All Briefs
Mission Brief · 019—

Pirxey World: our second brain at the leadership table.

We could only discuss the facts someone had time to gather. Pirxey World brings records from the tools we can access into one discussion, so our leadership team can question assumptions and explore options together. These are examples from our work, with the gaps left visible.

4
board questions, before and now
25
tools on our map, read with each leader's own logins
6 of 15
departments with a strategic repository readable by owners and directors, September 2026
Read the four questions
Four questions · before and now

Bring the records. Discuss the choices.

Someone remembered the conversation. Someone else knew the staffing plan. Bringing their evidence together took work before the discussion could begin. Our second brain helps gather accessible records; it does not turn them into a verdict.

Recorded work · audits and staffing radar

Can we take the next project without hiring?

Before

The official staffing table had fallen behind the conversations. Asking around gave us fragments of the picture.

Now

The radar compares headcount numbers week to week. Numbers only, no personal data.

  • Slack
  • ClickUp
  • PirxeyOS
  • company wiki
Decision

Delivery can discuss a move, a later start or a hire using the same evidence.

Still unknown

Logged hours do not prove spare capacity. A person caught hours assigned to the wrong person. We have not measured avoided hires.

Recorded work · separate directory and outreach records

Who should hear about the new offer first?

Before

We assembled lists and checked colleagues' prior contact by hand. One campaign carried a claim we could not defend.

Now

We check earlier contact and return to call transcripts. The campaign was stopped; unsupported fields in the brief now stay empty.

  • Slack
  • Gmail
  • Fireflies
  • HubSpot
  • LinkedIn + Sales Navigator
Decision

Sales approves the recipients, wording and order of contact.

Still unknown

The directory and outreach log are separate records, not a conversion funnel. Interest in a previous conversation is not current buying intent.

Recorded review · June 2026

Are we making decisions on a false assumption?

Before

We believed agreements signed after one date all used the new template.

Now

Checking the files disproved the clean cut-over. Our first comparison also hid a changed clause; a person caught the filter error.

  • Drive + Docs
  • company wiki
Decision

Operations could plan one template per agreement type, with missing files and unresolved cases visible.

Still unknown

Drafts are not signatures. Unreadable scans and substantive differences still require a person to review them.

Composed example · combined note still proposed

Where does my team need support this week?

Before

An unread badge or a manager's recollection told us where attention was needed. Opened messages disappeared from that count.

Now

Separate skills check replies and meeting promises. A reply elsewhere can close a question; an acknowledgement may leave it open.

  • Slack
  • Gmail
  • Fireflies
  • accounting system
Decision

The manager chooses where to step in and which commitment to clarify.

Still unknown

The combined note is not routine. Payment information stays missing without authorised access; silence does not establish dissatisfaction.

Contracts · the filter hid it

The filter hid it.

By hand: about 90 hours of people time over four to six weeks. With our second brain: about 17 hours. 224 contractors, eleven sections per agreement, one belief: “everyone who signed after a certain date is on the new template.” Checking it by hand means opening every PDF against the template and keeping a sheet; few companies ever do it. We did it at about half that size, and the belief was false: there was no clean cut-over.

Our first automated comparison was wrong too. Its list of differences to ignore swallowed the very clause we needed to find, because hyphenation after PDF extraction made the passage look like noise. A person noticed, the method changed, and the list of who is on which version became true, missing files and unresolved cases included. The 17 hours are edge cases, checking, signing and the four hours it took to find and fix the filter; classification and drafts were minutes of machine time.

Bank transfers · the daily collections job

Who actually paid?

Every day: match transfers to invoices, mark what is paid in six places, decide whom to chase. About thirty invoices a month. The work hides in the exceptions: a sister company pays, an intermediary sends money with no reference, one transfer covers several invoices to the cent, a wire arrives short by the fee, gross meets net, an arrears client pays the oldest debt first, a year in an invoice number is impossible.

Our rules match only when the payer is known and exactly one open invoice fits; everything else is listed for a person. A script runs twice a day from the bank's own emails and writes three of the six places; a person does the other three. Try the fixture yourself, then see what the rules propose.

Try the fixture: eight invoices, six receipts

Invented exercise · eight invoices · six receipts

Who actually paid?

Every name, reference and amount below is invented. DMO and DLC are demo currencies. All invoices start open in the books; the receipts have already arrived at the bank.

Demo rules: known payer, one invoice within 2.00 DMO or 0.02 DLC of the receipt, or an exact total of all that client’s open invoices. Domestic amounts use gross. Arrears accounts stay with a person.

The invoice ledger

Your working allocations appear beside the saved rules. “Paid” describes that allocation, not independent verification.

  1. DEMO-A/2026
    Halden Clinics
    Why this one is tricky

    Known sister company pays this client's invoices.

    100.00 DMO due

    Your ledger
    open · 100.00 DMO open

    Saved rules
    paid · 0.00 DMO open

  2. DEMO-B/2026
    Kestrel Robotics
    Why this one is tricky

    One of this client's two open invoices.

    60.25 DMO due

    Your ledger
    open · 60.25 DMO open

    Saved rules
    paid · 0.00 DMO open

  3. DEMO-C/2026
    Kestrel Robotics
    Why this one is tricky

    The other open invoice; there are no more.

    39.75 DMO due

    Your ledger
    open · 39.75 DMO open

    Saved rules
    paid · 0.00 DMO open

  4. DEMO-D/2026
    Tern Logistics
    Why this one is tricky

    Wire may arrive short by a bank fee.

    200.00 DMO due

    Your ledger
    open · 200.00 DMO open

    Saved rules
    paid · 0.00 DMO open

  5. DEMO-E/2026
    Lumen Dairy
    Why this one is tricky

    Domestic invoice: compare the receipt with gross, not net.

    123.00 DLC due gross (100.00 DLC net)

    Your ledger
    open · 123.00 DLC open

    Saved rules
    paid · 0.00 DLC open

  6. DEMO-F/2026
    Mosswood Hotels
    Why this one is tricky

    Older arrears invoice. Statement uses oldest first.

    75.00 DMO due

    Your ledger
    open · 75.00 DMO open

    Saved rules
    needs review · 75.00 DMO open

  7. DEMO-G/2026
    Mosswood Hotels
    Why this one is tricky

    Newer arrears invoice. A title cannot override the statement.

    125.00 DMO due

    Your ledger
    open · 125.00 DMO open

    Saved rules
    needs review · 125.00 DMO open

  8. DEMO-H/2026
    Orchard Pay
    Why this one is tricky

    Overdue. Finance checked the statement: no receipt for this client.

    90.00 DMO due

    Your ledger
    open · 90.00 DMO open

    Saved rules
    open · 90.00 DMO open

Select a bank receipt
Your allocation

HC Group Holding
100.00 DMO received · DEMO-A/2026

Select invoices, then save. A single invoice takes up to its open balance. For a split, edit the amounts; the initial fill follows ledger order. Different currencies cannot be combined.

Your ledger starts open. Select a receipt, then the invoices it pays.

No time limit or score. The worked proposal below is always readable. JavaScript enables your own attempt.

What the rules propose

This is a worked example of saved matching conditions. A match means “met the matching rules”. It does not establish a production success rate.

  1. Receipt 1 · HC Group Holding

    Met the matching rules

    DEMO-A/2026: 100.00 DMO cash.

    Why receipt 1 gets this result

    Payer known, exactly one open invoice fits within fee tolerance.

  2. Receipt 2 · Kestrel Robotics

    Met the matching rules

    DEMO-B/2026: 60.25 DMO cash.

    DEMO-C/2026: 39.75 DMO cash.

    Why receipt 2 gets this result

    Payer known; receipt equals the sum of ALL this client's open invoices, to the cent. No subset guessed.

  3. Receipt 3 · Tern Logistics

    Met the matching rules

    DEMO-D/2026: 198.00 DMO cash + 2.00 DMO fee allowance.

    Why receipt 3 gets this result

    Payer known, exactly one open invoice fits within fee tolerance. The impossible year is ignored, not repaired: this match uses payer and amount.

  4. Receipt 4 · Lumen Dairy

    Met the matching rules

    DEMO-E/2026: 123.00 DLC cash.

    Why receipt 4 gets this result

    Payer known, exactly one open invoice fits within fee tolerance. Receipt matches gross; the ledger's net figure is smaller.

  5. Receipt 5 · Mosswood Hotels

    Needs review

    125.00 DMO remains unallocated.

    Why receipt 5 gets this result

    Arrears client — oldest first? Needs a person to check the statement, whatever the title says.

  6. Receipt 6 · Payflow Settlements

    Needs review

    125.00 DMO remains unallocated.

    Why receipt 6 gets this result

    Payer unknown, no usable reference — needs review. An amount alone does not identify a client.

For the arrears receipt, oldest first would put 75.00 DMO on DEMO-F and 50.00 DMO on DEMO-G, leaving 75.00 DMO open on DEMO-G. A person checks the statement before saving that split. The intermediary’s separate receipt remains unresolved.

Which reminder leaves the desk?

Preview consequences from

Scenario preview: finance owns the reminders. Nothing is sent from this page or by the matching script. Your preview follows your allocations, including mistakes.

Showing the saved rules’ proposal.

  • Withdrawn · DEMO-A/2026

    Halden Clinics. Balance settled under this allocation; withdraw the reminder.

  • Withdrawn · DEMO-B/2026

    Kestrel Robotics. Balance settled under this allocation; withdraw the reminder.

  • Withdrawn · DEMO-C/2026

    Kestrel Robotics. Balance settled under this allocation; withdraw the reminder.

  • Withdrawn · DEMO-D/2026

    Tern Logistics. Balance settled under this allocation; withdraw the reminder.

  • Withdrawn · DEMO-E/2026

    Lumen Dairy. Balance settled under this allocation; withdraw the reminder.

  • Kept as draft · DEMO-F/2026

    Mosswood Hotels. The match or remaining balance still needs a person.

  • Kept as draft · DEMO-G/2026

    Mosswood Hotels. The match or remaining balance still needs a person.

  • Sent · DEMO-H/2026

    Orchard Pay. Overdue and finance's statement check found no receipt. Finance sends the reminder.

In the saved-rule proposal, Halden’s reminder is withdrawn once the sister-company payment is matched. Mosswood’s stays a draft while the statement and intermediary receipt are unresolved. Orchard’s is sent by finance after the statement check confirms no receipt.

“Paid” has six places to go

These ticks show ownership in the saved-rule workflow, not a log of completed writes. Automatic updates apply to accepted matches; unresolved receipts stay recorded for review. Check off the human follow-up you have accounted for in this exercise.

  • ✓ Automatic · Transfer record

    Receipt recorded; client, invoice periods and match comment filled for accepted matches.

  • ✓ Automatic · Collections record

    Accepted invoices marked paid; balance, payment date and comment updated.

  • ✓ Automatic · Monthly invoice rows

    Unpaid amount cleared on every row of each accepted invoice; verify the write succeeded.

  • If it exists: add the payment, update totals, balance and invoice status.

  • Revise priorities and reminder drafts; remove settled arrears from the lists.

  • Where applicable: record the match and check oldest-first or partial allocations.

One “paid” flag is not the end of the job. The client ledger, the chase list and, where applicable, the partner statement still need a person.

The useful result is a different chase list: a reminder withdrawn because the sister company had paid, one unresolved match kept as a draft, one genuinely unpaid invoice chased. The repeatable part is a script; what our second brain changed is who can get such a script and keep it correct — the finance lead describes the rule in a sentence, the exceptions stay hers.

The price of keeping the answer current

By hand · about 8.8 hours of people time per month
Roughly 32 matching passes × 8 minutes, plus 5 awkward cases × 30 minutes of extra investigation, plus 2 hours deciding whom to chase. The matching-pass count is a workload assumption, not a claim that every invoice receives exactly one transfer.
Assisted · about 1 hour of people time per month
Attention goes to what the rules refused and the remaining human follow-up. The six-place checklist still applies; the script does not complete all of it.
Machine time · scheduled twice a day
The script reads notifications, applies the saved conditions and writes the supported updates.
One-off setup · about 12 hours of people time
Writing and checking the script. Ongoing rule changes need attention too; the recurring estimate is not a guarantee of future maintenance effort.
Clients quietly waiting

Who is waiting on us?

Six invented clients, five channels, one week, Thursday afternoon. Decide who owes whom a reply, then compare with the reconstruction our recipes would make.

Try the exercise: six invented clients, five channels

It is Thu 16:00. Six clients, five channels. The last thing in each channel is shown; a dash means nothing this week.

Six clients, 24 cells to open. Decide who owes whom, then check.

ClientTeam chat · brand ATeam chat · brand BMailMeeting notesPaymentsfinance onlyWho owes whom?
Halden Clinics—us · Tue 15:10Answered here: the export runs nightly, we added the two fields.client · Tue 14:02Can the export include the two extra fields we discussed?—not yours to open
Kestrel Roboticsclient · Mon 10:40 · openedAny update on the staging environment for our testers?———not yours to open
Tern Logistics——us · Wed 09:20Re: “when is go-live?” (their question, Tue 17:45) — Thanks, we’ll confirm the date shortly!—not yours to open
Lumen Dairy———us · Mon 11:30We promised the revised estimate by Thursday.not yours to open
Mosswood Hotelsus · Wed 12:05You missed Tuesday's demo — does Friday 10:00 work instead?——note · Tue 14:00Demo: client did not join.not yours to open
Orchard Pay————not yours to open
The morning note the recipe would print for you
  1. Halden Clinics — asked by mail Tue 14:02, answered in chat B Tue 15:10 (68 min). Nothing open.
  2. Kestrel Robotics — waiting on us since Mon 10:40 (chat A, opened, no reply). Reaction target 3 h: missed by three days.
  3. Tern Logistics — asked for the go-live date Tue 17:45 (mail); our Wed 09:20 reply is an acknowledgement, not an answer. Still owed.
  4. Lumen Dairy — promise from Mon 11:30 meeting: revised estimate by Thursday. Nothing sent. Due today.
  5. Mosswood Hotels — skipped Tue demo; our reschedule Wed 12:05 unanswered for 28 h. They owe us; worth a call.
  6. Orchard Pay — payments: not visible to you — ask finance.

Before it prints a status it rebuilds the whole timeline of each conversation, across channels, and lists every source it searched — even the empty ones. Reply targets it measures against: a reaction within 3 working hours, an answer within 24. The rest — who to call, what to promise — is a person's.

Real ingredients: every trap is one we have fallen into — a message opened but never answered, gone from the unread count that minute; a thank-you mistaken for an answer; a reply in another channel; a promise only in a meeting transcript; a column you cannot open. The exercise shows how many cells a person has to check, and what a recipe reads before printing a status.

The reply targets we measure are a reaction within 3 working hours and an answer within 24. A workspace's complete unread state takes one call, about nine seconds. Before a client call, the previous meeting's transcript is re-read for promises; every source searched is listed, even when empty. An AI on your own logins cannot see finance's system, so it says "not visible to you" instead of guessing. One client-health table across all of this is the composed version of five recipes we run separately.

Two more jobs, as records

A launch list and a month of hours.

Records of work we did, dated where the sources give a date, with the mistakes beside the work they affected.

We have a new offer. Who should hear about it first?

Outreach log, June 2026. Nine targets after a person approved them; six actioned — three invitations, two direct messages, one left for a human to handle. Invitations go out one at a time, ten to fifteen a day, with a note under two hundred characters. Prior contact is checked in two mailboxes, two chats and the LinkedIn inbox first, to reduce duplicate contact. A campaign was stopped five days after it went out because one claim could not be defended; a week went into digging the decisions out of call transcripts. Now the campaign brief is filled from the raw transcript, with a link to the minute behind every answer, and unsupported fields stay empty.

A separate record is a directory of 277 companies turned into a checked sheet; the sources do not join it to the outreach log as a funnel. Enrichment ran in about five minutes: website found for roughly nine in ten, a LinkedIn page for about three in four, with the rule unsure → leave it empty. An earlier parsing bug had silently lost 12 of 304 entries before it was caught; the fixed run counts them.

By hand this is a sheet, a week of copy-and-paste and a list that is stale in a fortnight; assisted, the enrichment and the cross-checks repeat on demand. A person approves each message and everything a client actually reads.

Did we log the month properly, and who is actually free?

Hours audit, May 2026. Seven detectors ran over one recorded month of 7,878 entries: empty description; under five characters; a single entry of eight hours or more; twenty hours or more (a forgotten timer); suspiciously round hours in bulk; generic words; more than half of someone's hours in overhead. The pull takes two to three minutes and the report is name-free by default. Hours are not a productivity score, and a long entry is a question, not a verdict. Our wiki's rule: time data exists to protect the team, not to control it.

A person-matching step that reported “confidence 0.87” assigned someone else's hours; a person caught it. An alarm of “−190 h / +340 h” between two trackers came from how entries longer than a day are split; bucketed by reporting date, every gap was under two hours. Both are now checks in the recipe.

Staffing uses different evidence. On one August day the official table had three rows nobody had touched since March, while a single team post held fifteen open processes and forty-four candidates. The weekly staffing radar compares headcount numbers week to week: numbers only, no personal data.

Fictional decision · scripted scene

Would you still do it?

Illustrative records · February 2026 · no AI runs here

Talk

GmailEmail preference superseded by a newer note.Jun 2024 · Stale

Meetings

FirefliesThe client missed a demo; the reason is unknown.Feb 2026

Sell

HubSpotThe client asked about month-end matching.May 2025
LinkedIn + Sales NavigatorThe intended recipient has left.Jan 2026

Deliver

ClickUpNo capacity before April; availability still needs confirmation.Feb 2026

Run the company

accounting systemSlower than last quarter. Cause and overdue status unknown.Feb 2026
Drive + DocsRenewed. No bearing on this decision.Feb 2026 · Irrelevant

Know

strategic repositoryQuarter goal: reduce onboarding effort.Jan 2026
One person

What we brought to the meeting

Announce the offer, propose March, publish the case study.

An earlier client question and a plan excerpt. The proposal is still an assumption.

Still to establish

Current need, recipient, capacity and publication consent.

Publication consent · not visible to you. Ask an authorised owner. No switch opens this source.

Proposed · not routine

Still open.

Project margin. Compare hours, rates and invoices. The promise after handover. Compare sales commitments with delivery scope. The quarter against the objective. Bring departmental evidence to a discussion about what to continue, stop or investigate. These remain proposed analyses, not measured results.

Recorded means described in our existing work records; the scene above is invented. 015 maps the tools, 016 follows the index, and 017 explains own logins, repositories and where text goes.

Mission control standing by

More room to think together.

We built this for ourselves. The useful change is the evidence we can bring to a discussion. The decision stays with us.