreme-mover data, factor evidence, layer deltas since #358 by seathatflowsinourveins · Pull Request #463 · native-agent-stack/native-agent-stack

seathatflowsinourveins · 2026-09-28T16:37:11Z

Scope

This PR is lane:trading and adds one dated research record, catalogs/us-equities/research-20260928-movers.md. The catalog index links it, and both files are re-registered in manifests/evidence.json following the docs/lanes.md recipe. It changes no code or selections.

It answers the north-star questions the user asked on 2026-09-28, while restarting the paper lane:

R1: where to get historical daily extreme movers.
R2: which factors have published evidence for getting in before a jump versus trading after a detected jump.
R3: what changed after the September 26 convergence (Trading research convergence 2026-09-26: 13 layers, GPT-6 live refutation, verbatim-checked votes (verdict-wave input) #358) in the engine, full-market monitor, strategy-framework and research-RAG layers.

Findings

The v1/v2 study's mover set misses verified events. It covers 498 of 594 verified +20% events (83.8%), under the protocol's own 95% bar.
- Likely but unmeasured cause: bars were collected by today's symbol, so a reused ticker resolves to its current holder.
- Free fix: re-collect from Alpaca SIP with asof.
- Paid alternatives are listed with prices and overturn conditions. Buying any of them is the user's decision.
The evidence does not support getting in before a catalyst jump like KOD's +149%. Lottery and MAX traits predict low returns, and MAX survives only in microcaps under Hou, Xue & Zhang 2020.
- The supported direction is news-aligned continuation after the jump, held t+1 to t+5.
- The record ranks three preregistration candidates, each with its overturn condition.
No recorded selection changes since Trading research convergence 2026-09-26: 13 layers, GPT-6 live refutation, verbatim-checked votes (verdict-wave input) #358. The engine selection is in runtime-target.json; the strategy-framework and research-RAG layers have none.
- NautilusTrader 2.0.0rc5 is still the latest tag. rc6 is untagged and carries 68 breaking changes.
- alpaca-py 0.44.0 is still the latest release.
- One monitor documentation fact is new to the repo record: movers resets at the open.
- The strategy-framework and research-RAG layers stay open, each with a named comparison.

Evidence-class table

Claim	Evidence class	Source
R1–R3 findings	Research by three read-only Claude `stack-researcher` subagents (latest Opus, effort max), 2026-09-28 16:03–16:40Z	Tags D (documented), O (observed), A (abstract only) and U (unverified) in the record; citations inline
Account data limit of 10,000/min (Algo Trader Plus)	Re-checked by the coordinator against repo source	`blueprints/us-equities/data-lake/README.md:3-5` (limit); `blueprints/us-equities/incentive-monitor/README.md:7` (tier)
The paper key returns 401	Native probe on this host, plus #360's own note (re-checked)	`tools/credentials/alpaca_rate_limit_probe.py`: `GET /v2/account` → 401 at 16:03:25Z
Publication integrity	`local_integration`	`python3 scripts/validate.py` → passed (7,727 hashed files)

Paper series pre-registration (hash commitment)

The PC mover series pc-20260928 was frozen at about 12:05 ET, before any broker I/O. Its files stay private until it runs, and results come in a separate PR.

FREEZE.md sha256 905032d61b95448109e73eded63d493bdfe8202e75abe6fb0dac243360e81df3
config sha256 4b14a1b5b465ba6de6070567cb0478131ba71313a32af05b87b0002878d2dfd9
run script sha256 3f014762fdcf3a56d14f853b47a9063731072c0f4d93a3a68faad9f4070e2df0
engine sources: main 3058b237, 60 files, SHA256SUMS 77de4228864b73c5aa436399359c13a5e2c4388956ca1e8c200c97282b41134e

It has not started. Trial 1's scan got HTTP 401 at 12:02:53 ET, so no order was sent.

SOTA sources

The record cites these published sources inline (tags D, O, A and U). The main ones behind each finding:

R1, historical movers and data:
- Alpaca market data: https://docs.alpaca.markets/us/docs/about-market-data-api, https://docs.alpaca.markets/us/reference/stockbars (asof, feed, adjustment), https://docs.alpaca.markets/us/reference/movers-1 and https://docs.alpaca.markets/us/docs/streaming-market-data.
- Paid alternatives, informational only: https://massive.com/docs/rest/stocks/aggregates/daily-market-summary, https://sharadar.com/prices, https://databento.com/equities and https://eodhd.com/pricing.
R2, factor evidence:
- Hou, Xue & Zhang (2020), on MAX surviving only in microcaps, as cited in the record;
- https://arxiv.org/abs/2409.13740 and https://arxiv.org/abs/2510.21652;
- the stocks-in-play opening-range reference, https://www.quantconnect.com/research/18444/opening-range-breakout-for-stocks-in-play/.
R3, layer currency:
- NautilusTrader, Port the Interactive Brokers adapter onto ibapi 5.0.0 nautechsystems/nautilus_trader#5041, and the tag list (rc5 latest);
- alpaca-py, https://github.com/alpacahq/alpaca-py/releases (0.44.0 latest);
- paper-qa, https://github.com/Future-House/paper-qa/releases/tag/v2026.08.12;
- SEC EDGAR APIs, https://www.sec.gov/search-filings/edgar-application-programming-interfaces.

This PR changes no component or selection; it records research only.

Addendum 2026-09-29 (trading lane: ecosystem-roadmap-2026)

The pc-20260928 series ended not_started. Every probe got a 401, and the freeze's 15:00 ET stop ended it.
Keyring-launched series ran later. The user stored two paper key pairs in the kernel keyring. The extended-hours series ext-20260928 then ran on account 2 from 16:26 ET. Its receipt comes in a separate lane:trading PR.
Both accounts were flat afterwards. A read-only check at 2026-09-29T00:3xZ returned 0 positions and 0 open orders on each.
Freeze timing, checked against the private workspace. The three committed files match the hashes above: FREEZE.md 905032d6…, config 4b14a1b5… and run script 3f014762…. Their modification times are config 11:58:58, FREEZE.md 11:59:49 and run script 12:02:44 ET, all before trial 1's 12:02:53 ET scan that got the 401. So the frozen files predate every broker request. The "about 12:05 ET" in FREEZE.md is an approximate stated time. No order was sent.
The asof re-collection that this record proposes (R-NOW-1) is now running as its own unit and PR.

Mover v3 guard

No Mover v3 sealed file, outcome data or pinned study data was opened. The record publishes coverage totals only.

🤖 Generated with Claude Code

…e, layer deltas since #358 Three read-only research units (R1-R3) for the north star, recorded in catalogs/us-equities/research-20260928-movers.md and linked from the catalog index: - R1: historical daily extreme movers; the v1/v2 candidate set covers 498 of 594 verified +20% events (83.8%, under the protocol's 95% bar) because bars were collected by today's symbol; free fix is an asof re-collection; paid options are user decisions. - R2: published evidence does not support pre-positioning ahead of catalyst jumps; post-detection news-aligned continuation is the supported direction; top three preregistration candidates with overturn conditions. - R3: no layer selection changes since #358 (NautilusTrader rc5 still latest tag, alpaca-py 0.44.0); monitor, strategy-framework and research-RAG deltas. Co-Authored-By: Claude Opus 5.5 <[email protected]>

…idence review An evidence-reviewer checked every repo-cited claim against source: - holdout: the study's holdout outcomes are unread, but the window is partly exposed (broad-universe 2024-2026 evaluation; the price audit's disclosed per-event gains) - engine: rc5 pins Rust ibapi =3.3.0; runtime-target recorded #5041 at =4.1.0 - monitor: only the movers reset-at-open fact is new; the rest was already recorded - volume breakout: 11.7% is the +100-300% tier, precision about 0.25% vs 0.0043% - tier naming, requested vs served symbols, cause of the coverage gap marked unmeasured, "none passed development" instead of "all negative", non-news day-2 continuation noted, selections cited from runtime-target and the incentive monitor, trial deviations stated Co-Authored-By: Claude Opus 5.5 <[email protected]>

seathatflowsinourveins · 2026-09-28T16:53:00Z

Independent review round (one review, one repair). An evidence-reviewer subagent checked every repo-cited claim in the record against source, read-only, at 325fdb18. It found 3 contradictions and 7 overstatements. The coordinator re-checked the main citations and fixed all ten in a09daf58. The validator passed (7,727 hashed files).

#	Finding	Fix
1	"Holdout never read" overstated: the window is partly exposed (`broad-universe/README.md:95,130-131`; `mover-early-entry/protocol.json:7`)	The study's holdout outcomes are unread; the window's partial exposure is stated
2	"runtime-target pins 4.1.0" was wrong	rc5 pins Rust ibapi =3.3.0; runtime-target recorded #5041 at =4.1.0 (`runtime-target.json:72,74`)
3	Monitor "doc facts not in the repo" was wrong for 3 of 4 facts	Only the `movers` reset-at-open fact is new; the rest cite `incentive-monitor/README.md:14,23` and `architecture/README.md:112`
4	Volume breakout 11.7% set recall against prevalence	Now the +100–300% tier, 19 of 163, derived precision about 0.25% against a 0.0043% base rate
5	Algo Trader Plus cited to a file lacking the tier	Tier now cited to `incentive-monitor/README.md:7`
6	"No selection changed" against #358's "not a selection"	Now "No recorded selection changed"; selections cited from runtime-target and the incentive monitor
7	Coverage-gap cause stated as fact	Marked likely but unmeasured, with the four known gaps cited
8	24,664 symbols	Now 24,664 requested, 17,997 with bars
9	"All negative after costs"	Now "none passed development"; non-news day-2 continuation noted as already lost
10	Tier naming and "split artifact"	Now the +300–1,000% tier and price-basis artifacts (`deviations.json:77-79`)

Residuals, not verified by the reviewer:

Web and paper claims: the vendor docs and papers keep the D/O/A/U tags from the researchers.
The PR Mover v3 round 18: pre-freeze residual coverage, seal recovery for dry runs and live samples, ABJK/LPS literature correction — not frozen #360 body and the two 401 observations: the coordinator checked these directly.
The subagents' statement that they opened no Mover v3 sealed file.

The reviewer found no Mover v3 leak.

seathatflowsinourveins · 2026-09-28T19:00:31Z

Paper series pc-20260928 outcome: not_started (recorded under the freeze's stop rule).

Result. No order was sent, and the account was never touched beyond read-only probes.
Failed scan. Trial 1's scan got HTTP 401 at 12:02:53 ET.
Probe watcher. From 12:37:52 to 14:58:02 ET, a read-only watcher sent one GET /v2/account every 2 minutes through tools/credentials/alpaca_rate_limit_probe.py.
- All 71 probes returned 401.
- Its job was to start the frozen trial as soon as a key worked.
Stop. The freeze's series procedure stops the series at 15:00 ET, and the watcher exited there.

This is a failed condition, recorded as-is: a credential input the policy names as a concrete input to resolve (docs/paper-lane-policy.md). It is not a strategy result.

Next steps:

The user stores a new paper pair with tools/credentials/open_credential_terminal.sh alpaca-paper --probe.
A new series is frozen for the next session date, at the then-current main, from the same config and procedure. The FREEZE, config and run-script hashes above stay as the record of this date's plan.

seathatflowsinourveins added the lane:trading Trading lane: sim, paper and live north star label Sep 28, 2026

seathatflowsinourveins enabled auto-merge (squash) September 29, 2026 00:49

seathatflowsinourveins merged commit 11648f9 into main Sep 29, 2026
32 of 50 checks passed

seathatflowsinourveins deleted the claude/north-star-paper-research-20260928 branch September 29, 2026 01:07

This was referenced Sep 29, 2026

Mover coverage asof re-collection (R-NOW-1): frozen plan, runner, E2 totals; E1 blocked on pinned private inputs #485

Merged

Roadmap update 2026-09-29 (later): trading lane owned, paper series frozen, foundation sweep running, ai-memory candidate only #496

Merged

Navigation Menu

Repository navigation

Mover research refresh 2026-09-28: extreme-mover data, factor evidence, layer deltas since #358 - #463

Mover research refresh 2026-09-28: extreme-mover data, factor evidence, layer deltas since #358#463
seathatflowsinourveins merged 2 commits into
mainfrom
claude/north-star-paper-research-20260928

seathatflowsinourveins commented Sep 28, 2026 •

edited

Loading

Uh oh!

seathatflowsinourveins commented Sep 28, 2026

Uh oh!

seathatflowsinourveins commented Sep 28, 2026

Uh oh!

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

seathatflowsinourveins commented Sep 28, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Scope

Findings

Evidence-class table

Paper series pre-registration (hash commitment)

SOTA sources

Addendum 2026-09-29 (trading lane: ecosystem-roadmap-2026)

Mover v3 guard

Uh oh!

seathatflowsinourveins commented Sep 28, 2026

Uh oh!

seathatflowsinourveins commented Sep 28, 2026

Uh oh!

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

seathatflowsinourveins commented Sep 28, 2026 •

edited

Loading