#Rate objections: from the PDF to policy ids

Source: docs/source/ct-rate-objection.pdf (text in docs/source/ct-rate-objection.txt), Stevan's training playbook. Method: never defend the rate; change what is being compared.

Approval. Stevan approved every quoted rep line in the PDF as final wording on 2026-09-25: the "Say" lines, the three no-quote beats, the four qualifying questions and the closing quote. They are in policy/src/playbook-*.json as status: "approved", approved_by: "stevan@currencytransfer.com", approved_at: "2026-09-25", source: "ct-rate-objection.pdf", locked: true (verbatim; never rewritten). The word "Say" is a label the UI shows once on the block, never part of a line.

#Where each part of the PDF lives

Situations sit under a move's situations array. Ids are unique within a playbook. Point text_ids are <situation>.<n> and are stable. core_text_id names the line that shows the situation was handled.

Each situation carries headline (the client's words), what / not_for, hint (the PDF's delivery guidance), points, do_not (the PDF's Do-not text, verbatim) and do_not_flags. All five rate moves above are rewrite_allowed: false.

#Questions (all in policy/src/shared.json, routed in both rules files)

Rewrite verifiers (policy/src/verify.json, threshold 0.30 in both rules files): improves_rate_or_offers_discount (any price move, even conditional: only an approved line may offer one) and explains_spread.

#Risk flags (the do-nots) and their live notes

Live notes are drafts for Stevan (risk_flags[].live_note, the field DESIGN.md names).

#Checklist and gating

• rate_pressure_handled ("Rate pushback handled"), both scenarios, applies: {when_fact: "rate_pressure_raised"}, so it is n/a on calls with no rate pushback. M1 ticks it from the same-id noul on a rep turn. A latched do-not flag shows beside it but cannot untick it, because nothing un-persists. M2 rule (met_when in the rules file): rep_qualified_the_quote, or the core_text_id of the situation rate_objection_type picked, was said, and none of the five rate flags latched.
• rate_objection_type gating. Read it only while the open concern type is rate, fees ("price") or alternative_provider (the signal's when_concern). The Choice also answers none by itself when there is no rate pushback. concern.ts reads the first concern_type signal, so the entry must stay after the scenario Choice. The rate question group (COPILOT-081) asks the rate questions only while a rate-type concern is open, and the engine's rate episode (src/engine/rate.ts, COPILOT-086) follows the situation from this Choice: it opens one decision after the concern, re-types in place and closes with the concern. Show qualify_the_quote first when the situation has qualify_first: true, then the situation's points. Show concession_terms when a price move is on the table (SessionState.rate.concession: a rep turn in the episode with rep_obtained_something_for_concession or rep_improved_rate_unasked ≥ signal_on).

#Lines (verbatim; approved 2026-09-25 unless marked)

The same ids appear in both playbooks with identical text. That is 15 approved and 4 draft points per playbook.

#Drafted lines retired (status: "retired", each with a note naming the rule broken)

Onboarding: show_the_rate_against_mid_market.1, .2, .3; handle_rate_objection.1, .2, .3; handle_alternative_provider.1. CS: handle_rate_complaint.1, .2, .3. Kept as draft, because they fit the framework: handle_alternative_provider.2 (an open question) and .3 ("compare the settled amounts"). A retired line is still refused by the loader like any unapproved line. Delete it once you agree.

#Needs Stevan's ruling

1. said_rate_transparency vs the PDF. The onboarding checklist item applies on every call: mid-market is shown next to your rate, there is no transfer fee, and the margin is inside the rate. The PDF says never explain the spread and never make mid-market the benchmark. I retired all three lines that deliver the item. Do you want to drop or reword it, or keep it for neutral fee questions only (and un-retire .3)?
2. Core lines. Are the core_text_ids in the table above right? The biggest choice is no_quote_in_hand.2 ("No. That's the price.") over .1.
3. qualify_first. The PDF sets it only for a named competitor. Should bank_matched (a matched rate is also a competing number) be qualified first too?
4. Headlines. The mid-market headline drops the PDF's numbers, because a concrete rate is refused in approved text. The bank_matched headline is 9 words, against DESIGN.md's 8-word cap.
5. Concession terms 1-4 are descriptions, not quoted lines, so they are still draft. Approve them as points?
6. Alerts. Does a market order count as "forward cover"? Only the CS forward move is linked for now. offer_alert_or_market_order.2 ("What level would you be happy at?") is shown during CS rate concerns, and it could sound like "name your price".
7. Judgement retirements. Check handle_alternative_provider.1 ("fine for small amounts") and show_the_rate_against_mid_market.2 (a bank quote against mid-market).
8. rep_improved_rate_unasked is read as "moved on price without getting something back", not "moved on price without the client asking". Is that the right reading?
9. Authority. The PDF does not say who may move on price, or by how much ("I'll price it properly").

#Stevan's rulings (2026-09-25)

1. Rate transparency stays a must-say, narrowed to "the rate you see is the rate you get, no transfer fee"; the mid-market correction line handles comparisons; no line may explain how the spread is built.
2. The four concession terms are approved as points.
3. "My bank has matched you" runs without the qualifying sub-checklist.
4. Forward cover means a forward contract only; the alert line is never shown while a rate concern is open (blocked_by: rate_pressure_raised on the alert moves).
5. No guidance yet on who may move on price or by how much; the concession flag only reports a concession that obtained nothing. To be revisited.

#Live probe findings (COPILOT-086, 2026-09-25, jev-1.13.0)

After the rep quote "Today's rate on your 20,000 is 1.2690.", each of the five client lines picked its own situation with confidence ≥ 0.95. For the red-pen:

1. Two lines do not open a concern on their own. client_objecting scored 0.41 on "I'll just check what the others are doing." and 0.17 on "I spoke to my bank and they've matched you.", both under concern_open (0.60). The rate question group is asked only while a rate-type concern is open, so on these two lines the episode waits for a turn that does open one. The other three scored 0.84 to 0.96.
2. rep_improved_rate_unasked fires on any price move. "Let me see if I can take a few pips off that for you." scored 0.86 after the quote alone and 0.80 after "Can you do better than that?" (the client did ask). "No. That's the price." scored 0.02. So Jev reads the flag as any price move, asked for or not, and without checking whether something was obtained (question 8 above asks for the second reading). The engine treats it as a price move on the table (concession).

#L1 rate cases (COPILOT-092, 2026-09-25, jev-1.13.0)

eval/labelled/rate.json holds 25 cases (13 onboarding, 12 CS). node scripts/probe-questions.mjs --cases eval/labelled/rate.json passed 24/25 cases and 46/47 checks. For the red-pen:

1. rep_rubbished_competitor reads the mid-market correction as a put-down. The approved line mid_market_comparison.1 ("That's the mid-market rate. … Nobody trades at it, banks included.") scored 0.50, above the 0.40 negative bar. Criteria repair: add "explains that the mid-market rate is a reference number nobody trades at" to the false criterion, as rep_explained_spread already does.
2. Near the line: competitor_named.1 ("They probably do on the first one. …") scored 0.33 on rep_rubbished_competitor. It passes, but the margin is thin. The false criterion already quotes it.
3. The two client lines that open no concern on their own (serial shopper, bank matched; see above) still pick their situation (0.96 to 0.98) and their client noul (0.85 to 0.94) when asked directly.