57%
of working time goes to communicating —meetings, email and chat— and 43% to creating
57 of every 100 hours
Criterio EA
Executive office
Evidence before the decision
Ofelia sets the executive’s week beside the meetings and commitments: what clashes, which meeting has no materials, which minutes are still unsigned, which agreement has no owner or date and which email is waiting for a reply. She prepares each decision with its options and evidence, without ranking them. She does not send, schedule or decide.
What it costs, measured by third parties
57%
of working time goes to communicating —meetings, email and chat— and 43% to creating
57 of every 100 hours
55%
say the next steps at the end of a meeting are unclear
55 of every 100 people
68%
do not have enough uninterrupted focus time in their workday
68 of every 100 people
Third-party figures, each with its source, year and sample, verified against the publisher’s own text. The agent’s own figures are further down, in the test evidence.
Criterio is open source. If it creates value for you, help fund testing, documentation and the next capabilities.
Ofelia is Criterio’s EA family, an open platform of agent families. See the families
Offices
Ofelia processes nothing specific to a country: no invoices or regulations, only the calendar, the minutes and the email. That is why her test is counted in offices, not countries: two with defects seeded on purpose and two controls, where any signal is a failure. What does change from one office to another is how its calendar and email provider exports.
A bank’s president’s office
Colombia
32 documents · 10 defects seeded on purpose
Round 1
10 of 10 signals, none extra
17 of 17 data points
Verification after the fix
50 of 50 signals, none extra
51 of 51 data points
What failed, with its defect: EA-9
A credit union’s general manager’s office
Ecuador
42 documents · 6 defects seeded on purpose
Round 1
4 of 4 signals, none extra
16 of 17 data points
Verification after the fix
35 of 35 signals, none extra
45 of 45 data points
What failed, with its defect: EA-6, EA-9
Control office
Colombia
23 documents · no defect: any signal would be a failure
Round 1
no signal expected · 1 extra
21 of 22 data points
Verification after the fix
no signal expected
22 of 22 data points
What failed, with its defect: EA-7, EA-8
Control office
Ecuador
23 documents · no defect: any signal would be a failure
Round 1
no signal expected
22 of 22 data points
What the material also tests
Criterio · EA family
She prepares the calendar, each meeting with its decisions, the draft minutes and what needs a reply. She does not decide or send anything.
CRITERIO · EA FAMILY
Where the family stands: built and tested. The engine passes its deterministic test against four synthetic offices —two with seeded defects and two controls— and the agent passed the full round with a key she never sees. Next are travel and the exports of each provider.
Where each name comes from
Ofelia and the portal
Ofelia installs on her own, and Rostrum publishes her report only if the office decides so. The screenshot is Rostrum on the test office of a financial entity’s president, as the round with the agent left it.
criterio-assistant
From the Greek ōphéleia, “help”: she is the one who prepares.
Processes the exported calendar, each governing body’s meetings and the saved correspondence: the week against the meetings and commitments, each meeting’s pack with the record of each decision, the draft minutes and what is waiting for a reply. You send and decide.
What only she does
Checks every agreement in every set of minutes against its owner and date, and counts how many times each commitment moved. Prepares each decision with the options the documents bring and their evidence, without ranking them.
What she answers
Never
Sends, schedules, accepts or moves anything; recommends an option, prioritises for the executive or evaluates a person.

Its commands, by group
ea-setupea-wakeea-scanea-agendaea-prepea-minutesea-commitmentsea-mailea-weekly9 commands · 6 skills
Ofelia’s report is not published in Rostrum by default: the executive’s office is among the most sensitive things an organisation has. If the configuration says so, it goes in like any other family, and Rostrum shows what her engine already wrote, with the date it wrote it.
How it enters the portal
/criterio-assistant:ea-setup, leaves its configuration in .criterio/direccion/<executive>/ and marks its root folder with _CRITERIO_DIRECCION.txt.If its configuration says it is published, Rostrum adds it to the map and raises the portal; if not, server-setup names it among those not published:
/plugin install criterio-server@criterio /criterio-server:server-setup /criterio-server:server-start
What it has in Rostrum

What it costs you today
Ofelia extends whoever assists the executive. Enter how many people assist them, what they cost and how many hours a week each task takes them. Nothing you type leaves your browser.
Write the costs in the currency you choose. There is no conversion: changing the currency clears the costs, because an exchange rate written into a page is wrong within weeks.
The people
Hours per week on each task
Hours are per person, per week. The starting people and hours are examples: replace them with yours. The cost is yours to write: we do not print one we cannot source.
Your numbers · per year
Ofelia
Whoever assists the executive · 14 h a week each (35% of the week)
644 ha year
—a year
Write the monthly cost of each group to see what it costs.
What this number is and what it is not
It is the cost of the hours the office spends today on tasks Ofelia prepares: checking the calendar, building the meeting pack, drafting the minutes, chasing commitments, sorting the correspondence. It is not a saving. The person still reviews, decides and sends.
How many of those hours come back we will publish in the evidence section, with each round’s tests. Until that figure is measured, we do not put it here.
Test evidence · updated with every round
Every figure comes from a real run of the repository, with its count and its 95% confidence margin. The engine is tested on every change to the repository; the agent, in rounds on one office’s material.
How it is tested
2 offices and 2 controls
The engine, on all of them. The agent, so far on all 4 offices of the material, in 2 countries. The control offices carry no defects at all: any signal is a failure.
2 commands per office
Ofelia runs her setup and processing —the commands that process the documents and write the records—, with a token cap set before starting. The rest calculate on those records with the engine, which is tested on every change to the repository.
120 documents
What the material of the measured offices holds, in text, eml, msg, CSV and ics. The key is not counted: the agent does not have it.
9.9 million tokens
The work of the language model (LLM) in round 2, with its verifications. No command was repeated.
A key she never sees
What is expected is written in each office’s key before the run: every seeded signal and every data point the written documents carry.
95% margin
Every figure comes with its confidence margin: it says how much the sample supports, which a percentage alone does not.
What she wrote is graded against the key she never saw: every signal by its name and its key, and every data point in the records against what the written documents carry. Each office counts with its latest run, whatever round it came from; if a defect was fixed, the one that verified it. The first runs, with what they missed, are in the chart and in the detail.
Record data
97.4% or higher
Measured: 100%. The headline is the floor of the 95% confidence margin: what the test supports.
146 of 146 data points extracted from the written documents match the key.
Measured
Verification of EA-9: Ofelia on a bank’s president’s office · Colombia, 51 of 51 data points.
Verification of : Ofelia on a credit union’s general manager’s office · Ecuador, 45 of 45 data points.
95% confidence margin (Wilson): 97.4–100%
Signals found
95.6% or higher
Measured: 100%. The headline is the floor of the 95% confidence margin: what the test supports.
85 of 85 signals the key expected.
Measured
95% confidence margin (Wilson): 95.6–100%
Correct signals
95.6% or higher
Measured: 100%. The headline is the floor of the 95% confidence margin: what the test supports.
85 of 85 signals raised were correct.
Measured
95% confidence margin (Wilson): 95.6–100%
Controls without false alarms
2 of 2
2 of 2 control offices without a single extra signal.
Measured
8 commands run on 4 offices, each with its output, its cost and its grade published.
How the family moves from one round to the next
Each dot is one office in its first run of the round; the line is its 95% confidence margin. Verifications are shown apart: they repeat what was fixed. The axis starts at the lowest value published.
Round 11
Round 1 · verification2
Round 23
Round 2 · verification4
Results by country · each country’s latest run
Each country with its offices in their latest run with the agent. The margin is the 95% Wilson interval.
| Country | Entities | Round | Record data | Signals found | Correct signals | Controls without false alarms |
|---|---|---|---|---|---|---|
| Colombia | 2 | 2 | 76 of 76 · 100%95% CI 95.1–100% | 50 of 50 · 100%95% CI 92.8–100% | 50 of 50 · 100%95% CI 92.8–100% | 1 of 1 |
| Ecuador | 2 | 2 | 70 of 70 · 100%95% CI 94.7–100% | 35 of 35 · 100%95% CI 90.1–100% | 35 of 35 · 100%95% CI 90.1–100% | 1 of 1 |
Defects the rounds found · 9
Every error is logged with its origin and the commit that fixes it. It counts as verified only when a run after the fix gives the expected result.
Where they came from
For auditing
The full detail of every run. Each block opens on its own.
Ofelia ran her commands on each office’s material, with a token cap set before starting. What she wrote was graded against the key she never saw: every signal by its name, and every data point in the records against what the minutes and the emails carry.
| Test material | Agent · commands | Signals | Record data | Tokens |
|---|---|---|---|---|
| a bank’s president’s officeRound 1 · Colombia | Ofelia · 2 | 10 / 10 | 17 / 17 | 1.4 M + 1.2 M repeated |
| control officeRound 1 · Colombia | Ofelia · 2 | 0 / 0 · +1 | 21 / 22 | 1.3 M |
| control officeRound 1 · Colombia · verification | Ofelia · 2 | 0 / 0 | 21 / 22 | 1.5 M |
| control officeRound 1 · Colombia · verification | Ofelia · 2 | 0 / 0 | 22 / 22 | 2.1 M |
| control officeRound 1 · Ecuador | Ofelia · 2 | 0 / 0 | 22 / 22 | 1.5 M |
| a credit union’s general manager’s officeRound 1 · Ecuador | Ofelia · 2 | 4 / 4 | 16 / 17 | 1.7 M |
| a credit union’s general manager’s officeRound 1 · Ecuador · verification | Ofelia · 2 | 4 / 4 | 17 / 17 | 1.8 M |
| a bank’s president’s officeRound 2 · Colombia | Ofelia · 2 | 50 / 50 | 50 / 51 | 1.6 M |
| a bank’s president’s officeRound 2 · Colombia · verification | Ofelia · 2 | 50 / 50 | 51 / 51 | 1.6 M |
| control officeRound 2 · Colombia | Ofelia · 2 | 0 / 0 | 25 / 25 | 1.9 M |
| control officeRound 2 · Ecuador | Ofelia · 2 | 0 / 0 | 25 / 25 | 1.2 M |
| a credit union’s general manager’s officeRound 2 · Ecuador | Ofelia · 2 | 35 / 35 | 44 / 45 | 1.6 M |
| a credit union’s general manager’s officeRound 2 · Ecuador · verification | Ofelia · 2 | 35 / 35 | 45 / 45 | 1.9 M |
EA-1Fixed and verified
Setup stopped: the engine required the entity's tax ID, and no document in the office carries it. Ofelia did not invent it and left the configuration unvalidated.
The office is identified by its executive: the engine no longer asks for the tax ID and the report header does not show it when missing. Both commands were repeated on banking: 10 of 10 signals and 17 of 17 data points.
the code · a bank’s president’s office · Colombia · d11470ea
EA-2Fixed and verified
The inventory flagged the .ics calendar as an «unsupported format», although Ofelia's engine processes it.
The inventory treats as readable the formats the family's structure declares as its own by extension. Verified in the banking repeat.
the code · a bank’s president’s office · Colombia · d11470ea
EA-3Fixed and verified
The citations of the agreements and commitments named the section, not the line: «Commitments, first item».
The skills say each citation points to a line —`line N` or `item N`— and that quoted text goes with it, not instead. Verified in the banking repeat.
the agent’s instructions · a bank’s president’s office · Colombia · d11470ea
EA-4Fixed and verified
The key contradicted two skills: it cited the decision in the minutes that defer it, while decision-brief cites it where the material writes it, and it took a request for a meeting with the executive as routine, while correspondence-record only takes an already approved appointment as routine.
The key now follows the skills. Ofelia had done both right.
the answer key · a bank’s president’s office · Colombia · d11470ea
EA-5Fixed and verified
The harness did not recognise as the same citation one that names the section before the item —«Agreements, item 1»— or one that quotes the line's text in « ».
`tests/ronda_familia.py` resolves the line by «line N», by «item N» even after the section, or by the quoted text, and also applies it to the deadline's citation.
the test harness · a bank’s president’s office · Colombia · d11470ea
EA-6Fixed and verified
The citations of the agreements carried the office folder in front —`direccion/gerencia-general/organos/…`—, so they did not open from the office.
The skill says each citation is relative to the office folder, and the engine reports a citation that is not there as a record issue, to be corrected before calculating. Ofelia was rerun on the credit union: 17 of 17 data points.
the agent’s instructions · a credit union’s general manager’s office · Ecuador · d11470ea
EA-7Fixed and verified
In the Colombian control, Ofelia left open a commitment whose completion was in the meeting's material, and the engine raised an overdue commitment that was not.
The record skill says that, before leaving an overdue commitment open, its evidence is looked for in the governing body's meeting material, and `ea-scan` repeats it. Ofelia was rerun on the Colombian control: no extra signal and the commitment met with its evidence.
the agent’s instructions · control office · Colombia · d11470ea
EA-8Fixed and verified
In the run that verified EA-7, Ofelia added to what is still unknown about a decision a fact the minutes already state: that it is decided at the next meeting.
The decision skill says `missing` holds only what the material does not answer, not what the documents already say. Ofelia was rerun on the Colombian control: no extra signal and 22 of 22 data points.
the agent’s instructions · · d11470ea
EA-9Fixed and verified
The batch's proposals carried no figures, and the regional office one had the same options as the vendor one: «carry on as today» and «change vendor». Ofelia said so in what is still unknown for each decision, and she was right; the key expected nothing to be missing.
Each proposal carries its two options, specific to the matter, with what each one costs, and says the offers are complete. Setup and processing were repeated on banking and on cooperatives: 50 of 50 and 35 of 35 signals, and 51 of 51 and 45 of 45 data points.
the test data · a bank’s president’s office · Colombia · 10bb31d7
Synthetic material from a bank’s president’s office (Colombia), a credit union’s general manager’s office (Ecuador), plus 2 control offices with no defects: 4 offices with their calendar, meetings and correspondence, and a key the engine never sees.
85
signals planted on purpose, each with its key
12 of 12
family signals planted at least once
2
control offices, where any signal is a failure
The engine raises every expected signal and none extra, and the gate that checks it runs on every change to the repository. The material also tests the rules of real operation: a clash that only appears after the clock change, an old calendar that gives no clashes, a draft in the minutes folder and an excluded folder.
What comes next
A results page that only says what passed is advertising. What makes it auditable is what it says is not known yet.
Roadmap
Each increment is a version with its full test and its statistical analysis published, with the same metrics, so the evolution can be compared. Versions add up: the latest released one carries everything before it and the fixes for what its tests found.
StableIts evidence says so, not its number: in the latest run of each entity in its material, the lower bound of the 95% margin is 95% or more for signals and for data, and no control raised a signal. Each whole number has ten versions, from x.0 to x.9; then comes x+1.0.
0.12.0
ReleasedReleased on 2026-10-27
Ofelia, built and tested with the full round with the agent
Test
Round 1: Ofelia on the four offices, 14 of 14 signals and 76 of 78 data points, with one extra signal in a control; every defect it found, with its fix and its verification
Latest released
1.0.0
ReleasedReleased on 2026-10-29
Carries everything before it, with its fixes
Travel, and each provider's exports
Test
Round 2 with the agent on the four offices, each with its provider —Outlook or Google—: 85 of 85 signals and 144 of 146 data points on the first run, no signal in the controls; the defect it found, in the material, with its fix and its verification
The same roadmap lives in the repository, in families/criterio-ea/hoja-de-ruta.json.
How she works
Five steps and one boundary. The language model (LLM) extracts what only a written document holds; the code calculates everything else. It is generative AI plus a deterministic engine, applied to the executive’s office.
Your exports
The calendar as .ics, the emails saved as files and each meeting’s folder: notice, materials, notes and signed minutes. As they are, where they are.
She does not connect to the mailbox or the calendar.
The model extractsCalendar and mail
The engine processes the calendar —recurrences, exceptions and time zones— and the headers of each email. What it cannot process it declares, with the reason.
An ambiguous event is declared, not guessed.
The model extractsThe record
One per meeting and one per email. The agreements, the commitments, the decisions with their options and what each email asks for, every item with its citation.
“Not found” is a valid answer.
The model extractsThe engine
Calendar clashes, materials by the configured days, pending minutes, every agreement against its owner and date, every commitment against its evidence, and the deadlines.
Deterministic and tested.
The code calculatesSignals
Only what crossed a threshold: what clashes, what is missing, what falls due and what is waiting for a reply. If nothing crossed one, it says nothing.
Silence is the feature.
The code calculatesNo date, clash or deadline in a report comes from an estimate. And what is said nowhere is declared, not filled in.
Why this boundary matters →Seven invariants that hold her up
She prepares; she does not recommend or decide
A decision’s options go in the order the documents bring them. The engine rejects a record that ranks them.
She never sends, schedules, accepts or moves anything
She writes drafts only in 90-borradores/, with «BORRADOR» in the name and in the text. A person sends.
She writes only in what is hers
Her state, her configuration and her drafts. She does not edit original minutes, calendars or emails.
A commitment is not met by saying so
It is met only with a document to back it. Each belongs to the minutes where it was born.
The model extracts, the code calculates
Every date, clash, deadline or count in a report comes from the engine. If the engine does not give it, it is not written.
She does not process what is excluded
Excluded folders are stated in every report, and nothing from them goes into a record.
Not published by default
Her report goes into Rostrum only if the office’s configuration says so.
The detail
The commands with their input, process and report; the signals the code calculates; the folder structure; the sources, and the links to the repository.
In order of importance. The first one sets up the configuration the others need.
ea-setupThe configuration, the inventory and the governing bodies with their meetings
00-referencia, the excluded folders, the thresholds —days before a meeting for the materials, days for the minutes, date changes that count, maximum age of the calendar—, whether it publishes to Rostrum and the termsea-scanThe records of the meetings, the correspondence and the calendar
10-agenda, each meeting in organos/, the correspondence in 20-correspondencia and the itineraries in 30-viajes (F-02, F-03)ea-agendaThe executive's week: conflicts, trips against the calendar, meetings without materials and meetings without minutes
ea-prepThe meeting pack, with the record of each decision
20-material and prepares the record of each decision: the options the documents bring, the evidence for each with its citation, what is still unknown and the dates that set the deadlineea-minutesThe draft minutes, with each agreement and commitment cited
30-notas90-borradores, with «BORRADOR» in the name and in the text, and the list of what is missing to close itea-commitmentsThe status of the executive commitments, by governing body and by owner
ea-mailThe correspondence that needs a reply, with its drafts
90-borradores. It never sendsea-weeklyThe weekly summary: what arrived, what is due and what still needs preparing
ea-wakeWhat the office needs today; silent when nothing does
share-findingsA summary with no data to share, only if you say yes
.criterio/corridas/12 signals. Each is raised with its date, its owner if the minutes say so and the citation of the document; the days, the counts and the calendar’s age are each office’s configuration.
Ofelia · 12
meeting_conflictTwo of the executive's meetings overlap in the exported calendar
meeting_without_materialWithin the configured days before a governing-body meeting, its materials are not in the folder
meeting_without_minutesThe configured days after a meeting have passed and its signed minutes are not there
agreement_without_ownerAn agreement in the minutes has no owner
agreement_without_dateAn agreement in the minutes has no date
commitment_overdueA commitment is overdue and no document supports it as done
commitment_rescheduledA commitment was moved more times than configured
correspondence_unansweredA saved email asks for a reply and its deadline passed with no saved reply
decision_deadlineA pending decision has a date that sets it within the configured days
meeting_during_flightA meeting in the calendar falls during a flight in the executive's itinerary
meeting_while_awayAn in-person meeting in the executive's home city falls while the itinerary has them travelling in another city
trip_not_in_agendaA flight in the itinerary is not in the calendar: that time looks free
Recommended, not mandatory: Ofelia processes the folder as it is. Setup says which file is out of place and where it would go, without creating, moving or renaming anything. It looks for the office folder, proposes it with its reason and, if you say yes, marks it with a visible file, _CRITERIO_DIRECCION.txt. Among your documents, besides that mark, she only writes her drafts, in 90-borradores/.
One folder per executive being assisted. Inside it, the office's reference files, the calendar, the correspondence, the drafts Ofelia prepares and one folder per meeting of each governing body.
direccion/
└── <ejecutivo>/
├── 00-referencia/
├── 10-agenda/
├── 20-correspondencia/
├── 30-viajes/
├── 90-borradores/
└── organos/<órgano>/<AAAA-MM-DD>/
├── 10-convocatoria/
├── 20-material/
├── 30-notas/
└── 40-acta/
estado-direccion/<ejecutivo>/ · One executive's office
| Folder | What goes there | Agent |
|---|---|---|
00-referencia/ | The governing bodies being followed —board, executive committee, board committees—, their members and their meeting calendar | Ofelia |
10-agenda/ | The calendar exports in .ics, with the export date in the name | Ofelia |
20-correspondencia/ | The emails saved as files, one per message: .eml, or Outlook's .msg | Ofelia |
30-viajes/ | The itinerary of each trip —the agency's, the airline ticket—, one per trip, with the departure date in the name | Ofelia |
90-borradores/ | The only place among the documents where Ofelia writes: draft minutes and draft replies, with «BORRADOR» in the name | Ofelia |
organos/<órgano>/<AAAA-MM-DD>/ · One meeting of a governing body, with its date as the folder name
| Folder | What goes there | Agent |
|---|---|---|
10-convocatoria/ | The notice of meeting and the agenda | Ofelia |
20-material/ | The materials sent to members before the meeting | Ofelia |
30-notas/ | The notes or transcript of the meeting | Ofelia |
40-acta/ | The signed minutes. Ofelia never files a draft here | Ofelia |
estado-direccion/ · What Ofelia writes: the configuration, the record of each meeting, thread and export, and the reports. No person edits it, and Ofelia writes nowhere else except `90-borradores/`.
The role, from O*NET; the calendar and the email, from the standards that define their formats. Nothing is built on a source still to be verified.
O*NET · 43-6011.00 Executive Secretaries and Executive Administrative Assistants
Datos de 2026 (O*NET OnLine, sitio actualizado el 6 de octubre de 2026) · onetonline.org ↗
IETF · RFC 5545, Internet Calendaring and Scheduling Core Object Specification (iCalendar)
Septiembre de 2009, Proposed Standard; actualizada por las RFC 5546, 6868, 7529, 7953, 7986, 9073, 9074 y 9253 · rfc-editor.org ↗
IETF · RFC 5322, Internet Message Format
Octubre de 2008, Draft Standard; actualizada por la RFC 6854 · rfc-editor.org ↗
Installation, setup and terms
Three steps, and none of them asks you to edit a file. One instance per executive: if you assist two, each office has its own.
To receive every new version, turn on auto-update once: in /plugin, Marketplaces → criterio → Enable auto-update. Without it, the installed version does not change on its own.
What it needs from you
Each agent comes with its own setup command. It looks at your folders, asks a few questions and shows you a first result on your own documents.
/criterio-assistant:ea-setup
How it asks
It does not schedule itself. It comes with the command a clock calls (ea-wake), and someone sets it once as a scheduled task. If your organisation does not want unattended runs, it is run by hand.
Nobody reads a disclaimer in a repository. So the warning reaches the person who runs the agent and the person who later receives the document, in four layers that every agent implements.
One acceptance per version
Every version of the family is accepted again and leaves on your machine a receipt in Markdown and PDF, with the SHA-256 fingerprint of the terms and of the receipt itself. Until the one for the installed version is recorded, the agent answers questions but produces no full analysis. It never assumes it or infers it from you continuing to talk.
On every report
Every report carries a footer saying it is a draft, based on which documents, how many fields had no data and how many contradictions remain open. The document travels by email and reaches a committee: the warning travels with it.
At the start of every run
Before the first finding it says which records it loaded, when they were refreshed, and whether any is stale or missing.
In the legal notice
The full terms, the legal notice and the version register —what changed in each version, which terms govern it and their fingerprint— are published on this site, and every receipt points to them.
What you agree to, in plain words
Terms version 1.3, of 2026-10-28, for version 1.0.1 of the family. Every new version of the family asks for acceptance again.
Legal notice and version register →Apache 2.0
The method, the record schema, the code that computes and the acceptance criteria, in full and with no conditions. Use it inside your organisation without asking, adapt it, and build the capability you are missing: the agent for a function should be written by someone who knows the work.
What it does not do
Frequently asked questions
What it is
She prepares the work of the executive's office. She sets the week beside the meetings and the commitments, builds the pack for each governing body's next meeting with the record of each decision, drafts the minutes from the notes, follows the commitments in each set of minutes and says which email is waiting for a reply.
No. She prepares and checks: which meetings clash, which meeting has no materials, which minutes are still unsigned, which agreement has no owner or date. What is accepted, sent and put first is always decided by a person.
No. Each decision arrives with the options the documents bring, in the order they bring them, with the evidence for each one, what is still unknown and the date that sets the deadline. The engine rejects a record that ranks the options or marks one as recommended.
She does not send, schedule, accept or move anything. She does not edit original minutes, calendars or emails, does not declare a commitment met without a document to back it and does not evaluate a person or a team.
How it is used
The ones the office exports: the calendar as .ics, the emails saved as files, and each meeting's notice, materials, notes and signed minutes. She does not connect to the mailbox or the calendar: a direct connection is a separate decision, with its own privacy review.
As it is: Ofelia processes it that way. If you start from scratch, the family recommends one folder per executive with its reference files, calendar, correspondence, drafts and one folder per meeting of each governing body. Setup looks for the office folder, proposes it and, if you say yes, marks it with a visible file, _CRITERIO_DIRECCION.txt. Among your documents, besides that mark, she only writes her drafts, in 90-borradores/.
Within the first fifteen minutes, ea-setup shows the calendar for the next two weeks with its clashes, the meetings without materials and those still without minutes.
Any. Ofelia processes no invoices or regulations, only the calendar, the minutes and the email, so the office's country is a data point, not a limit. What does change from one office to another is how its calendar and email provider exports: that is what gets tested next.
In each command's report, in her state folder. Her report is not published in Rostrum unless the configuration says so.
How it is tested
On synthetic material from two countries —Colombia and Ecuador—: a president's office and a general manager's office with seeded defects, and one control office per country, with no defects at all, where any signal is a failure. Every signal is seeded at least once, and the material also tests the rules of real operation: a meeting in another time zone that only clashes after the clock change, an old calendar that gives no clashes, a draft put in 40-acta/ and an excluded folder.
The material carries a key with what was planted, and the engine never sees it. The test compares every signal by its name and its key: an extra one counts the same as a missing one. The gate that checks it runs on every change to the repository.
The engine is, against the key of the four offices. The agent's extraction —the agreements, commitments and decisions in each set of minutes, and what each email asks for— is tested in the round with the agent on the same four offices, published with every output, its cost and its grade. Nothing has run yet on a real executive's office.
Data and privacy
They are processed on the AI platform where the agent runs, like any conversation with it. The plugin does not connect to the internet or send us anything. The executive's office is among the most sensitive things an organisation has: that is why she processes exports, not accounts, and respects the folders you exclude.
On your computer: the meeting and correspondence records, the reports, the drafts, the log of each run and the receipt for accepting the terms, which each version asks for again.
The same answers live in the repository, in families/criterio-ea/PREGUNTAS.md.
Support the work
Criterio remains open source. A voluntary contribution helps fund testing, documentation and the next capabilities without changing access to the software.
This is a voluntary contribution. It does not unlock features, purchase consulting services or create an advisory relationship.
Support Criterio
Choose an amount and how you want to contribute.
Choose your country to see the payment methods available to you.
dLocal Go currently covers Latin America, Indonesia, Malaysia, Kenya and Nigeria. Global coverage will be added separately.
The idea
What the governing body agreed was always written down. Nobody had set it beside the executive’s week.
An assistant that processes the calendar and the minutes does not hand you a control you never built. She shows you, every week, what clashes, which agreement has no owner and which commitment moved three times, and leaves the decision where it was: with a person.
Principles
José Ñáñez