Skip to content
Research library
Findings update 002 July 16, 2026 Locally source-verified

One artifact crossed the deadline

One local fixture crossed a frozen completion threshold. The surrounding controls keep that result narrow, expensive in bytes, sensitive to one host, and unrelated to production or market validity.

Waggle + Kea

8/8 vs 6/8

The resident native-state arm completed a passing fixture; the matched resident full-text control remained incomplete at the same deadline.

Frozen budget

30.422 s vs 45.000 s

The native path included planning, serialization, qualification, assembly, and testing; the control reached the deadline after six decisions.

HEATWAKE

0 empirical bytes

The operator representation advanced, but the empirical model, forecast, edge, and trading claims did not move.

Jump to a findings section
Executive finding

A narrow positive result and a decisive negative result.

Waggle + Kea first failed an eight-module fresh-process deadline gate, then completed a different preregistered eight-part executable fixture through one resident native-state Executor while the matched resident full-text control reached six decisions and timed out. The positive result is real inside that single local model, host, task family, and 45-second envelope; it is not a learned language, general quality result, or overall-efficiency win. Separate provider-session studies preserved exact fixture decisions but did not produce a cost advantage. HEATWAKE improved its synthetic operator representation and local lifecycle while admitting zero empirical bytes, opening no final test, and retaining ABSTAIN.

How this was verified

Claims, controls, receipts, reruns.

1

Snapshot comparison

The July 15 public source labels were compared with later committed reports, protocols, receipts, test definitions, and exact source commits for both research programs.

2

No-call replay

Four changed Waggle/Kea gates were replay-verified from their sealed artifacts. The verification launched no provider calls, local model inference, external calls, or authority effects.

3

Private evidence boundary

HEATWAKE's tracked source snapshot was verified locally with two non-public retained evidence inputs. Their contents and locations remain private; the public ledger exposes only aggregate status and non-reversible record hashes.

4

Claim grammar

A completed fixture is reported as a completed fixture. Failed gates remain negative results, proxies remain proxies, and no local source verification is described as independent validation.

Program 01

Waggle + Kea

Machine coordination with an independent qualification plane

Full program

Verdict

Supported only in one preregistered local fixture: a resident native-state arm completed 8/8 decisions and a passing allowlisted artifact inside 45 seconds while the matched resident full-text control reached 6/8 and remained incomplete. Not supported: general quality, a learned language, or overall efficiency.

The new checkpoints separate four questions that should not be collapsed: provider continuity, provider fan-out economics, fresh-process deadline capability, and resident logical-context deadline capability. The first preserved bounded decisions, the second and third were negative, and the fourth crossed its exact fixture bar.

Finding 1.1 Boundary finding

Provider-session continuity preserved bounded decisions

2/2 session parity · 2/2 no-context abstentions

In the sealed study, two fresh resumed Executors returned the same exact multi-fact decisions as two full-text controls. Two no-context controls abstained, and Kea refused two ambiguous or unsupported tasks before a provider process.

Boundary: Planner plus resumed-session input totaled 36,336 reported tokens versus 17,957 for the paired full-text controls. Cost was not reported for every call. This is ordinary provider-managed continuity, not native state, independent semantic decoding, or an efficiency result.

Finding 1.2 Negative result

Provider-session fan-out failed its utility gate

8/8 parity · task-8 cost proxy 14.6513% worse

Eight independent provider-session forks and eight full-text controls returned byte-semantic exact decisions, while Kea preflight-abstained two unsupported lanes. The forked route showed no public-API-equivalent break-even through task eight.

Boundary: The dollar curve is a frozen public-API-equivalent proxy, not the user's actual subscription bill. Only three of eight forks reported any cache read, and none showed meaningful retained-context cache affinity.

Finding 1.3 Negative result

Fresh processes did not complete the required artifact

Native 6/8 · full text 4/8 · builds 0/2

Under the preregistered 45,000 ms deadline, native state produced two more exact decisions than repeated full text. Both arms still lacked required modules, so both deterministic integration builds failed.

Boundary: The result remains negative at the frozen eight-module artifact bar. Partial throughput was not reinterpreted as completed capability after the fact.

Finding 1.4 Supported in scope

Resident logical contexts crossed the fixture deadline

Native 8/8 + PASS · full text 6/8 + incomplete

A separate Planner plus one resident native-state Executor completed all eight exact allowlisted decisions, assembled the release-manifest fixture, and passed its integration test at 30,422.008834 ms. The matched resident full-text control reached six decisions and the 45,000 ms deadline.

Boundary: This was one unsigned local run on one host and one task family. The 315,696,149-byte state represented a 9,223-byte shared prefix; scheduling, thermal state, and OS cache were not controlled. No storage, network, energy, cost, or overall-efficiency advantage is supported.

Not demonstrated

  • No learned Waggle protocol, private language, neuralese, cross-model transfer, or independent semantic decoding of opaque state.
  • No general coding or quality result: the model selected bounded identifiers and an allowlisted deterministic assembler emitted the fixture source.
  • No token, dollar, storage, network, energy, or overall-efficiency claim.
  • No production traffic, customer result, revenue, deployment, or authority grant.

Next legitimate gates

  1. 1. Replicate the resident deadline result across seeds, artifact families, system states, and independent host runs before generalizing it.
  2. 2. Measure variance, I/O, energy, retained-state lifecycle cost, and matched compact symbolic controls.
  3. 3. Keep Kea qualification and authority envelopes separate from the opaque transport in every follow-up.
Program 02

HEATWAKE

A synthetic-only liquidity research system that still abstains

Full program

Verdict

Supported at synthetic operator scale: a deterministic twelve-channel market-state representation, bounded descriptive overlays, a read-only physical-data gate, and a product-owned local lifecycle. Not supported: any empirical forecast, market edge, profitability, or trading readiness.

The tracked HEATWAKE changes improve how synthetic source state is represented and how the local operator is handed to the founder. They do not admit market data or promote a model. The empirical verdict therefore remains unchanged.

Finding 2.1 Supported in scope

Field View made the synthetic state more inspectable

12 channels · 720 × 96 field · 9,427 contours

The operator compiles 3,600 one-second synthetic book states into a fixed-order time-by-price volume, then derives 96 displayed-state flow vectors and 40 execution-with-limited-move observations under hard ceilings.

Boundary: The representation is descriptive only. It does not identify participants, infer intent, detect spoofing, expose hidden liquidity, predict direction, or establish empirical validity.

Finding 2.2 Boundary finding

The physical-data gate stayed closed

0 external disks · 0 source bytes · WAITING_FOR_ORDERED_DRIVE

A GET-only local projection now presents CONNECT, ENCRYPT, CUSTODY, and ENVELOPE as a fail-closed founder workflow. It observed no eligible external research disk and executed no mutation.

Boundary: Formatting, encryption-secret handling, activation, acquisition, provider access, training, final-test access, orders, capital, and spend remain separately gated.

Finding 2.3 Supported in scope

The local operator gained durable process ownership

28 UI files / 95 tests · 64 service tests

The reviewed local operator can start, report status, restart, and stop through a product-owned lifecycle instead of depending on a transient verification shell. The sealed safe-local audit covers 288 runtime sources with zero high findings.

Boundary: This is a loopback-only founder workflow, not a public deployment, live feed, login service, production runtime, or execution authority.

Finding 2.4 Boundary finding

The empirical conclusion did not move

0 admitted third-party bytes · final test unopened · ABSTAIN

No empirical sample was imported, trained, evaluated, or opened as a final test. The operator continues to withhold its probability field and disables orders.

Boundary: The synthetic representation and green product tests are engineering evidence, not evidence of forecast quality, market edge, profitability, or trading readiness.

Not demonstrated

  • No admitted live or historical market corpus, empirical fit, empirical generalization, validated forecast, market edge, or profitability result.
  • No order, trade, capital deployment, provider acquisition, public deployment, or production readiness.
  • The safe-local audit is reproducible only with locally retained non-public evidence inputs; it is not a Git-only or independent reproduction.
  • The storage-workflow visual check remained blocked by loopback browser policy; no visual PASS is claimed for that specific surface.

Next legitimate gates

  1. 1. Wait for William to attach and manually qualify the ordered encrypted research volume.
  2. 2. Complete the exact rights, custody, and acquisition envelope before requesting any third-party byte.
  3. 3. Run frozen empirical baselines and controls before opening any final test or making any directional claim.
Cross-program finding

Cross-program conclusions.

A deadline can be a capability bar

The completed-artifact criterion prevented partial throughput from being promoted. Only the later resident fixture crossed the exact preregistered bar.

Negative controls stayed first-class

The provider fan-out cost proxy and fresh-process artifact gate remained negative even though both contained locally interesting sub-results.

Product work remained separate from scientific validity

HEATWAKE's representation and lifecycle improved while its empirical conclusion remained unchanged at zero admitted bytes and ABSTAIN.

Opaque transport still carries no authority

Kea qualifies lineage, compatibility, bounded outputs, and refusal. Neither provider references nor native state can authorize an external act.

Limitations and disclosures

Limits on the reported findings.

  1. 1All verification described here is local source verification by K&E Studios Research, not peer review or independent validation.
  2. 2The archived provider-session studies used provider processes during their original formal runs. This publication replay launched zero provider calls and did not treat prior paid or hosted output as a fresh run.
  3. 3HEATWAKE's safe-local audit correctly refused the dirty active research tree. The exact tracked source commit plus two non-public locally retained evidence inputs regenerated the sealed zero-high-finding record; unfinished work was excluded.
  4. 4The public evidence ledger contains aggregate measurements, commit labels, and non-reversible hashes only. It does not expose private repositories, paths, payloads, model state, prompts, raw receipts, credentials, or reconstruction-enabling artifacts.
Fresh local rerun ledger

Rechecked from the source trees.

Waggle + Kea changed gates

PASS

Provider-decision, provider-fan-out, native-deadline, and resident-deadline focused suites passed against the committed reports; TypeScript and the production build passed. The replay used zero provider calls, zero local model inference, zero external calls, and zero authority effects.

HEATWAKE tracked operator snapshot

PASS

The operator passed lint, strict TypeScript, 95 tests across 28 files, and a production build; the operator service passed 64 tests and the lifecycle suite passed 10 tests.

HEATWAKE safe-local audit

PASS WITH PRIVATE INPUT BOUNDARY

The exact tracked commit plus two non-public retained evidence inputs regenerated audit v57 over 288 runtime sources with zero high findings. A Git-only checkout does not reproduce this audit and is not claimed to do so.

External effects from heartbeat verification

ZERO

No provider call, paid request, email, source acquisition, empirical byte admission, final-test open, order, trade, capital movement, authority grant, or research-harness deployment occurred.

The material delta is one bounded completed-artifact result under a frozen local deadline, surrounded by negative controls that prevent it from becoming a language, general-quality, or overall-efficiency claim. HEATWAKE's operator advanced, but its empirical result remains exactly where it was: unopened, untrained on market data, and ABSTAIN.

Author and research contact

William Keenan · K&E Studios Research

william@kestudios.dev
Author profile