Beth's two questions — the funnel, hours saved, and what customers actually said

Measured, no hedge needed — these are counts read from our production database on 2026-10-27, and ratios of those counts:

• At Camellia, 4 in 10 inquiries become a booked tour (40.6%), and a bit over 1 in 5 of those tours becomes a signed lease (22.0%).
• Clara had worked 21 of the 27 leases signed at Camellia — meaning she had replied to that person at some point in the life of their enquiry. On the stricter reading, where her first reply has to land at or before the signing date, it is 15 of the 19 leases we can date that way.
• At Western Slope, 13.9% of inquiries become a booked tour, and Clara had worked 13 of the 33 leases signed there since August — though for 3 of those 13 her first reply came after the signing, because she re-engaged the person later.
• Western Slope's tour→lease rate is not available and should not be offered. Only 3 of the 33 people who leased there ever had a tour booked with us — their leases mostly arrive already past the tour stage — so any percentage would measure how their leases enter our system, not how well they convert.

Estimated — the hedge wording is written out, use it as-is:

“That's our own estimate, not a number they've given us — we count the calls, texts, emails, tours and renewals Clara handled for them, and we price each one at what we think it would have taken their staff. On that basis it's roughly 20 hours a week at Western Slope, closer to 24 if you count the renewals a person still approves, and a little under 5 at Camellia. The counts are real production data; the minutes-per-task are our assumption, and we haven't yet done a time-and-motion study to confirm them.”

The two strongest quotes. Both verbatim from our meeting recordings; the rest, with their caveats, are in Part 2.

“No, it was a lot of help. Having the tours get scheduled and answering calls, that does help a lot.”

— on-site property manager, Camellia (JPCO), 2026-10-15

“…I know that the plan wasn't necessarily that Clara would, like, reinvent the way that we answer the phone, but it has done that for us, and it's awesome, and I think it's a major selling point for you guys…”

— principal, Western Slope, 2026-10-07

Every figure in this section is Measured: a count of records read from the production database on 2026-10-27, or a ratio of two such counts over the same group of people. Nothing here is estimated, and nothing blank has been filled in.

How the window works, because it is the thing that makes a rate honest. Each row picks the people who arrived in that period and then follows those same people as far as they got, whenever they got there. The tempting alternative — tours held this quarter over inquiries received this quarter — divides two different groups of people and produces a confident wrong number. One lead means one person, which is the unit the product itself counts. Recent arrivals always read low at the bottom of the funnel, because an October lead has not had time to sign, so both customers' rates are floors, not ceilings.

Why Camellia gets two rows and Western Slope one. Camellia has been live far longer, and a three-month window throws away half its history — so its own all-time row keeps the comparison fair to it, while the shared window is what makes the two customers comparable at all. Western Slope needs no second row: its leads begin in August, so all-time and the shared window are the same group of people but for a single February lead (246 inquiries instead of 245, 13.8% instead of 13.9%).

Why Western Slope's tour→lease cell stays blank

“Tours” means a tour was booked

That is the product's own definition, and it deliberately includes tours later cancelled, because booking is the event. We do not record whether a tour was actually attended — there is no attended, completed or no-show field anywhere on a tour, only cancelled. So tour→lease reads “of the people who booked a tour, the share who went on to sign”. There is no honest show-up-adjusted rate available and the lane did not invent one.

Leases Clara had worked — both readings, side by side

The phrasing matters more than the number. Say “Clara had worked 21 of the 27 leases signed at Camellia”. Do not say “Clara had 21 leases” or anything that sounds like she produced the paperwork: the loose rule means she was in the conversation, and the shorter phrasing is a claim we cannot support and exactly the one an investor would repeat back to us.

Why both, rather than one. The leases AppFolio confirms for us carry no signing date to order against, so the strict rule cannot be applied to every row. Applying one rule to some rows and a different rule to the rest would be worse than publishing both and saying which is which. The two denominators are different populations — 27 is every person who signed, 19 is the subset we hold a signing date for — so the two columns are not two slices of one number.

What the loose rule does and does not claim. It does claim Clara talked to this person: that is a timestamped event, not an inference from a status field. It does not claim she closed the lease alone, and it does not claim the lease would not have happened without her — a human leasing agent was very likely involved in most of them.

How “leases signed” is counted, and why one stamp was not enough

A lease is visible to us two independent ways: our own signed stamp on the lead, and the person's AppFolio record being linked to a tenant — the property management system itself saying this lead became a resident. The counts above are the distinct people carrying either signal, so nobody is counted twice. The gap is real: at Camellia 19 leads carry our stamp and 8 more are AppFolio-confirmed without one, so using our stamp alone would have understated Camellia's leases by about a third. These are people who signed, not lease documents — two roommates on one lease count once where they share a person record and twice where they do not.

Why there is no screenshot, and why the screen would not have answered this

The two lead populations, and which one “Inquiries” counts

Each customer has two kinds of lead record with us: the leads PropFlow handled itself, and the customer's own AppFolio lead book, imported in late September.

Western Slope's imported book reaches back to September 2025 — roughly a year of their leasing before Clara, with 209 of those people already linked to a tenant — and it carries no tour information of any kind, so those leads could only ever sit in the top bar of a funnel and never in one below it. Putting about 940 pre-Clara leads into the denominator would report Western Slope's inquiry→tour at about 2.7% instead of 13.9%, which is the same mistake as mixing two windows. So Inquiries counts PropFlow's own leads, and the imported book is used only for the lease signal, matched person by person.

The figures

The two are not like-for-like on size either: Western Slope is 90 buildings, Camellia is one. And do not pair one customer's four-week figure with the other's since-live average — that is the comparison this work exists to prevent.

The window, and how it differs from the funnel's

The hours figures use the last four weeks, 2026-09-29 to 2026-10-27. The funnel table above uses leads who arrived between 2026-08-01 and 2026-10-27. These are deliberately different periods, chosen for different reasons — the funnel needs a group of people followed over time, the hours figure needs a recent rate — and they have not been harmonised. Do not present them as one period.

The method, part 1 — the work Clara actually did Measured

Counted by the product's own calculation over each customer's own buildings, read from production. These are volumes since each customer launched.

What the product throws away, which matters for believing the number: calls under 15 seconds (wrong numbers, accidental dials) earn nothing; calls the customer's own team answered are not Clara's volume; test and sandbox buildings are dropped; our internal dogfood channel is excluded; a conversation with nobody on the other side earns nothing. Maintenance conversations earn zero minutes because the per-work-order credit already prices that work — the model deliberately refuses to count the same job twice.

The method, part 2 — the minutes a person would have spent Assumption

These minute values are assumptions. They are not measured and they are not an industry benchmark. They come from the product's own savings model, where they were calibrated twice against real production data — a May audit found two repeat-dial phone numbers generating about 70 of 212 “prospect inquiries” and tightened the rules, and a July recalibration found that fix had over-corrected, with 220 Camellia calls of 15 to 59 seconds across 116 different numbers earning nothing despite being real reception work. Both corrections moved the number down.

Two things the audits are easy to conflate, and the difference is the whole honesty of this figure.

• Which pieces of work count — this IS calibrated against real production data. The two audits changed eligibility: which calls earn credit, at which tier, with what floor for a real conversation. That work was done against real transcripts and real call durations.
• How many minutes each piece is worth — this is NOT measured anywhere. The 45-minute leasing credit, the 10-minute text, the 15-minute email and the 60-minute renewal come from our own reasoning about what a property manager's full workflow involves — the call, the note afterwards, the entry in the system, the follow-up. They have never been checked against an observed person doing the task. There has been no time-and-motion study at any customer, and no published benchmark exists to check them against — our own code says so, noting that the industry publishes outcome metrics and never coordinator minutes-per-touch.

The single external cross-check anywhere in the model is the 5-minute reception credit: it prices to about $3.90 a call at a $47/hour wage, which sits inside the $0.98–$2.15 a call that commercial answering services charge, plus the interruption it replaces.

How to bound it instead of defending a point. Every hour here is volume multiplied by assumed minutes, so the figure scales linearly with the assumption: halve the minutes and the hours halve (Western Slope about 10 to 12, Camellia about 2.4). That sentence is the honest sensitivity and is better said out loud than buried. No confidence interval is offered on purpose — nothing in the data would justify its width, and a made-up plus-or-minus would be a second invented number dressed as rigour.

The method, part 3 — how long each has been live Sourced

Both dates are the product's own launch definition: the earliest piece of Clara activity we have stored for that building. That makes each a lower bound on how long they have been live — Clara demonstrably ran on that date — and never an upper one. Camellia's date is corroborated by an internal review of our homepage statistics, which independently chose 2026-04-06 as the start of its production window; no human-written go-live date for Camellia exists anywhere, which is a named gap below.

Western Slope's dates are human-written, from its own go-live tracker:

• 2026-09-12 — first live check of the phone line on a real tour.
• 2026-09-17 — the planned go-live switch, when their main number was forwarded to us.
• 2026-09-26 — autonomous leasing on: Clara answers new leads by email with no review hold. Before this her email replies waited for a person. This is before the four-week measurement window opens, so every email conversation counted in the window was answered on her own; the earlier review weeks affect only the since-live average.
• 2026-10-09 — renewals set to supervised (one monthly list, approved by reply), and the emergency line live so Clara takes maintenance calls.

The arithmetic, so anyone can redo it

One structural caveat first: the maintenance part of the figure is not windowed — work-order events carry no single timestamp to window on, so a windowed run returns all of it. So leasing and renewals are divided by the window and maintenance by the full weeks live, then added. Mixing those two would have quietly inflated the weekly figure.

Western Slope — estimated 24 hours a week
  last 4 weeks (2026-09-29 → 10-27 = 29 days = 4.143 weeks):
    leasing 79.7 h + renewals 19.9 h = 99.6 h ÷ 4.143 wk = 24.04
    maintenance (all time)             1.4 h ÷ 6.571 wk =  0.21
                                              TOTAL = 24.25  → ~24
  excluding the supervised renewals:
    79.7 h ÷ 4.143 wk = 19.24 + 0.21 = 19.45  → ~19.5
  since launch: 138.1 h ÷ 6.571 wk = 21.02    → ~21

Camellia — estimated 4.7 hours a week
  last 4 weeks:
    leasing 17.5 h + renewals  0.6 h = 18.1 h ÷ 4.143 wk =  4.37
    maintenance (all time)             9.2 h ÷ 29.43 wk =  0.31
                                              TOTAL =  4.68  → ~4.7
  since launch: 205.6 h ÷ 29.43 wk = 6.99     → ~7.0

Volume ramped — in opposite directions, and we are not hiding it

Camellia's decline is almost all renewals, not leasing. Leasing went from 5.14 to 4.22 hours a week — a modest fall. Renewals went from 1.53 to 0.14, because zero renewal cycles opened at Camellia in the last four weeks, against 29 since it went live. Whether that is a lease book that renews in clusters, a quieter autumn, or something worth looking at, this data cannot say, and we are not going to guess at a cause in an investor reply. It is here because quoting Camellia's 7.0 since-live average would be quoting a rate it is not currently running at.

One thing to know before this figure goes anywhere public

Every quote below is word for word from our own meeting recordings, with the speaker's role, the company and the meeting date. Nothing is tidied up and no qualifier has been dropped. Exact links are in the sources section.

The five strongest, in order

The ranking changed, and the new first place is a Camellia quote, not a Western Slope one. The earlier answer led with the Western Slope phone line. A closer read of an already-in-scope Camellia meeting turned up something better, on exactly the axis Beth cares about. Widening the scope to other teams is not what produced it — the newly added teams yielded almost nothing usable; it came from re-reading a meeting that had already been searched, where the filters had missed it because the line contains neither the product's name nor any superlative.

1. “No, it was a lot of help. Having the tours get scheduled and answering calls, that does help a lot.” — on-site property manager, Camellia (JPCO), 2026-10-15. The new first place. It names the two things Beth asked about; it is a correction of our own modesty rather than an unprompted rave — Fede had just said “I don't know how much we help, but as long as we help a little bit, I'm happy” and she pushed back to say it was a lot; it comes from the longer-running deployment; and in the same minute she offered to write a review, so its permission question is nearly settled. It sits in a conversation where Sean had just said occupancy was at about 80% and is now about 92% — his spoken figure in the room, not something measured for this page.

2. “…I know that the plan wasn't necessarily that Clara would, like, reinvent the way that we answer the phone, but it has done that for us, and it's awesome, and I think it's a major selling point for you guys…” — principal, Western Slope, 2026-10-07. Previously first, and still the most quotable sentence we have: unprompted, names a specific function, says the product exceeded what they signed up for, and the customer himself calls it a selling point.

3. “…I think I asked it about an application, and it was like, I'll ask the leasing team, and I was like, wow, what a great answer.” — principal, Western Slope, 2026-09-09. The honesty quote: a customer praising Clara for saying “I don't know” is strong with an investor thinking about whether AI can be trusted.

4. “…I think it's, you know, it's working really well…” and “…it's iron some kinks out, but I think we're all, we're all really excited…” — principal, Western Slope, 2026-09-22, one week after go-live. The balanced one, credible precisely because it admits kinks.

5. “…hearing, like, recording those conversations is really awesome, and then you also, you know, you get the summaries of those…” — principal, Western Slope, 2026-09-22. Names a concrete feature they value, so it is evidence of use rather than goodwill.

Each quote above is trimmed only at its head or tail, and every cut is marked with an ellipsis; the untrimmed version is the matching row in the tables below.

If you want exactly one: the first. If you want two, the first and second — one from each live customer, covering “it genuinely helps with tours and calls” and “it beat our expectations”, which also happens to be the two companies Beth asked about.

Western Slope Property Management

Live on leasing since 2026-09-17 by their own tracker, and the richest source of praise in everything we have recorded.

Camellia Apartments — which is the JPCO team

Gera's ask for “good things the JPCO team have said” is answered by this section. JP & Co (JPCO) is the management entity and Camellia is the building, so they are the same people: the on-site manager at Camellia plus the regional and corporate staff above her. These quotes serve both asks and are listed once, here.

The full exchange behind the first-place quote, because the context is most of its value:

Sean: So, but yeah, we've come a long way. I mean, we were at 80%, basically, and now we're at 92, so we're definitely making progress.

Fede: Getting in at 80. So it's like, I don't know how much we help, but as long as we help a little bit, I'm happy.

On-site manager: No, it was a lot of help. Having the tours get scheduled and answering calls, that does help a lot.

Gera: Can you say that again so I can write it down and put it on the website?

On-site manager: Where can I write a review? I got you.

ConAm

Other teams — supporting colour only, added at Gera's request

Citus / Situs Group (the Denver portfolio, called “MMG/Denver” in older notes). Their leasing only went live around 2026-10-20 and they were still not fully onboarded on 2026-10-26, so they have barely used the product and have little experience to praise — the same structural reason ConAm has nothing. What survives as genuinely theirs is all forward-looking.

Teams checked and deliberately left out: an asset-management prospect running about 2,700 units (2026-09-02) — a prospect, and the recap says they were not ready to sign; a Zillow employee (2026-09-08) — a recruiting conversation, not a customer; and an adjacent vendor who joined a Western Slope call (2026-10-21) — a third party talking about his own product.

Borderline — the founders decide

• “I'm, like, trying really hard, I'm like, I don't know what else to do for these people, but…” and “I mean, I want to try to get this moved forward for you guys, but I'm getting stuck because of the Yardi integration stuff” — ConAm, 2026-09-29. Evidence that our contact is pushing for us inside a company that is blocking us. Genuine warmth, but it is about us as a vendor she likes, not about the product working, and it sits in a call about why we cannot launch. The lane would not send it.
• “Yeah, like, it saves me… the idea is that it is saving me…” — Western Slope, 2026-10-07. Reads like a time-saved quote, but in context he is describing what a proposed voicemail-to-email flow would do, not reporting a saving he has banked. Using it as a time-saved testimonial would misrepresent it.
• “No, I think it can save us VAs. You know, we have 3 VAs right now. I think we're spending…” — Citus / Situs Group, 2026-09-18. The closest anyone has come to putting a number on labour, which is why it is flagged rather than buried. But it is “can save”, said before they went live, by a team that still is not fully onboarded — an expectation, not a measurement, and it cannot answer Beth's staff-time question. Use it as clearly labelled colour or not at all.
• The Camellia speaker label. Three of the older Camellia quotes come from a speaker labelled with the company name rather than a person, because someone joined from a room device. That meeting's own recap attributes the on-site report to the property manager, so that is almost certainly who is speaking, but it could not be proved line by line. The 2026-10-15 quotes, including the first-place one, are not affected — that speaker is labelled by name.
• Which Western Slope person said the 2026-10-07 quotes. That meeting's recap names two Western Slope attendees and the transcript labels them as two speakers, one abbreviated to a single initial, which is where those quotes come from. He speaks unmistakably as a principal with authority over the account — he refers to “our maintenance manager”, “our main office phone number”, and to paying AppFolio for their answering service himself — so the company is certain and the seniority is clear, but which of the two named men it is cannot be resolved from the transcript. A founder will know on sight.
• “Clary answers everything. Great. Every phone call comes in, goes through… Our current message is terrible. Terrible. Yeah, that's what I'm saying, and so it was perfect, because we didn't even have anything good to start with.” — Western Slope, 2026-09-16. Two cautions: it is the day before go-live, so he is describing the plan rather than lived experience, and the “terrible” is about their own existing voicemail greeting, which reads badly if the quote is cut short.
• The 2026-10-21 Western Slope session — the one meeting Gera pointed at — yielded no usable quote, and that is worth knowing before their next meeting. It was a maintenance scoping session and the customer's substantive remarks are constructive criticism: Clara oversharing on first-touch texts, Clara stepping on a staff member's manual actions, and a wish for a deliberate handoff between Clara and staff. Its two positives are both forward-looking rather than testimonial. Every customer speaker in that recording is also redacted, so nothing from it could be attributed to a person anyway.
• The enthusiasm quotes with profanity (“holy shit, this is amazing”). Authentic, and the most energetic thing any customer said, but you may not want it in an investor email verbatim.
• “That's awesome. Oh, man, that's awesome.” Thin as text on its own; it only lands with the context, which is worth paraphrasing around the quote.

Paraphrase, not verbatim — do not quote these

These are our own recap summaries of customer meetings, written by us, not customer words. They read like testimonials and must not be used as quotes. They are listed because they point at where a real quote could be confirmed — the verbatim original for each sits in the transcript under that same recap.

• “The system is working and Jason is happy” — our recap, Western Slope working session, 2026-09-22.
• “Western Slope is getting real value: Clara is booking tours with no staff involvement” — our recap, Western Slope check-in, 2026-09-30.
• “On-site reports Clara is working decently for tours and applications” — our recap, Camellia, 2026-08-06.
• “The team is enthusiastic…” — our recap of the 2026-08-25 Denver-portfolio demo. This is the sentence that produced the false “warmest in the corpus” claim retracted above. It is our words about them, and the transcript underneath it cannot be attributed to anyone.

Kept rather than smoothed away. Every one of these is a named gap, not a closed question.

About the hours figure

• The minutes-per-task assumption rests on our own reasoning and nothing else — no time-and-motion study at any customer, no published benchmark to check it against, and now no customer statement either. If that reasoning is generous, the figures are generous by exactly the same proportion.
• No customer has ever stated a realized time-saved figure on any recording. Checked across 655 meetings with a control, not assumed.
• Which of Western Slope's 35 renewal cycles a person approved is not separable in this data, which is why the figure is a range rather than one number.
• No hours-saved figure has ever been published before, per customer or otherwise — so this is the first, and there is no earlier number to stay consistent with.
• Hours saved is not stored anywhere, so there is no history of this metric to plot. Only the maintenance saving in dollars is kept daily.
• No human-written go-live date for Camellia exists. Its 2026-04-04 is derived from the earliest activity we hold, corroborated by an independent review's 2026-04-06 window start — a lower bound, not a date somebody wrote down. Western Slope's dates are human-written.

About the funnel figures

• Test traffic is not excluded, and that is a different thing from test buildings. No building in these counts is flagged as a test property — that was measured on all 24 properties carrying these leads, with a control that proves the check discriminates. But a team member's own real phone texting a live building is caught by no building flag, and that exclusion is not applied on this path. So a small number of Camellia's 101 inquiries may be internal testing. It cannot move 40.6% far, but it is not zero and it was not measured.
• Company staff are not dropped from these counts, though the product's own screen drops them, so the inquiry counts may be a hair high on that axis too.
• The product folds leads together by email and phone on top of the person-level collapse used here, which was not reproduced — so its own people count could be slightly lower than 245 at Western Slope, which would make 13.9% a floor rather than a ceiling.
• Camellia is one building here. All 208 lead records on that company sit on Camellia Apartments; its second building has none. That is the right reading of the question, but worth knowing before anyone says “two properties”.
• Leases are counted per person, not per lease document, and households were not collapsed, because the product's own funnel counts records rather than households.

About the quotes

• Email was not searched. Gera's scope-add mentioned Slack threads or emails where staff praised the product. The meeting-recap channel and the task thread were covered thoroughly; nobody's mailbox was, because there was no read-only path to one that did not risk touching a live mail account. A named gap, not a closed one.
• One meeting's speaker labels are worthless, and it is a standing hazard for any future quote harvesting. The 2026-08-25 Denver-portfolio demo has only two speaker labels across 426 lines — one customer name and one device name — and no label at all for anyone from PropFlow, even though the recap says Sean ran the demo. So our own pitch sits under the customer's name: lines like “We're the orchestration layer on top” are ours, not theirs. Nothing from that meeting can be attributed to a person and no quote from it is used here. Every other quote was checked against this failure mode before being listed.
• Sarcasm is a live hazard in these recordings and it caught the filter once: on a 2026-07-08 Camellia call the on-site manager says “so their kitchen started flooding a bit, so I was like, oh, that's fantastic” — a textbook false positive that reads as praise to a keyword search and is the opposite. Every quote listed was read with its surrounding turns, which is how that was caught and dropped.
• The best quote was missed by the first two searches, because understated praise contains no product name and no superlative. A third sweep specifically for low-key phrasing returned 12 candidates and that was the only strong one, so the set is reasonably complete — but not exhaustive, and nobody should claim it is.
• Speaker roles are only as good as our recaps state them. Where a recap names a role it was used; no job title was inferred from a name.
• No earlier quote list exists to build on — the task thread was read in full, and “I'd done that a while ago” referred to creating this task, not to a finished list. One qualification: Gera had already spotted the first-place quote live on the 2026-10-15 call and asked the speaker to repeat it for the website, so the best quote here was already known to him in the moment. It was just never written down.
• Two lines that look like quotes and are not: on 2026-10-14 the Western Slope contact says “Wow, that's a… that's a strong compliment. I appreciate it, though” — he is receiving a compliment from Fede. And on 2026-08-06 a Camellia regional staff member says transfers have “been working good” — but “you guys” there is the on-site team, not Clara. Both excluded deliberately, and noted so nobody re-finds them.

Where the lanes used different anchors for the same fact

Everything below was read on 2026-10-27. All reads were read-only. No customer and no investor was contacted. No stored customer login was decrypted. Customer speakers are attributed by role, not by name, so the founders can decide about attribution themselves; one first name appears inside our own recap wording in the paraphrase list. Full names are in the research lane's own sources file.

The funnel figures

• Source: the production database, read partition by partition for each customer's prospect list, counted by a script replicating the product's own funnel rules and the product's own arrival bucketing. Research lane part1-funnel-and-clara-leases; its SOURCES.md carries the exact commands, the per-figure table and the scripts.
• Controls: a server-side count matched the rows walked on both customers (298 and 1,458). A fabricated partition returned zero, so a zero is evidence about the data rather than about the query. Five predicate controls were asserted before counting (a confirmed tour must reach the tour stage, a bare inquiry must not, a signed stamp must reach signed, and so on) and the script refuses to run if any fails.
• Lease population: the distinct people carrying either our own signed stamp or an AppFolio link to a tenant. Both the lease column and the tour→lease rate use that same population, so each percentage is reproducible from its own row: Camellia 9 of 41 = 22.0% in the window, 15 of 80 = 18.8% all-time.
• For comparison only, never as the headline: our own stamp alone would give Camellia 19.5% in the window and 16.2% all-time — lower, because the stamp misses leases AppFolio confirmed.
• Two corrections applied mid-lane, both from an independent fresh read of the production loader: the merge-duplicate rule is broader than first written (no effect — the two rules select the same rows in production), and the product collapses leads by person before counting, which moved Western Slope's inquiry→tour from 12.6% to 13.9%. The figures on this page are the corrected ones.
• Test-building check: all 24 properties carrying these leads read as not-a-test; a known test property reads as one through the same command, so the check discriminates.
• Not read: AppFolio directly (a stored customer login may never be decrypted — their leads are visible as the imported records instead); the live product screens (see the no-screenshot section); tour records themselves (they carry no attended field at all, so they could not have produced a show-up-adjusted denominator).

The hours figures

• Source: the product's own savings calculation — the same function behind the hours-saved figure on Clara's command centre page — run unchanged against production per customer, scoped to each customer's own buildings. Research lane part1-hours-saved-per-week; its SOURCES.md carries the command, the script, the raw output and a per-figure table marking each value measured, assumed or derived. Run at 2026-10-27 23:00 UTC.
• Status of each layer: the volumes are measured; the minutes per unit are an assumption from the product's own constants; the go-live dates are measured lower bounds, with Western Slope's milestones human-written in its go-live tracker; every hours-per-week figure is therefore estimated.
• Controls: the property rosters total 92, matching exactly the 92 the independent homepage-statistics review reported for the two live customers. A fabricated property id returned no rows while Camellia returned 238, so the search discriminates. Every October metric row read carried real rather than synthetic provenance. The documented “windowing does not touch maintenance” behaviour was verified rather than assumed, on all three cohorts.
• Prior art searched: our docs, 2,168 published pages and the team transcript mirror — no published hours-saved figure exists. The 2026-10-07 homepage-statistics review is the page that ruled the metric out of the website headline.
• Side effects of that run, disclosed. Despite asking for stored accounting only, the module queued read-only report calls to both customers' AppFolio accounts, and Western Slope's request allowance was briefly exhausted during the run. That allowance is a per-few-seconds rate bucket rather than a daily quota, and the system is built to absorb it as retries, so it refilled on its own within seconds. No writes occurred anywhere. One thing was NOT checked: whether an alert fired for that account in the window. The alerts channel is stored by name and the reading tool needs its id, so it could not be read from that session. That is an unread, not a pass — do not report it as “no alerts fired”. Also disclosed by that lane: one unbounded property read before a repo guard refused the pattern, and one 10,059-row read that should have been narrowed. Both read-only.

The quotes

• Source: our meeting-recap Slack channel — one recap per meeting with the full transcript in its replies. 9,164 messages covering 2026-06-30 to 2026-10-27, 655 meetings. Read live for the threads Gera named, and through a local index of the same channel for the sweep. Research lane part2-verbatim-customer-quotes; its SOURCES.md carries one row per quote with the speaker's name, the company, the meeting date and an exact permalink, plus the searches and their controls.
• Controls: a positive-sentiment sweep of about 45 terms across all 655 meetings, restricted to non-founder speakers, returned 292 candidate lines, of which 16 survived being read for scope, subject and sincerity — so the zeros reported here are about the subject rather than the search. Founders' own speech was excluded deliberately: a founder saying a customer loves it is hearsay, not a quote. The realized-time-saved sweep returned zero usable customer lines against 71 matches on founders' speech, which is the control proving that search worked. ConAm: 88 of 655 meetings mention them and the client calls were read line by line.
• Each quote was read with roughly seven turns either side before being listed, to rule out sarcasm and confirm the subject was the product. Two changed status as a result and are in the borderline list.
• Dates are taken from each recap's own header line, not from when the recap was posted to Slack.
• Sources checked that did not answer: the transcripts folder named by one of our own tools (it belongs to a different machine); a Zoom transcript tool that is not installed on this box; Clara's own call recordings with tenants and prospects (they cannot contain a customer praising us to us); and Google Drive, which was not needed and is named as unexplored rather than empty.

Driver's rulings this page follows, for anyone reading it later: lead with the hours range rather than the single number; publish both readings of “leases Clara had worked” with the loose one leading and the mandated phrasing; leave Western Slope's tour→lease cell blank with its reason beside it; and treat the screenshot route as closed. All four were in-plan and reversible presentation calls, so no founder was asked.