Skip to content

blog: SAR ADC netlist to FPGA entry (Closes #13210) - #13211

Merged
gHashTag merged 7 commits into
masterfrom
claude/netlist-work-architecture-419804
Nov 7, 2026
Merged

gHashTag merged 7 commits into
masterfrom
claude/netlist-work-architecture-419804

Conversation

@gHashTag

@gHashTag gHashTag commented Nov 7, 2026 •

Copy link
Copy Markdown
Collaborator

What

specs/blog/sar_adc_netlist.t27 (one new file, 242 lines): the blog entry "A SAR ADC from its SPICE netlist to an FPGA: what the circuit is and what was measured", slug sar-adc-netlist-to-fpga, dated 2026-11-07.

It says what the circuit is (the 11-bit SAR ADC of the uberWAVE golden deck, IHP SG13G2, PSP103, 16 subcircuits / 1265 MOS / 128 cap_cmim), what problem it solves (binary search over a capacitor DAC, 31 conversions in the deck's 16 us), what the t27 pipeline builds from it, what was measured, and what is not claimed.

Where every number comes from

  • docs/analog/v0.1-release-notes.md (netlist counts, table errors, comparator 613.07 ps vs 602.54 ps, board step 17.0 to 19.6 us, converter 30/31 codes with 2 fitted numbers, comparator unit 59.46 MHz)
  • docs/analog/accept-sar-2026-10-20.md and specs/analog/accept_sar.t27 (13 blocks: 9 PASS / 0 FAIL / 4 NOT MODELLED; 30 rows: 26 / 0 / 4)
  • docs/analog/board-psp103-inv-buildb-2026-10-24.md and docs/analog/board-psp103-inv-2026-10-20.md (inverter receipts 1217 / 1216 / 1218 / 1219 / 1217 on tt / ss / ff / fs / sf; ss to sf one run each)

The tests tie the headline constants to the arrays they summarise: the acceptance counts, no block with more open rows than rows, every corner receipt line a pass, the +1.75 % comparator figure recomputed from its two times, the board step slower than real time, the comparator unit below the 100 MHz board clock, 31 conversions from the deck timing, and the converter not end to end.

No speed, power, noise or chip claim. The deck URL is not repeated here; specs/analog/sar_adc.t27 carries it with its hash.

Checks

  • t27c check specs/blog/sar_adc_netlist.t27: Typecheck OK, 0 errors, 0 warnings.
  • The spec's tests were NOT run locally: builds and tests do not run on this Mac, and the T27 OS pane table was full (16/16) with no pane of this session. CI's suite runs them; read that result, not this line.
  • No seal: seals come from a master-built t27c, and seal-coverage only checks existing seals.
  • ASCII only; lefthook foreign-budget, own-language and secret-gate passed on commit and push.
  • The two master-red fixes this PR carried earlier (specs/tri/t27b/timeout_terminal.t27, docs/igla-coder-kernels-literature.md) are reverted on the branch: ci: clear three master reds from #13179 and #13175 (Refs #13179) #13225 landed the same fixes on master first. The diff against master is now the blog spec alone, so the path-filtered loop-tools-tracked gate no longer runs on this PR; its master red stays filed as t27b ledger test counts pass_ct as non-pass, loop-tools-tracked red on master #13213.

Not in this PR

  • The TypeScript body (t27c gen-ts) and the trinity blog registration, and the Russian body in trinity apps/website/i18n/blog.ru.json: a trinity PR after this lands.
  • Publication on t27.ai or X: only after the owner's APPROVED line.

Closes #13210
see #10486

🤖 Generated with Claude Code

specs/blog/sar_adc_netlist.t27: the blog entry about the SAR ADC of
the uberWAVE golden deck read into .t27 and run on the AX7203. Every
number comes from docs/analog (v0.1 release notes, the acceptance
report of 2026-10-20, the corner board receipts of 2026-10-24); the
tests tie the headline counts, the comparator percentage, the board
step and the conversion count to the arrays they summarise. No speed,
power or chip claim. The Russian body goes to the trinity bundle.

Refs #10486

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@github-actions

github-actions Bot commented Nov 7, 2026

Copy link
Copy Markdown
Contributor

PR Dashboard

Generated at: 2026-11-07 05:25:17 UTC

Summary

Status Count
Total Open PRs 36
PRs with Failing Checks 33
PRs with All Checks Green 3
READY 2
FAILING 33
PENDING 0
NO CHECKS YET 0

These columns do not partition: 2 + 33 + 0 + 0 = 35, and there are 36 open PRs. A PR is being counted twice or not at all.

Seal Status

  • ⚠️ STALE -- sha256(compiler.rs)=e75b498e5e79 != manifest seal=87e5cbd3ad94.
    The committed NMSE numbers were certified against an older compiler.rs.
    Run scripts/reseal-check.sh locally for the two-step reseal command (advisory; not a merge gate).

…(Refs #13210)

Master has been red on duplicate-bodies and secret-scan `scan` since
f451395 (#13179), and every open PR inherits both reds:

- the private `same` helper was a twelfth copy of the bytes_eq body
  (ledger 11,3). It is removed; the spec imports std::mem::bytes_eq,
  as specs/t27f/live_preview.t27 already does. tools/dupe_scan.py now
  passes (1802 duplicate bodies, was 1803) with no ledger change.
- line 41 named the toolchain by a developer home path, which the scan
  job refuses; it now reads <repo root>/target/release/t27c.

No behavior change: bytes_eq is the same body. Typecheck OK; the spec
has no seal. Carried in the blog PR because both reds block it too.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@github-actions

github-actions Bot commented Nov 7, 2026

Copy link
Copy Markdown
Contributor

PR Dashboard

Generated at: 2026-11-07 05:31:27 UTC

Summary

Status Count
Total Open PRs 36
PRs with Failing Checks 33
PRs with All Checks Green 3
READY 2
FAILING 33
PENDING 0
NO CHECKS YET 0

These columns do not partition: 2 + 33 + 0 + 0 = 35, and there are 36 open PRs. A PR is being counted twice or not at all.

Seal Status

  • ⚠️ STALE -- sha256(compiler.rs)=e75b498e5e79 != manifest seal=87e5cbd3ad94.
    The committed NMSE numbers were certified against an older compiler.rs.
    Run scripts/reseal-check.sh locally for the two-step reseal command (advisory; not a merge gate).

Master has been red on "Documented t27c subcommands exist" since
15e3276 (#13175): docs/igla-coder-kernels-literature.md names
`tri papers` three times as a command, and no `tri papers` resolves
on any surface the gate checks, so the dead `tri` mentions rose from
the recorded ceiling of 68 to 71. Each of the three mentions now says
"not built yet". The local gate run is back at exactly 68; no tool or
ceiling changed. Prose only.

Carried in the blog PR because this red blocks every open PR.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@github-actions

github-actions Bot commented Nov 7, 2026

Copy link
Copy Markdown
Contributor

PR Dashboard

Generated at: 2026-11-07 05:34:51 UTC

Summary

Status Count
Total Open PRs 36
PRs with Failing Checks 34
PRs with All Checks Green 2
READY 2
FAILING 34
PENDING 0
NO CHECKS YET 0

Seal Status

  • ⚠️ STALE -- sha256(compiler.rs)=e75b498e5e79 != manifest seal=87e5cbd3ad94.
    The committed NMSE numbers were certified against an older compiler.rs.
    Run scripts/reseal-check.sh locally for the two-step reseal command (advisory; not a merge gate).

This was referenced Nov 7, 2026
gHashTag added a commit that referenced this pull request Nov 7, 2026
)

- timeout_terminal.t27: drop the local `same` body, use std::mem::bytes_eq
  (duplicate-bodies: bytes_eq copied [12, 3] back to [11, 3]).
- timeout_terminal.t27: replace the developer home path with a
  repo-relative one (scan: hardcoded developer home path).
- igla-coder-kernels-literature.md: mark the three `tri papers`
  mentions "not built yet" (documented-commands ceiling back to 68).

Same content as the two fix commits in #13211, applied unchanged so the
later merge of that PR is clean on these hunks.

Co-authored-by: Claude Sonnet 5.5 <noreply@anthropic.com>
github-actions Bot pushed a commit that referenced this pull request Nov 7, 2026
…er reds (#13257)

* ci: clear three master reds from #13179 and #13175 (Refs #13179)

- timeout_terminal.t27: drop the local `same` body, use std::mem::bytes_eq
  (duplicate-bodies: bytes_eq copied [12, 3] back to [11, 3]).
- timeout_terminal.t27: replace the developer home path with a
  repo-relative one (scan: hardcoded developer home path).
- igla-coder-kernels-literature.md: mark the three `tri papers`
  mentions "not built yet" (documented-commands ceiling back to 68).

Same content as the two fix commits in #13211, applied unchanged so the
later merge of that PR is clean on these hunks.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>

* ledger: restore the five pass_ct blocker fields dropped by c15eabe

The bless in c15eabe wrote the five pass_ct entries back with an empty
blocker, undoing 5c74c06. loop-tools-gate asserts that every non-pass
entry has a blocker and a known reason, so it is red on master since
f451395 and on every PR touching specs/tri/t27b. The five strings are
restored byte for byte from 5c74c06; nothing else changes.

Refs #13184

* feat(t27os): Telegram account link contract

specs/t27os/telegram_link.t27 states when a verified Telegram account may be
bound to the current local profile: one profile, one fresh attempt, a 300 s
deadline, a result taken once while pending, an explicit confirm that checks
the attempt again, issuer and client binding, no automatic replacement of a
linked subject, and an explicit unlink. Every end of an attempt clears its
handle, so a late result is refused. 11 tests, sealed. No host call, UI,
provider key or profile file format change.

Closes #13255
Refs dmitrii-f-t27#17

* gen(t27b): regenerate gen/c/tri/t27b/steward.c from steward.t27

steward.t27 counts pass_ct as a pass since #12792, but steward.c was not
regenerated, so the loop-tools gate check that gen/c/tri/t27b/steward.c is
t27c gen-c specs/tri/t27b/steward.t27 byte for byte fails. On master it
stayed hidden because the ledger step before it failed first. Regenerated
with t27c built from e96c497; the diff is the three lines of the pass_ct
branch.

Refs #13184

* fix(t27os): name a linked account by issuer, client and subject (FR-002)

A stored link was compared by subject alone. FR-002 names an account by
issuer, configured client and verified subject together, and OIDC only
makes sub unique within one issuer (one client with pairwise subjects),
so the same subject under another issuer or client is another account:
link_decision and link_confirm_candidate now take link_cfg for the stored
link and return LINK_CONFLICT when it does not match. The comment no
longer overstates the OIDC rule for sub, and the subject's character
edges (33, 126, 32, 127) are tested. 12 tests, resealed.

Refs #13255

---------

Co-authored-by: Dmitrii Vasilev <admin@t27.ai>
Co-authored-by: Claude Sonnet 5.5 <noreply@anthropic.com>
@gHashTag gHashTag reopened this Nov 7, 2026
@github-actions

github-actions Bot commented Nov 7, 2026

Copy link
Copy Markdown
Contributor

PR Dashboard

Generated at: 2026-11-07 17:29:52 UTC

Summary

Status Count
Total Open PRs 43
PRs with Failing Checks 39
PRs with All Checks Green 4
READY 2
FAILING 39
PENDING 0
NO CHECKS YET 0

These columns do not partition: 2 + 39 + 0 + 0 = 41, and there are 43 open PRs. A PR is being counted twice or not at all.

Seal Status

  • ⚠️ STALE -- sha256(compiler.rs)=b97adc88eb30 != manifest seal=87e5cbd3ad94.
    The committed NMSE numbers were certified against an older compiler.rs.
    Run scripts/reseal-check.sh locally for the two-step reseal command (advisory; not a merge gate).

@gHashTag
gHashTag merged commit a27e196 into master Nov 7, 2026
45 of 50 checks passed
@gHashTag
gHashTag deleted the claude/netlist-work-architecture-419804 branch November 7, 2026 17:30
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Blog entry: the SAR ADC netlist read into .t27 and run on the AX7203

1 participant