Alpha Agent 1.40.0 — experimental Foundry v2 research integration
Local prospective source edition. Publication and complete qualification remain pending.
Begin with the operator start guide, the existing research catalogue, paper/resource publication guide and enterprise workspace guide. One catalogue connects the Foundry v2 paper, the exact retained-evidence interactive HTML, the named reproducible research companion and its distinct source/evidence release. English/French action labels, static links, descriptive accessible names and the existing reading order remain supported. The same catalogue also prominently lists the three existing enterprise manuscripts: Build the Means of Building, Reusable Launch Kits with Rejection-Only Improvement Transfer, and the earlier SUCCESSOR Ω Harness integration specification. Their PDF/TeX serving copies preserve the original bytes and root URLs. These are bounded research, local studies and modeled integration claims; the Foundry card retains its separate paper, interactive experience, research-companion and source/evidence destinations. The earlier SUCCESSOR Ω manuscript is an integration specification. Its corresponding Harness implementation remains separately tracked in unresolved PR #4863; listing the original paper does not publish or approve that implementation.
This edition composes the existing paper parent 19a13e4440da93668dd42c537ac43a011cd2a0c2
and resource child 0b6630a15c8d2765b5c242856a4525e5490bdb83 with accepted 1.39 main
e7fbd7e7fd0599f66c70ea8fe5e26b178e04f541, retaining corrected software source
5d5adc2e49a7cde4fd5619216296bf4cde948167 and boundary source
75b2ec6ed4b3cdab483bb7797f8bd84df64d8ce9. The parent’s original two inventory/index failures
remain recorded. Their catalogue-driven correction preserves every fixed historical PDF blob
assertion while checking all current records and reading-index entries. The failing parent is
not an independently mergeable stage. No held harness merge or restricted probe is included.
The corrected homepage ownership boundary preserves the existing enterprise preview under both gallery and paper generator orders. Original paper PDF/TeX bytes, all seven historical catalogue records, the Foundry interactive HTML and existing runtime implementation retain their source identities. Current release metadata, demo guides, generated pages and cache workers advance together. The explicit checkout repair retains every earlier exact edition and admits one additional complete 61-file closure for the final generated workers.
The strict classifier and passive PDF/TEX overlay allowlist are unchanged. New or changed HTML, including reverted history, requires full software acceptance. A later passive paper overlay must find linked HTML byte-identical in the qualified software source, restored site pack and current checkout; the overlay never copies HTML. Metadata hashes alone do not prove public availability. The named research companion remains separately authenticated research evidence; GitHub’s automatic repository archive is not that companion.
Research and execution limits
The experience replays a finite computational thermal-fluid study. Its retained discrepancy failure remains visible: a 4.9477% predicted/observed residual passes the 5% runtime rule while true calorimeter error is 8.6964%. Conventional engineering ties useful accuracy with fewer readings, a centralized lab is CAD 460 cheaper, and withdrawal/underutilization cases fail. Economics are modeled and contracts are Python models. Physical construction, physical deployment and commercial/EVM operation remain NOT RUN. Interactive navigation is not a numerical rerun, physical validation or proof of general autonomous capability.
The inherited bounded enterprise workspace and six-stage Discovery Lab keep their original
mission/history, evidence and recovery contracts. Imported evidence grants no execution,
promotion, commissioning, funding, signing, payment or settlement authority. USDC6 and
AGIALPHA18 stay separate zero-cap inspection routes. Installed enterprise receivers,
same-original-mission native return/reopen, external model/tool pilots and real-user outcomes
each retain their own unfinished acceptance. The four held native-source helper/browser
tests are listed for exclusion from the two current release invocations. Pre-run
EXCLUSION_PLANNED records describe intention only and leave their outcomes unqualified;
actual deselection requires observed job output. Their original restriction is not cleared.
The two scoped invocations now have a mandatory identity-only guard for the five Alpha-side source/test files named in the finite coverage policy. It rejects critical source drift before the selected tests run. It neither executes nor qualifies the four excluded behaviors; they remain unqualified. Updating those fixed identities requires a new eligible exact-source coverage disposition. Every other selected gate and current budget is preserved. This bounded source-preservation condition addresses future silent changes without claiming native behavioral acceptance or clearing the original action restrictions. The earlier 6d capability attempt is retained as a failed selection safeguard; its pre-run record does not prove nonexecution. Earlier hosted outcomes remain separate historical facts. All other selected tests and existing named release-job requirements remain applicable.
Integration and release sequence
Version 1.40.0 is a local prospective edition following the accepted 1.39 sequence. Recheck actual tag/release availability and current main at the permitted integration point. The existing paper owner retains the shared stack. No public branch, main merge, version publication, tag move or asset replacement follows merely from local preparation. Preserve every original parent/child commit and earlier failed review.
The final combined source needs fresh configured review and applicable PR checks, followed by its own full exact-main Agent/software, security, recovery, installed-package, browser, accessibility, copied-PDF/public-path, deployed-interaction and downloaded-byte acceptance. Local noninteractive checks establish only their recorded source scope. Existing 1.39 review and release evidence cannot be rebound to this source. The earlier versioned notes and research release remain unchanged; any eventual publication-status annotation is separately dated.