The page called "All leads" dropped every closed lead from its query, so
a lead that was lost, declined or onboarded vanished from the one place
somebody goes when they cannot find something. That is the page whose
entire job is to answer "where did it go?".
Closed leads are shown now, dimmed, with the stage they ended in. A
"Hide N closed" toggle is offered only when there is something to hide
and is OFF by default — hiding is a choice, not the behaviour.
The filter also moved off the query and onto the render. A view that
throws rows away when it fetches them cannot be un-filtered by a toggle,
and the header can now say how many are closed rather than silently
showing fewer than the count claims.
Dimmed rather than struck through or greyed to unreadability, and hover
restores it: the row is muted, not disabled — it still opens, and the
lead file behind a lost lead is often exactly what someone wants.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Recording is now on for Meera (93) and the transcript is kept on the lead
(94), so the call stops being a thing only one AI ever saw.
The wake that ends a call already carried far more than Engage used — a
summary, a sentiment analysis, and the FULL TRANSCRIPT, turn by turn —
and all of it was dropped once the three-sentence write-up was done. So a
customer said "I think it's 12,000 rupees" and "lower premium", and the
file held a paraphrase.
That matters here more than it would elsewhere: consent_artefact =
verbal_call is written from what the agent HEARD, and it is the artefact
justifying every later contact. The evidence for it should be the
customer's own words.
"Hear the call" now sits on the Log Contact entry and opens the
conversation, laid out like the WhatsApp thread — agent right, caller left
— because it is the same kind of thing and should not need learning
twice. It also files under Contact in the lead record.
Turns are split on the speaker labels rather than on newlines: a single
turn wraps, and splitting per line shattered one sentence into four
bubbles. A transcript that will not parse is shown raw — it is still
evidence, and "nothing to display" would be a lie.
NOT INCLUDED: the audio. Recording is enabled and the WAV is uploaded,
but recording_url is delivered to the agent's post-call WEBHOOK
(voice_internal_handler.go:429), not in the wake payload, so nothing puts
it on the lead. Wiring that webhook to an external-API activity is a
separate change, worth doing when someone asks to listen rather than to
read.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The trail rendered twelve fields as prose and four as money. The other
hundred-odd were invisible — so Capture Motor Risk, which writes engine
capacity, fuel, year, NCB, previous insurer, policy number, IDV and the
add-ons, appeared as a heading and a timestamp. The only way to see what
an agent had decided was to open the Lead file and guess which values
came from which step.
Each entry now carries a folded "N fields written" list: every field that
step committed, by label, with the values already shown above it
(narrative, money, documents) excluded so nothing repeats. Folded because
most steps write a dozen and the trail has to stay readable as a story.
WHAT THIS IS NOT, and it matters: it is not the model's internal
reasoning, and it is not its tool calls. Both ARE recorded — ai_reasoning,
ai_confidence, ai_model and ai_tool_calls sit on the same audit row — but
GetAuditLogsForApp selects eight columns and none of them are those, so
an app JWT cannot read them at all. Surfacing them needs four columns
added to that query and to ActivityLogEntry, which is a view-service
change and an image build.
This is the next best thing and arguably the more useful half: not what
the agent thought, but what it committed, itemised against the step that
committed it.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The page said "Waiting for the customer to reply on WhatsApp" directly
above a panel saying "Accepted by the customer, over WhatsApp. KYC and
underwriting are running." Both rendered from the same instant; only one
was true.
Quote Presented has the same shape as Document Pending — the state holds
still while the situation underneath it changes completely. Before the
customer answers, the lead is genuinely with them and nothing is running.
After they accept, KYC and the underwriting screen run inside the SAME
state, because the lead does not move until underwriting clears. phaseOf
splits it, so the strip now says "Verifying KYC" and then "Screening the
risk for underwriting", and the Next step line records only the fact of
the acceptance rather than repeating what is running.
That also puts this stretch under the stall detector for the first time,
which matters: the KYC agent talked itself out of the underwriting screen
earlier today and the lead sat in Quote Presented with a passed KYC and
nothing happening. With this it would have said so after five minutes and
offered "Run the underwriting screen" — both steps ops_admin can perform,
so the recovery needs nobody's help.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The lead page told an operator an automated step "typically completes
within two minutes; this view refreshes automatically" — and went on
saying it indefinitely. Today a lead sat forty minutes under that
sentence while the model provider returned 502s.
Agents stall, and for reasons this app does not control: the provider
slows from 6s a call to 90s or drops the request, a thinking budget runs
out mid-sentence, a late webhook wakes the wrong employee. From the
console every one of those looks identical — a lead that stops — so the
screen now reports the only thing it can honestly know (nothing has
happened for N minutes), says the work so far is safe, and offers the
way out.
THE WAY OUT IS NOT A RETRY BUTTON, because the platform has none and no
agent can be woken directly. But every agent is woken BY AN ACTIVITY, so
performing that activity again wakes it again. nudgeFor() holds that
mapping, and inside Document Pending it picks by how far the chain
actually got — three agents work that state in sequence and the one to
restart is the one that did not finish.
That matters most for the Advisor. It is the only AI step nobody may
perform by hand — permitted to ai_advisor alone — so re-performing the
activity that wakes it is the ONLY route back, and it is the step that
failed twice today. Without this table an operator's only option was to
wait or to call me.
Five minutes before it says anything: an employee wake plus a slow model
is legitimately three or four minutes, and a console that cries stall on
a working lead is one nobody reads.
The clock lives in state, ticking every 20s, rather than Date.now() in
the render body — reading the wall clock while rendering is impure, and
the counter would otherwise only move when something else happened to
re-render the page.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The order was right — the audit rows are in sequence and the panel
rendered them in sequence. Two other things were wrong.
A MESSAGE THE CUSTOMER GOT WAS NOT IN THE THREAD. After "Okay I confirm
the policy go ahead" the panel showed nothing from us until their next
question, so it read as though we had ignored somebody agreeing to buy.
We had not — the confirmation went out immediately — but it was composed
inside a trigger node and never written to a field, and the console can
only show what the workflow records. 90 stores it in customer_answer, so
it appears here and in the trail. An operator reviewing an acceptance no
human took needs to see exactly what the customer was told at the moment
they said yes.
ORDERING IS NOW (timestamp, id). One submission writes three rows in the
same second — the trigger commit, its repeat, the settle — so a timestamp
alone left their order to the sort's stability, and a reply could print
before the message it answered whenever the two landed in the same
second. It had not happened yet; it was waiting to.
DE-DUPLICATION ONLY COLLAPSES AN IMMEDIATE REPEAT, not any repeat
anywhere in the thread. A customer who asks the same thing twice because
the first went unanswered has said it twice, and a thread that silently
showed it once would hide exactly the impatience an operator needs to
see. Only a same-side, same-words turn directly after its twin is the
platform talking to itself.
The footer says which sends are still absent and why: the quote (a
registered template, not text we compose) and the read receipt (which
would sit between every question and its answer).
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
TOOLS ARE VISIBLE NOW. Clicking any agent — in the trail or the "Worked
by" strip — opens a card saying what it is, when it runs, what it can
reach and what it reads. The answer is short and it is the reassuring
kind: four of the five hold no tools at all and reason from the policy
wordings, and the only tool in the roster reads a partner registry.
That answer stopped being true once, silently. Intake held a tool that
could telephone customers, its charter never mentioned it, and it used it
on every lead believing it was a duplicate check (89). Nobody could see
that from any screen. This is the screen that would have shown it — which
is the actual argument for building it, beyond a demo looking better.
Where an agent has no tools the card says so in words rather than showing
an empty section. "No tools" is the single most reassuring fact about
something that writes into an insurance file; a blank reads as missing
data.
The roster is a hand-maintained mirror of the employee config, like
STAGES and ACTIONS, and carries the same hazard: a tool added there and
not here is described wrongly, silently. Named in the file.
THE CONVERSATION WAS EATING THE TRAIL. Every WhatsApp turn was its own
entry — heading, actor, quote box — so a four-message exchange occupied
more of the trail than the entire underwriting chain, and a resend
printed the same sentence four times because the customer sent it four
times. A contiguous run now collapses into ONE entry: a WhatsApp mark, a
count, the latest line, and a link that opens the thread. A second
exchange later in the lead stays separate, because that is a different
episode in the story.
And the actor is right. The channel performs these as the system, so the
trail read "System · Customer Reply" — the opposite of what happened.
It says "the customer" now, in the channel's own green, and the rail dot
takes that colour so a scan shows where the conversation was without
reading a word.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Five of the six asks. The sixth — the whole row as the click target —
shipped in 704afe1 and is already live on both grids.
THE AUDIT TRAIL WAS FROZEN. Timeline fetched the rows once on mount and
never again, so a lead being worked by five agents in three minutes
showed the trail as it was when the page opened — on the one screen whose
job is watching work happen. The rows move to the lead page, which
already polls every 12s, and Timeline becomes presentational. One fetch,
one poll, and the conversation view reads the same rows so the two cannot
disagree.
THE ACCEPTANCE BUTTON OUTLIVED THE ACCEPTANCE. Recording an acceptance
twice is not a loop, a recovery or an alternative — it is meaningless,
and offering it invites someone to overwrite a customer's WhatsApp
acceptance with a worse record of the same event. Gone once
acceptance_ref is set, replaced by a line saying who accepted and when.
THE AGENTS HAD NO FACE. Five of them carry this workflow and the console
named them three different ways — "ai_engage" here, "Engage" there,
"Engage AI" elsewhere — so nobody could see that the thing which called
the customer and the thing which wrote the quote were one worker. One
roster now (api/agents.js): a short name, a colour and two initials each,
used wherever an agent is named. People get a disc too, in grey — a trail
where the machines are decorated and the humans are plain text reads as
though the machines are the important ones, which is backwards on a
screen built for oversight. A "Worked by" strip above the trail shows the
team at a glance.
THE CONVERSATION WAS A LOG, NOT A THREAD. Each turn sat as its own entry
among twenty others. Now one button opens it as a thread — theirs left,
ours right, oldest first — and the footer says plainly which parts it
holds: the quote, the read receipt and the acceptance confirmation are
sent by trigger nodes and never written to a field, so they are not
there. A thread that quietly omitted them would be worse than one that
admits what it is.
CONFIRMING A PREMIUM IS A SIGN-OFF, NOT A TASK. It is the only step
locked to one role and the only one that touches money — an employee that
could mark a premium received could put a customer on risk for a policy
nobody paid for. It sat fourth in a list of six queues. It gets its own
band now, shown only to the role that owns it, amber only when something
is actually waiting: a permanent alert colour on an empty queue teaches
people to stop seeing it.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
86 moved Log Payment Nudge onto a timer — booked 24h out by Request
Premium, re-booked at 48h to a cap of three — because it was firing 47
seconds after the request and recording an outcome for a conversation
nobody had had.
A scheduled activity has to carry ZERO roles or the scheduler's own
perform fails silently, so "Chase the payment" would now 403 for
everyone who pressed it. It stops being offered.
A person who actually wants to reach the customer has Reply to the
customer, which sends a WhatsApp rather than recording that one was due.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Engage can now answer a customer's question from the policy wordings
(82), so the lead file has a conversation in it rather than a single
recorded reply.
The audit trail renders both halves — "The customer said", then "We
replied" — so the exchange reads as one, which is what an operator
reviewing an acceptance nobody in the office made actually needs.
The lead file gets a Conversation group beside Proposal: the question,
the answer, and the count. Acceptance stays in Proposal, one tab away
from the words that produced it.
"Reply to the customer" is offered to ops_admin only. Engage declines
anything it cannot answer from the wordings, anything that is a
negotiation, and everything past three answers — so a person needs a way
to take over, and it has to be through the same thread or the reply
lands somewhere the lead file never sees.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Recording the acceptance was the third human touchpoint, and it was a
person in the office typing what a customer had said somewhere else.
The customer's WhatsApp reply now drives it (81), so this stage has no
step of its own.
"Record the acceptance" drops from a step to a folded recovery lever —
kept, because a customer who says yes on a CALL still needs somebody to
record it, and because a reply Engage judged ambiguous has to be
actionable by a person.
That left Quote Presented with an empty screen at exactly the stage an
operator is most likely to wonder whether something has broken, so the
waiting state now says so. Grey, and no pulse: nothing is happening, and
that is the correct state — unlike the AI strip, where a still dot would
mean something is wrong.
The reply itself surfaces in two places: the audit trail carries it as
"The customer said", and the lead file files it under Proposal beside the
acceptance it produced. The evidence and the decision it justifies are
one line apart, which matters when nobody in the office made it.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
THE SIDEBAR NOW CARRIES COUNTS. It deliberately did not, on the argument
that a tally could only come from a second full list call that would then
disagree with the queue's own total. Right about the cost, wrong about
the conclusion: with no numbers the only way to learn whether anything
was waiting on you was to open all nine queues in turn, which is the
question navigation exists to answer.
There is no second call. The overview's existing fetch moved into a
PortfolioProvider that both surfaces read, so this is one call fewer than
before and the two agree by construction. Bounded at 200 rows, as the
overview always was; past that the honest answer is an aggregate
endpoint, not a bigger limit.
What the badge counts is what needs a PERSON — phaseOf again, so a lead
whose documents are in and whose AI chain is running is reported beside
the badge rather than inside it. An amber dot marks a queue holding a
renewal inside a week, which is the only reason to open one queue before
another and was previously invisible. Zero is shown rather than hidden:
"nothing here" is an answer, and a queue that disappears when it empties
makes the sidebar move under the cursor.
Closed is folded into a summary. Three of the twelve queues, opened about
once a week, and at equal weight they made the live ones harder to find.
REASONING OPENS IN A DIALOG, NOT INLINE. The audit rail is 320px wide
with one entry per step, and a 1,500-character rationale expanding in
place pushed a lead's whole history off screen to read one sentence of
it. Nobody reads a paragraph in a sidebar. The finding — the AI's own
first sentence, which is already the conclusion — stays on the line; the
working is one click away and one Escape back, headed by what is being
read ("Call — Log Contact") rather than by nothing. Escape closes, the
page behind does not scroll, and focus returns to the button that opened
it so a keyboard reader keeps their place.
Two headings that described the container rather than the contents:
Record → "Lead file", and the audit trail's subtitle now says what a
reader gets from it.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Traced the console against the workflow it renders. Five gaps, all of
which made a working chain look like a broken one.
THE AUDIT TRAIL WAS READING THE WRONG KEYS. An audit row's data is keyed
by the ACTIVITY's field ids, which carry a per-form suffix — the call
notes arrive as contact_notes_2, the document request as
documents_notes_3 — and the timeline matched the unsuffixed global ids.
Almost nothing ever matched, so the trail was a list of activity names
with the agents' reasoning invisible behind it. The one narrative line
that did appear was an accident. It now reads the platform's own typed
`fields[]` (label, data_type, value) and matches on the base id.
That is also what lets it answer the question that was asked: an upload
now renders as "3 documents received" with each one named and openable,
instead of a bare "Collect Documents".
DATA_UPDATE ROWS WERE CLASSED AS AGENT WORK. A bookkeeping row inherits
the roles of whoever caused it, and the AI check ran first — so an AI's
field write appeared in the trail as an entry titled "Data updated",
while the toggle underneath still offered to reveal the others. The
activity id says a row is bookkeeping; the roles say who triggered it.
DOCUMENT PENDING IS TWO SITUATIONS. Before the upload a person has to
act; after it the lead stays in the same state while three AI steps run.
Both rendered as "waiting on the partner agent", so a lead that had just
been served showed an Upload documents button under a panel saying the
documents had been received, and counted against the Action-required
queue. phaseOf() derives the difference once, from documents_status, and
the header, the queue count, the row status and the action list all read
it. The upload demotes to recovery — "Replace or add a document", folded
away with the other levers.
A STOPPED CHAIN LOOKED IDENTICAL TO A RUNNING ONE. The rating engine
refuses to price without an IDV and writes so into quoted_breakup;
Engage then declines to raise a quote it would have to fabricate. Both
are right, and nobody was told: the refusal sat in a field on a tab and
the lead never moved again. blockedOn() surfaces it as an amber strip
that names the missing value and opens the form that carries it.
THE STALL MEASURE WAS DATED FROM THE WRONG COLUMN. progress() fell back
to created_at when updated_at was absent — which it always was, because
neither view returned it — so every lead older than half an hour would
have reported stalled. It reads updated_at only; the view supplies it as
of 76.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Two changes to the queue and overview tables.
The row is the click target. An "Open" button was a 90px destination inside
a 1000px row that already reads as one object, and it put a second tab stop
on every line. The row now carries the click, Enter/Space, a focus ring and
the hover tint; the button column is gone, along with its CSS.
The column it freed shows what is actually happening. This workflow is run
by agents and watched by a person, so the question a queue has to answer is
not "which stage" — inside a queue every row shares it — but "is it moving".
An automated stage now reports working (under 3 minutes), waiting (under 30),
or stalled with the elapsed time, with a pulse on the live one and a red left
edge on the stuck one. Thresholds are generous on purpose: an employee wake
takes a minute or two and a scheduled retry can be hours out, so stalled
means "longer than any normal step", not "longer than average".
Before this, a lead three hours into "Calling the customer…" looked identical
to one thirty seconds in — which is the failure this console exists to catch.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
`documents_status` was a dropdown — not yet / partially / all uploaded —
put to the person who had just attached the files. It asked them to tell
the system what it could see for itself.
Not cosmetic. That field gates the AND-join that wakes Engage after an
upload (55). Left on its default, the lead sits in Document Pending with
all three documents attached and nothing happening, and it reads as the
AI having stalled rather than as a form field nobody filled.
It is derived in the trigger now (75), from the files themselves, after
the commit and before the join reads it — server-side, because a value
computed only in the browser would be right on screen and absent to the
API. `documents_notes` goes with it: the same script writes what is
still missing, and a person overwriting that would be arguing with the
file list.
So both are declared derived and no form asks for them, with or without
a lead behind it. Matched on the base id, so every per-activity suffix
is covered rather than the two that happen to exist today.
Uploading an RC read the document correctly, said so, and filled
nothing.
The endpoint answers keyed by the ocr_config's extraction_fields[].key
— `reg_no`, `make_model`, `mfg_year`, `engine_cc`, `fuel` — because
that is what the vision prompt was asked to produce. Its own package
doc says so. The form's fields are `motor_reg_no_6`, `motor_cc_3`,
`motor_fuel_3`. The console wrote the response straight into values, so
every extracted value landed on a key no field renders: the read
succeeded, the "we read it for you" panel appeared, and the boxes
stayed empty.
This is the third place the same mismatch has bitten. Prefill hit it
(the pipeline names globals, the form uses suffixed activity keys) and
the validation errors hit it (the workflow reports the machine key).
Different surfaces, one cause: nothing in the stack translates between
a global field name and a form's per-activity id, so every consumer has
to do it and each one forgot.
The bridge was already on the field. ocr_config.field_mappings maps
extraction_key -> target_field (the global), and the form field is that
global plus a suffix. Verified against the live Collect Documents
schema: all five RC keys now resolve to real fields.
Also: which value wins. Seeding the form from the lead record (6a6801d)
made every mirrored field non-empty, so the old "only fill what is
blank" rule would have blocked OCR from writing anything at all. Values
copied off the lead are now tracked, and a document may overwrite them —
it is the better source for what it states. Anything typed by a person
is never overwritten, and stops being overwritable the moment it is
typed.
Four operator-reported issues, one commit because two of them are about
the same form.
ALL LEADS WAS GONE. The team's rewrite of the queue screen dropped the
/stage/all view, so finding a lead meant guessing which of nine queues
it sat in. Restored: no stage filter, closed leads dropped client-side,
one request for the whole open book. Linked under Overview in the
sidebar so it is one click from anywhere.
THE DOCUMENT FORM ASKED FOR WHAT THE LEAD ALREADY KNEW. Collect
Documents mirrors fourteen record fields — registration, make and model,
previous insurer, expiry, policy number, PAN — so the OCR has somewhere
to write. Rendered blank, they read as fourteen more things to type, on
a form whose whole point is uploading three files. They are now seeded
from the lead by base key: an agent uploading for KA01MF6618 sees
KA01MF6618 already there. Server prefill wins, then anything typed;
files and generated ids are never seeded, and an INIT form has no
record to seed from. SME-only fields were already hidden on a motor
lead by fieldLine — that half was working.
"SEND A REMINDER" IS THE SCHEDULER'S JOB, NOT A BUTTON. It fires at
24h, 72h and 120h on its own. Offering it as a loop action invited a
person to send a reminder the platform was already about to send. It
and Retry Call move to `force` — reachable behind "Lead not moving?"
for the day the schedule itself fails, and otherwise out of sight.
THE UPLOAD STAYED THE "NEXT STEP" AFTER THE UPLOAD. Once
documents_status is complete the lead remains in Document Pending while
Engage captures the risk and the quote is built — and the panel kept
saying "Upload documents", which reads as though nothing was received.
Now, when complete, the upload becomes a loop (a fourth document, a
correction) and the panel says what is actually happening: documents
received, Engage AI is capturing the risk and preparing the quote.
Permissions unchanged; ACTIONS still matches
tbl_wf_state_allowed_activities exactly.
The app read as a control panel. Every activity a stage allowed was
rendered as an equal button, so operating it required already knowing
the workflow — which is the opposite of what a workflow tool is for.
Two concrete failures, both visible in the data:
Ops on Document collection got SIX equal buttons. Exactly one was the
next step: upload the three documents. Capture Motor Risk, Capture SME
Risk, AI Cover Recommendation and Generate Quote are the AI's own chain,
present only so a stalled lead can be pushed by hand.
And FOUR stages offered a partner agent nothing but "Mark Lost" — New
Lead, Contacted, Underwriting referral, Premium confirmation. At each of
those the system was actively working the lead, and the only thing the
app suggested was giving up on it.
So an activity now carries what it IS at a stage, not only who may press
it:
do the step this stage is waiting on. Usually one. Referred has
two, because clear and decline are a decision pair rather than
a step and the option of not taking it.
again a bounded loop — retry, reminder, re-quote. Legitimate, never
the answer to "what now".
force an AI employee's own job, offered to a person only so a stalled
lead can be moved. Folded behind "Lead not moving?".
exit Mark Lost. Always reachable, never presented as a step.
The panel is "Next step". With a step, one button at full weight. Without
one it says what is actually happening and who has it, which is the true
answer at six of the nine working stages and was previously rendered as
an empty bar or a lone Mark Lost.
Result per role, checked rather than assumed: an underwriter sees two
buttons, on one stage, and nothing anywhere else. A partner agent sees a
step at three stages. Ops sees a step at four and can still force any
AI activity from the folded section.
Nothing about permissions changed, and the action list still matches
tbl_wf_state_allowed_activities exactly — this is only about which of
them is presented as the thing to do.
Two things on the lead page.
THE AUDIT TRAIL REPEATED ITSELF. "Qualify Lead" appeared three times in
a row and "Partner Agent Lead" twice, which reads as the AI having done
the same thing three times. It had not. One submission writes several
rows, told apart by execution_state:
zk-act-qualify TRIGGER_PERFORMED the trigger's commit
zk-act-qualify TRIGGER_PERFORMED again, on the settle path
zk-act-qualify zk-state-qualified the one that moved the lead
Only the last means anything to somebody reading the file. They are now
collapsed into one entry carrying the earliest timestamp — when the
operator acted — and whichever state and payload is populated.
Matched on (same activity, within fifteen seconds) rather than on
execution_state, so a genuine repeat survives: Collect Documents really
is performed twice on a lead whose first upload was short, and those are
minutes apart. And matched by looking BACK through recent entries rather
than at the previous one, because the platform interleaves a DATA_UPDATE
row between the two halves of a submission — the rows to merge are near
each other in time but not adjacent in the list.
THE AI PARAGRAPHS HAD NO SUMMARY. attribution_reason and
eligibility_reason run to a paragraph each and opened folded, so the
finding could not be read without expanding. ClampText now lifts the
first sentence out as a headline and puts the rest behind "Show the
reasoning" — no new field, and nothing invented: the employees already
write a conclusion and then its evidence.
It falls back to plain folding when the split would be useless — no
sentence terminator, a first sentence that is the whole paragraph, or
one short enough to be a fragment rather than a finding.
The other half of that is in the charter (70_verdict_first.sql). The
headline is only as good as the sentence, and Intake was opening with
"Tool 29601 confirmed..." — an internal id — because it had been told to
"name the evidence, not the conclusion". Right about content, wrong about
order: it now leads with what it decided and gives the evidence second.
Submitting the new-lead form with a blank Product Line produced:
400 The workflow rejected this submission.
validation failed for activity zk-act-init-agent: product_line_4(required)
That asks an operator to know what a field_s_id is, that `_4` is a
disambiguating suffix and not part of a name, that "required" is a rule
rather than a value, and that 400 means them and not us. All four are
ours to know.
Two changes, and the first means the second is rarely reached.
The form now checks required fields before it sends anything. It marks
each empty one — on the label, because somebody scanning a thirteen-field
form is looking for the NAME they missed, not for a red box — scrolls the
first into view, and says "Product Line is required". No round trip, and
nothing to decode. The marking clears as the field is filled, since a
field still flagged after being corrected teaches people to ignore the
flagging.
If the workflow does reject a submission on fields — it validates more
than this form can know about — describeValidation now renders the reply
in the form's own labels. It parses every `key(reason)` pair, maps the
reason to a sentence, and resolves the key through the schema, including
the suffix rule the prefill seeding uses. A key the schema does not carry
keeps its raw name: a wrong label is worse than an ugly one, because the
operator goes looking for a field that is not there.
The status code is dropped from that surface. "400" is the first thing
read and the least useful thing shown; it stays on the errors an operator
genuinely cannot fix by typing.
"Sourced By", "Submitted By" and "Source Channel" rendered empty on the
new-lead form. The prefill pipeline was working the whole time — the
form-screen response carries
partner_code "Shetty Insurance Services (POSP-77341)"
rm_or_agent_id "arjun.posp@zurichkotak.example"
source_channel "agency"
and the form ignored all three.
The two sides key the same field differently and nothing reconciles it.
A form field is the ACTIVITY key, suffixed because one global may be
used on several activities in a version — partner_code_3,
rm_or_agent_id_3, source_channel_4. A fieldMapping node names the
GLOBAL — partner_code, rm_or_agent_id, source_channel. Neither is
wrong. The seeding loop matched on f.id exactly, found nothing, and
seeded nothing, with no error anywhere.
Now it tries the exact id, then the field uid, then the global name with
the numeric suffix stripped. Taking all three means this survives a
pipeline authored in either convention rather than breaking again the
next time one is written the other way. Empty and null are skipped so a
mapping that resolved to nothing does not overwrite a real default.
Worth noting the failure mode rather than just the fix: a prefill that
matches nothing looks exactly like a prefill that was never configured.
It is why the pipeline read as "empty" from the config tables — the
authoritative store is tbl_wf_activity_prefill_nodes, not the
pipeline_config JSONB, and the quickest honest check is neither: POST
/view/form-screens and read prefill_data.
Four things, and the first is the one that mattered.
ROLE-BASED ACCESS. The console showed every action to everyone on
purpose — "the refusal is the demo" — and the workflow refused
server-side. That is a defensible engineering position and a poor
product: an underwriter saw a row of six buttons, five of which 403.
src/api/permissions.js now mirrors tbl_wf_activity_permissions, and the
session already carries user.roles, so nothing new had to be fetched.
Action buttons, sidebar queues and the entry doors are all filtered.
It remains presentation only — the platform still refuses, and a role
added there but missing here hides a button that would have worked,
which is the failure mode to watch.
Each role's app now looks like their job. Ops sees the whole board and
still cannot clear a referral. An underwriter sees one queue and two
buttons. Agents see the three queues they work in. Reporting is NOT
gated: the overview counts the whole portfolio for everyone, because an
agent tracking leads they filed is reasonable and acting on them is not.
One deliberate divergence from the workflow, commented where it lives:
the two scheduled activities carry no roles at all, because that is the
only configuration under which the scheduler can perform them. Open to
the SYSTEM is not open to everyone signing in, so the UI narrows them —
otherwise an underwriter is offered "Send Document Reminder".
ANALYTICS. The overview was three lists. It is now a measured page:
open leads, pipeline value, written premium, conversion, commission —
then renewal exposure in three buckets, the queues needing a person with
the age of the oldest item in each, and stage distribution as bars.
Checked against the live book rather than assumed, which found a real
bug before it shipped: Policy Issued is written business AND not yet
terminal, so its premium was counted in pipeline and production both.
₹42,912 double counted on 37 leads. Pipeline now excludes anything
already on risk.
The lead page gained the same treatment: renewal countdown, lead age,
time in current stage, premium, commission, AI confidence. Age and
dwell are computed — neither is on the record, and they are the two
figures that answer "is this moving?", which no field could.
TERMINOLOGY. The previous pass over-corrected: fixing builder jargon
("the workflow decides, server-side") produced chat ("What you can do",
"The agents have it", "Nothing here right now"). Neither is how an
operations console reads. Now: Actions · Record · Audit trail · Action
required · Customer response · Automated · Scheduled · Document
collection · Underwriting referral · Premium confirmation.
LAYOUT. Prose subtitles cut to one line or removed. Measures render as
tabular figures in a bordered strip instead of label/value pairs. The
overview splits into work on the left and distribution on the right.
Stage bars replace a wall of equal-weight cards.
Pre-existing lint errors unchanged at 11; none in the new files.
Three problems, and the first one is the reason the other two were hard
to see.
THERE WAS NO DASHBOARD. "/" redirected to the underwriting queue — one
stage, usually empty, and belonging to somebody else. Every screen in
the app answered "show me this queue"; none answered "how is the book
doing, and is anything waiting on me?", which is the only question a
person has before they have picked a queue.
So: a Today page. Three blocks, in the order somebody cares about them.
What is waiting on a person, as three cards that colour only when they
have something in them. Renewals running out, worst first, capped at 30
days because a renewal six months away is not a thing to look at today.
Then every stage with its count, zeroes included — a stage quietly
receiving nothing is only visible if its zero is on screen.
It is one list call, counted in the browser. That is honest at this size
and it is the same call the queues already make. If the book outgrows it
the answer is a counts endpoint, not a bigger page.
THE WORDS WERE WRITTEN FOR WHOEVER BUILT THE WORKFLOW. "What happens
next" was followed by "these are the only activities this state allows
— but the workflow decides, server-side, whether you may". A failed list
said "this queue has no record view yet". Panels were called "Flow
details". A closed lead read "nothing runs from here". None of that is
wrong; all of it is addressed to the wrong reader.
Now: "What you can do", "Lead details", "History", "This queue could not
be loaded". Queue names say what they want — "Documents needed",
"Underwriter to review", "Payment to confirm". The sidebar group called
"Not yet" says "Not due yet", and "Running by itself" says "The agents
have it", which is the actual claim being made.
THE QUEUE HAD NINE COLUMNS AND LED WITH THE WRONG ONE. Two were internal
vocabulary: attribution status renders "clear" or "contested" and means
nothing to an operator, and the channel is background rather than
something anyone scans a queue for. Both fold into a subtitle under the
customer. That leaves six columns and puts the renewal countdown — the
number that decides whether to act today — second instead of fifth.
Also: the timeline printed a raw slug when a step came back without a
name. It is the screen this product is demonstrated on, so a
"zk-act-doc-reminder" in the middle of an otherwise readable story is
expensive. It now reads as English.
The workflow now schedules work for itself. Two activities exist that
nobody navigates to — the scheduler performs them — and both are offered
here as well, because a person needs to be able to do by hand what the
timer does on its own.
Retry Call, in Awaiting Contact. Booked automatically four hours after an
unanswered call, clamped into 09:00-21:00 IST, three attempts and then
the lead closes. It records the attempt; whether it also DIALS is gated
on the lead's `auto_retry` field, which ships unset, so nothing rings a
phone until somebody arms it deliberately.
Send Document Reminder, in Document Pending. Fires at 24h, 72h and 120h
from Engage's request. On the third unanswered one the lead leaves the
queue — Parked if the renewal is still weeks off, Lost if the cover is
about to lapse. That is the bound: a chase with no ceiling fills a book
with leads nobody will ever work, all of them looking active.
Both are labelled "Scheduled — or you" rather than given an AI or a role,
because neither is true. The performer on the timeline will read System.
The workflow dropped four stages on 7 September and renamed two. This
table is the console's mirror of it and was still describing the old
shape, which breaks quietly rather than loudly.
WHAT CHANGED IN THE WORKFLOW
Risk Captured, Proposal Accepted, KYC Verified and Underwriting Cleared
are gone. A lead entered and left each of them inside a minute, so they
were pipeline steps wearing a stage's clothes. Meanwhile the one genuine
stall -- a partner agent finding three documents -- had no stage at all
and happened inside "Contacted", which meant the queue that most needed
watching was the only one that did not exist. zk-state-docs is that
queue, and it is where the work those four stages used to represent now
happens: upload, capture, advise, quote, all without the lead moving.
Quote Presented absorbs the other three the same way -- acceptance,
identity and the underwriting screen -- and leaves in one move for
Payment Pending, Referred or Declined, decided by a single gate that
reads both answers. The two-gate chain it replaces could not express
"identity refers but the risk is fine": the first gate routed to
Referred before appetite had been asked, so an underwriter always
received half a file.
Qualified became Awaiting Contact and Quoted became Quote Presented.
"Qualified" meant pre-call here and post-call on the whiteboard -- the
same word on opposite sides of the dial.
WHY THE RENAME IS THE DANGEROUS HALF
STAGES.name is compared against current_state_name in three places: the
queue filter, the sidebar tally, and the action lookup on the lead
detail page. A workflow rename that is not mirrored here does not error
anywhere -- the record view returns zero rows, correctly, for a name no
lead is in. Both renamed queues would simply have read empty. The header
comment now says so.
Worse, zk-state-docs was absent from STATE_ACTIVITIES entirely, so a
lead sitting there rendered no buttons at all. Document upload is the
human step the entire rest of the flow hangs off.
ALSO
Parked / Nurture has been in the workflow since 4 September and was
never added here, so its Resume Outreach button did not exist. Until a
scheduler lands, that button is the only thing that wakes a parked lead,
so it gets a nav group of its own rather than being filed under
"running by itself", which it is not.
DROPPABLE follows the new state list. Mark Lost now also reaches
Referred, which the workflow had never allowed and now does: a lead
whose cover lapses while an underwriter holds the file previously had
nowhere to go.
doc_vehicle_photos becomes conditional. The comment said no break_in
flag existed to test -- one does now, derived at filing on all three
doors, so the slot honours its own "break-in only" caveat instead of
being shown to every motor renewal.
Verified by diffing STAGES.name and ACTIONS against tbl_wf_states and
tbl_wf_state_allowed_activities on dev; both match exactly.
Issue Payment Link is now Request Premium — the platform has no payment
integration and the fabricated link field is off the form. The action's
'by' was 'Payment webhook', which was aspirational; collection happens on
the insurer's own rails and ops records the reference.
Log Payment Nudge added to Payment Pending. Engage performs it; without an
entry here the activity is invisible to the console rather than broken.
/ocr-extract answers { extracted: {...}, raw: "..." }. The client checked
`fields` first and `extracted` second, which was right by luck — but an
extraction that returned nothing, or threw, did so silently: the button
went back to idle and the operator had no way to tell a read from a
no-read.
Now an empty or failed extraction says so under the field, and either
way the UPLOAD survives. The file is already stored and referenced; a
failed read only means the fields are not pre-filled, which someone can
recover by typing. Losing the upload because the read failed would not
be recoverable.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
/upload requires it and it is fixed for this app, but the client only
appended it when ctx happened to carry it — and FileField passes only
activity, field and instance. Every upload came back 400 with
"Missing required params: workflow_uuid".
Taken from config now, so it cannot be forgotten by a caller.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Collect Documents shipped with every document field rendering as a plain
text input — ActivityForm had no case for `file` or `ocr`, so they fell
through to the default. There was a form asking you to type an RC book.
Uploads happen ON PICK rather than on submit, because /ocr-extract takes
a REFERENCE to a stored file: the document crosses the wire once, it
survives a reload (a draft cannot carry a File), and re-extracting costs
no second upload.
For an ocr field the extracted values are shown inline — "read from the
document" — and filled into the fields they map to. Only into fields that
are still blank: a value a person typed is never overwritten by one a
model read.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The activity was added to the workflow on both Contacted and Proposal
Accepted, but not to the console's action map — so the platform allowed
it, the console never offered it, and nothing errored anywhere. There was
simply no way to upload a document.
Notes the hazard on the map itself: it is hand-maintained against
tbl_wf_state_allowed_activities, and anything missing from it is
invisible rather than broken.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The timeline carried timestamps; the two screens people actually act on
carried none. A queue with no dates cannot tell you which lead is going
stale, and on a renewal book the countdown to expiry is the number that
decides whether anyone should act today — it was buried in a field group.
Queue gains "Renewal due" and "Added", both absolute plus relative: the
absolute answers "when exactly", the relative answers "is this stale".
The countdown is toned — lapsed, urgent, soon, later — on the same
thresholds the workflow itself uses to decide whether to call, so the
screen and the agent are reading the same bands.
Lead header gains the countdown, the current insurer, the registration,
and when the lead arrived and was last touched.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The lead file listed the current field values and nothing about how they
got there. On an app whose whole claim is that agents do the work, that
is the wrong thing to show: you could see the outcome and not whether a
machine or a person produced it.
Adds a timeline from the audit trail — every activity in order, badged by
actor, carrying the reasoning the agent actually wrote (attribution,
eligibility, call notes, recommendation rationale, the premium breakup)
rather than the raw field dump that is already below it.
Colour encodes something real: machine work is blue, people amber, the
platform grey, so a file reads at a glance for how much of it was done by
hand.
Also fixes the audit path — the bare /view/audit is not an API route and
falls through to the SPA, returning HTML with a 200 that parses as a JSON
error rather than an HTTP one. It has to be app-scoped.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
This demo is about the individual agent who sources a lead and gets paid
when it onboards, so the console offers that one entry point and goes
straight to its form rather than showing a chooser with one option.
The direct and bancassurance INIT activities are hidden, not removed —
both still work over the API with their own permissions, and the
bancassurance door is what makes the cross-channel duplicate story
possible.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
lead_ref is now an id_gen, issued server-side and stripped from the
submission — so the form shows it as auto rather than as an empty box
somebody is expected to type a reference code into, and never sends it.
Also seeds the inputs from the prefill the form response carries. The
door's channel and the signed-in agent are stamped by a prefill pipeline,
so the form should show them already filled rather than ask.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The pipeline could be looked at but not driven. Adds the three pieces
that make the machine walkable end to end from the UI.
- ActivityForm renders whatever /view/form-screens returns — labels,
types, select options, mandatory flags — and submits it back. No form
is defined in this repo, so a field added in Studio appears here with
no code change.
- AddLead presents the three doors. Each is a separate INIT activity with
its own permissions; the bank one lands further along because the bank
already did CKYC.
- Lead shows the file grouped in the order it was worked, ending with
what the sourcing agent earns, plus the activities this state allows
and who normally performs each.
The "who normally performs this" label is presentational only. Nothing
here enforces anything — the workflow refuses server-side and the form
reports what it said, including a note when a 403 is the platform
declining rather than the console misbehaving.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The POST body is not the GET query params with a different verb. The view
is named by rv_template_uid (rv_id is the GET spelling) and paging, sort
and filters all live inside search_query. Sending them at the top level
returns 400 "Missing param: rv_template_uid (body)".
This would have failed the moment the record view was seeded, and the
400 would have read as "the view is missing" rather than "the request
was wrong" — the queue screen renders exactly that notice on any error.
Verified against dev: the old shape returns the missing-param 400, the
new one returns "record view not found", which is the view genuinely not
existing yet.
Also adds rows(), which normalises the response key — workflow views
return `data`, rdbms views have been seen returning `records`.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The build asserts on <!--BASE_HREF--> and placement fails without it, so
nothing was ever placed and nginx 404'd. The ./config.js tag is the other
half: it carries this environment's VITE_ZINO_API_URL, and without it
requireConfigValue throws and the app renders nothing even once served.
Order matters — the placeholder precedes the favicon href, or that
relative ref resolves against the wrong root once mounted.
Verified by simulating placement locally: filled the placeholder, wrote a
config.js beside the artifact, served it under /zurich-kotak/ and loaded
it. index.html, the JS bundle, config.js and the favicon all 200, the
login renders, and with config.js removed the root stays empty rather
than falling back to some other backend.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
RBAC is what this demo argues, so which persona you are signed in as
should never be a guess while presenting.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The scaffold had the brand right and nothing behind it: the login was a
600ms setTimeout with a TODO, and three pieces of the frontgen deploy
contract were missing.
- Sign in against POST /usr/login (org_id as a STRING — a number is
rejected by the gateway) with the session in a provider; surface the
gateway's own message rather than a generic "invalid credentials",
because a wrong password and a user without access to this app look
identical from here and are not.
- Runtime config: the API URL is read from the config.js the server
writes at placement, never compiled in, and requireConfigValue throws
so a build with no config.js fails loudly instead of calling whichever
backend built it.
- base: './' plus a router basename taken from <base href>, so one build
serves any mount path.
- Move the fonts and logo from public/ into src/assets/ — Vite rewrites
bundled asset URLs to be relative, while a public/ file referenced as
"/fonts/..." stays absolute and 404s under the /zurich-kotak/ mount.
- API client for recordview / detailview / form-screens / start /
activity, and the pipeline shell whose sidebar is the state machine in
the order a lead moves.
Queue and lead-file data wait on the record and detail views.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
