TASK-rectification-in-chat-step1-20261006, direct execution.
- The homepage 生时校正 pill and the text trigger (isRectificationHandoffQuestion)
are gone. The consult agent decides (RECTIFICATION_OFFER_RULE, one
definition) and calls offer-birth-time-rectification; the server decides
whether a card may show (self, product on, no case in progress,
feature_pricing price). The offer is stored with the answer and travels on
run.completed; the card hangs under that answer.
- Opening from the card binds the offer to the case in the consultation's own
messages (no table change); GET/POST /api/rectification/cases/[caseId]/source
find the source and go back: one divider (time changed or not), earlier
answers captioned 按校正前的时间, and a 按新时间重新看 button that sends only
when clicked. Auto-return after the post-adopt check or a terminal case, only
for work done in this view.
- Model history skips the divider and notes answers written with the old time.
Full suite 4987 / fail 24, identical to 6ec8efd5 (4973 / 24), no test name
lost; tsc 0; lint 0 errors; build keeps / Static, first-load gzip +0.20%.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017eEAG8HD3mm8gsKXgk8uU8
About 30 fixed rectification lines lose the receipt tone: no 「已记录。」
openings (收到 when the range holds, 好 when it moves), 麻烦 instead of 请,
时间 instead of 候选 in what the user reads, and no 「新建对话」 after
adoption. Numbers, clocks, ranges and the honesty boundaries are unchanged.
CHOICE_ACK_RE also recognises the new acknowledgements, so a model sentence
that repeats one is still dropped; the delivery guard matches the delivery
sentence by containment and does not depend on the opening.
Full suite 4973 / fail 24, identical to 61d6f248; build keeps / Static.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017eEAG8HD3mm8gsKXgk8uU8
Both rectification prompts now share RECTIFICATION_VOICE_HEAD, the same warm
voice as the consultation answer (no mind-reading here: these turns collect
facts). Opening and read-only turns used to send the whole bound Skill (11,812
characters for 10.0.33); they now send the sections they act on, plus the
product-identity section when the bound Skill has it. 9.0.0 still goes out
whole (BUG-621). Rule bodies, scoring, cards and adoption are unchanged.
Opening input 7,433 -> 3,892 tokens in a DeepSeek replay. Full suite 4970 /
fail 24, identical to c76fdda7; build keeps / Static.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017eEAG8HD3mm8gsKXgk8uU8
The method block bound into every chat turn quoted most of SKILL.md (benchmark
scores, CLI flags, oracle queues, file paths) plus the shared method, whose
output contract asks for JSON verdicts, A/B/C/D confidence, audit tables, raw
data and web verification - the opposite of the chat shape. Chat now quotes
only the truth-boundary sections and adds CHAT_METHOD_BOUNDARIES for the
limits from the dropped sections it still has to keep. SKILL.md and the shared
method file are unchanged for reports and skill_read.
Natal system prompt 55,242 -> 26,591 characters. DeepSeek A/B on public
golden charts (flash 10 questions): input 24,560 -> 14,025 tokens, 37.8 ->
30.3 s, no regression in automatic checks or reading; a high-rigor request
still says plainly that no external check was done. Full suite 4970 / fail
24, identical to b8385adb; build keeps / Static.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017eEAG8HD3mm8gsKXgk8uU8
Product 2026-10-07: the answer voice becomes a knowing, gentle older sister,
matching the homepage greeting, and may read feelings: after the answer, one
sentence, with 可能 / 多半, never a guess stated as fact or a question
rewritten into a need; another person's feelings need chart evidence and only
on an unafflicted line (afflicted lines keep the range rule, BUG-1182).
Overturns the feelings part of BUG-1070.
Also removes rules that contradicted the answer shape: 'restate the same
constraint', the rectification 'say progress' line, 'answer concisely', and
'every time say what to do' (now only when things are hard).
Full suite 4969 / fail 24, identical to 52d14e8f; tsc 0; lint 0 errors;
build keeps / Static, first-load gzip unchanged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017eEAG8HD3mm8gsKXgk8uU8
The slot wait was a fixed 5 minutes from the moment a file started queueing.
With 46 fixture files sharing 2 slots the queue outlasted that, so whichever
file came last failed the gate at random (runs 1742 and 1746). The deadline
now restarts whenever a slot changes hands; only 10 minutes with no slot
changing hands (a hung holder) fails. The 2-slot cap from BUG-281 stays.
Full suite 4967 / fail 24, identical to 171a3f25; tsc 0; lint 0 errors.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017eEAG8HD3mm8gsKXgk8uU8
BUG-1252: answer prose drops text-wrap: pretty. WebKit (every iPhone browser)
applies it to the whole paragraph, so lines came out evenly short with a blank
strip on the right and every streamed character re-broke the lines above.
BUG-1253: the condensed checklists and shared reading open with an
analysis-only note; PLAIN_SPEECH_RULE (one definition) tells the writer not to
quote layer names, signal counts, rules or house-counting, caps evidence in
brackets at two, and gives everyday words for card fields with no Chinese name
(no coined names like 合力星). Marriage layers become everyday wording; career,
wealth and health say not to announce layers.
Full suite 4966 / fail 24, identical to baseline d2c22ca2 (4962 / 24); build
keeps / Static; first-load gzip unchanged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017eEAG8HD3mm8gsKXgk8uU8
The answered choice persists the acknowledgement, a blank line, and the
rewritten delivery. The exit fill-in required equality and wrote the
deterministic delivery again. Check containment instead, and pin it on the
real answer path. The suspected widen staleness regression did not hold;
keep a guard test.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01N4f2nya58RoRu4yEmJgRGE
Ordinary consult first turns drop Markdown headings and the required
action list. People are named in the paragraph, and pinned examples no
longer contain copyable advice. BUG-1244, BUG-1245.
N1, N2, N3, and N5 do not pass. Switching ayanamsa moves level-5 lords and
equal-division D150 past the 20% wording line. Two full runs are byte-identical.
No production scoring change. closed_by_design.
Five-level Vimshottari and equal-division D150 helpers, plus a runner that
refuses to score until the preregistration file exists. First three lords
match the production chain on the v5 events. No production files changed.
Consult turns reuse a same-day official snapshot inside the rectification gate, or record deferred_in_consultation instead of starting another 4-second snapshot. Classification, step 0, step 1, and the time to the first answer character go into the existing usage metadata and a new admin usage column.
BUG-1231, BUG-1232
Round-7 option A is text-first, so Scorpio and Aquarius lords, the Ketu-only degree reversal, and node exaltation years now live in the Rath display layer. Scoring files stay unchanged.
Co-authored-by: Cursor <cursoragent@cursor.com>
The natal Sun target came from compute_chart_data while the solver iterates
on _get_sun_lon_jd (swe sidereal flags); the frames differ by up to ~17",
moving the return by up to ~7 minutes. The annual chart was also cast at the
truncated minute. Ports upstream yinduzhanxing 7cc6425d / 46ed9399 (these two
hunks only). PyJHora reference: 3 cases within 0.03 min. 243 public-figure
annual charts: annual Lagna changes 13, year lord 11, Mudda first lord 0.
v5 77 cases identical to 3c27c38e; consult goldens re-captured byte-identical.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017eEAG8HD3mm8gsKXgk8uU8
- Report p33 table: no Kranti column (no producer; materials.kranti stays blocked).
- Report p53 table: the "SAV" column was the with-Lagna total (386). Split in the
reader into SAV / Lagna BAV / SAV + Lagna with a totals row; ashtakavarga.py is
untouched (frozen rectification identity, ERR-115).
- /api/chart ascendant carries nakshatra / pada / lord (_nakshatra_fields shared with
the nine bodies); chart cache schema 4 -> 5.
- /api/varga_full bodies carry dignity_level from varga_dignity_level, the function
the Vimsopaka path now shares (A/B identical); the chart page shows the D1 status
wording and no longer derives a three-level dignity itself.
- Report page draws each chart once: the trailing block skips charts the body already
fenced (same rule as the export); the birth-time caveat moves under the body's
chart heading.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017eEAG8HD3mm8gsKXgk8uU8
Progress lists the delivery per item, F3 locations, the ERR-110 re-freeze
without a version bump (scores unchanged), the v5 evaluation (byte-identical
at a18f8d4e, bc9b9f89 and 0176cb30), upstream commits after 0a6696c5 not
taken and why, the Kemadruma difference from upstream, and open questions.
Backtest round 6 (86 answers read and scored): parents / requests / banned
phrases unchanged; career type conflict 5 -> 9 / 18, still failing the line.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01N4f2nya58RoRu4yEmJgRGE
