Promote: the Dogecoin and Zcash parity surfaces - #35

…ge-tap-over-tls

Write down how a source entry reaches an authority that speaks TLS

The two chains rendered a capability report where Bitcoin renders a
dashboard, and their navigation hid Mining, Charts, and chain Docs
behind an activeChain gate. This builds the missing product:

- The route registry gains mining, graphs, and docs categories, chain
  switching preserves the section, and a chart child survives a switch
  when the destination chain has an equivalent (a registry maps
  block-sizes-weights to block-sizes and block-health to
  block-interval).
- The chain dashboard: a full-width block timeline with one cube per
  upcoming target slot and one per confirmed block, heights above every
  cube, the direction toggle shared with the Bitcoin strip, fee
  guidance in chain-native units, a Universe Lens over the real pending
  set (the Bitcoin treemap renderer fed chain-read items, with a table
  fallback), mining facts, recent blocks with evidence-based miner
  attribution, and the old capability report filed as a status drawer.
- Zcash slots state eligibility, never order: tier chips stay on the
  cubes and the strip says selection is randomized under ZIP-317.
- A chain mining page: network facts, rewards, pool shares with an
  honest Unknown bucket and named evidence, recent blocks, and a
  how-this-chain-mines explainer covering AuxPoW and Equihash.
- Chain charts: an echarts shell with ten chart routes per chain,
  ranges disabled until stored history covers them, table fallback,
  and a pool ranking page.
- Chain docs at /dogecoin/docs and /zcash/docs: fourteen sections from
  dashboard field definitions to the API list, privacy limits on
  Zcash, and the candidate-versus-projected distinction.
- The mempool page keeps its bucket row; the shared cube grammar and
  page skeleton move to partials so the surfaces cannot drift apart.
- Gates: parity-regressions.test.mjs pins the structure in CI, the
  production smoke checks the new data families with a
  backfilling-aware honesty rule, the browser smoke audits the
  rendered timeline, and the visual matrix covers the eight new routes
  in both themes with fixtures for every new endpoint.

Local verification: production build green, 407 universe unit tests,
40-screenshot visual run with zero contrast, a11y, overflow, and
console findings.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

The mobile gate measured the range, window, lens, and navigation pills
at 31 pixels tall and as little as 39 wide, and the protocol and panel
links at 24 pixels. All of them repeat in a row or a column, which is
exactly the case the 44 pixel rule is about: a miss selects the
neighbour.

Each takes the house pattern already used by the protocol navigation:
a coarse pointer or a narrow window gets a 44 pixel minimum in both
directions. Desktop with a mouse is untouched. The two window rows that
did not wrap now do, because a thumb-sized pill is wider than its
label and the panel must not push sideways instead.

Verified: mobile gate clean on all four new routes across seven window
sizes; the visual matrix stays clean at 320, 375, and 1440 in both
themes.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

Give Dogecoin and Zcash the full explorer surface Bitcoin has

Dogecoin returns up to a thousand pending transactions in one request.
Zcash refuses anything above two hundred as a bad request rather than
trimming it. The dashboard asked both for four hundred, so the Zcash
lens and its arrivals list were empty on every real load while every
fixture and every unit test passed: a fixture answers whatever it is
asked.

The ceilings now live beside the request that has to respect them and
are clamped where the URL is built, so no caller can exceed one.

Two things follow from finding it this way:

- live-e2e.mjs is added. It loads the local build and forwards only its
  API calls to a deployed origin, which is the cheapest check that can
  see a contract the two sides disagree about. It reports rather than
  asserts, because the deployment it reads is live and a chain in a
  genuine outage would fail assertions for telling the truth.
- The fault panel's own styling moved into the shared chain-page
  partial. It had stayed behind in the old component, so the dashboard
  rendered "Why Dogecoin is not ready" at browser-default heading size,
  outranking the page title.

Verified against the live origin: both dashboards render the lens, the
timeline and every panel with no failure notice, over real blocks and
real pool attribution.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

The comment in this job predicted this failure and named the fix. The
deep pass is routes times states times themes times widths, it was
measured at 47 minutes over seven routes on 2026-08-29, and it was
measured again today at 1h11m on run 33323262662. The job cap is 100
minutes. On develop it reached the cap: run 33320958080 was killed at
1h41m with the deep pass still running, so the required check for the
promotion pull request never produced a verdict at all, and the same
happened to its pull request run beside it.

Three shards, split by route because the cost is per route, sized from
the measurement rather than by counting names: graphs and mining, then
protocols and home, then blocks, transactions and addresses with the
chain failure states. About 20, 20 and 34 minutes of a 60 minute cap.
Each builds its own copy, which is what the mobile gate beside it already
does and for the same reason: the gates cannot make each other wait, and
each one can be read and run on its own.

Nothing is dropped. Every route, state, theme and width the pass covered
is still covered, and the broad pass, the forced-colour pass and the
bundle gates stay in the build job, which now fits in 45 minutes.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

Escaping only the pipe leaves a value ending in a backslash able to
consume its own escape and split the row, so one cell reads as two. The
backslash goes first and a newline becomes a space, because a newline
ends the row outright. CodeQL flagged it as js/incomplete-sanitization on
the pull request that introduced it.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

Ask each chain for a pending set it will actually return

Run 33330398257 passed every gate in the frontend job and then reported a
failure:

  ##[error]Failed to CreateArtifact: Artifact storage quota has been hit.
  Unable to upload any new artifacts.

Nothing was wrong with the build. The deep pass uploads about fifty
megabytes of screenshots per run and the broad pass another seven, on
every run, green or not, and that is what filled the quota. Usage is
recalculated every six to twelve hours, so deleting faster does not fix
it; producing less does. 1.1 GB of artifacts from before 17:00Z were
deleted to clear the immediate block.

The screenshots are diagnostic. When a gate passes, nobody opens them;
when one fails, they are exactly what is needed. Both screenshot uploads
now run on failure, with `if-no-files-found: ignore` so a job that failed
before the pass ran does not fail twice for it.

Retention was already capped at three days by the repository, which every
run said in a warning while the file asked for thirty. It now asks for
what it can have.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

WebKit and Firefox was the one required check in this workflow that asked
for a GitHub-hosted runner. On 2026-08-30 the organization's plan stopped
providing them, and from about 19:39Z every run of this job failed while
the four self-hosted checks beside it kept passing. A required gate no
runner can accept blocks every merge in the repository, which is the same
failure as a gate that cannot finish.

It moves to `browser-heavy`, which already carries a browser and the
system libraries one needs. `playwright install --with-deps` wants root
and a runner account may not have it, so a refused dependency install
falls back to fetching the engines alone rather than failing a gate over
a permission.

Nothing about what is measured changes: the same WebKit window set and
the same Firefox subset, on the same routes.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

GitHub-hosted labels bypass the self-hosted fleet and the RunsOn AWS
overflow, and they are what consumed the organization's Actions minute
allowance. The official RunsOn images track the GitHub runner images closely,
so the toolchain is unchanged.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

Run 33333310466 passed every measurement in "Mobile and adaptive layout"
and then failed on "Keep the mobile report". The Actions artifact quota
is full, so any upload attempt fails the step, and a required check that
is green in substance reports red.

The two screenshot uploads were already narrowed to failure. These are
the three that were left: the mobile report, the engine reports, and the
chain page smoke. Same rule, same reason. On failure the upload may still
be refused, but the job is failing already, so nothing changes about the
verdict.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

# Conflicts:
#	.github/workflows/universe-ci.yml

The conflict in universe-ci.yml was one line, the runner target for the
third-engine job, and resolving it by taking the whole file from develop
threw away everything else on this branch: the sharded visual matrix, the
artifact policy, and the step that runs the new roster gate. The run that
followed proved it, failing on a step named for a script this branch
deletes.

The file is this branch again, with develop's RunsOn target for
mobile-engines kept.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

backend-apis main 100c0ec0 is live. It promotes op_inscriptions from
BLOCKED to VERIFIED READ ONLY, which production supports:
/api/v1/universe/sources reports index-opinscriptions ready, zero blocks
behind, zero consecutive failures.

So the pin is re-recorded from the deployed origin: seven protocols
readable rather than six, and the sourceSha is the commit production is
running. --against now agrees with what is served.

The README count test derives its numbers from the pin instead of
spelling them out, so the next roster change is a regenerate rather than
an edit to a test.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

…30-181216

Hold the protocol roster to the registry that owns it

The transaction reading now carries each input's and output's asset
inventory: protocol chips with the asset's own name (the id when no name
exists), exact quantities shifted by the asset's own decimals, links to
the protocol detail routes, and a named coverage state whenever emptiness
was not proven. Protocol actions stop being bare labels: they carry the
asset, the exact quantity, and the outpoints the action touched.

Address pages gain an asset holdings panel fed by the new holdings
endpoint: aggregates across unspent outputs, address-level ledger
balances in their own clearly explained section so a quantity is never
counted twice, an assets column on the unspent output table, the privacy
boundary verbatim for shielded Zcash addresses, and honest notes when a
reading is partial, unscanned, or out of an authority's coverage.

Protocol copy that called Doginals and DRC-20 another chain's business is
corrected, and the six chain protocols that had no copy at all get real
descriptions. The visual matrix fixtures exercise a multi-asset output, a
proven-empty output, an unscanned output, and both holdings views.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

…-20260830-2110

Show every provable protocol asset on chain transaction and address pages