How People Remember the Same Event Differently: How FindMeld Finds Hidden Connections

Ordinary search works well when the required information is already published: we enter a query, select keywords, and browse the results.

But how do you search for a person if one witness saw them in Germany, another — several days later in the Czech Republic, and a third doesn’t even know anyone announced a search? What do you do if everyone speaks their own language, describes appearances differently, and remembers time approximately?

Or another scenario: someone lost a document in one district, and it was found later in another city. Descriptions don’t match word-for-word, dates are approximate, and map points are far apart.

These are exactly the kinds of challenges that FindMeld addresses — a project of the MELD® system developed by DigiMeld UG.

After getting acquainted with the application, I would describe FindMeld not as a search system in the usual sense, but as a human memory matching system. It looks not for word coincidences but for probable connections between people, objects, and events.

Two sides of the same event

The main screen of FindMeld offers two primary scenarios:

• "I am searching";

• "I was a witness".

The first is designed for someone who needs witnesses or useful information. The second — for someone who saw, found, or remembers something.

This division contains one of the system’s main features.

Typically, witness searches start with an announcement: a person posts in social networks, local groups, messengers, hoping the right witness will be there at the right time.

In FindMeld, a witness can leave information even without knowing whether it may be needed. Meanwhile, a search request can appear in the system earlier or later than the witness’s message.

When the second side appears in the system, FindMeld checks whether both entries relate to the same event. Therefore, the order of their appearance doesn’t matter: the testimony can be waiting for a future search, and the search — for a future witness.

Searching by the overall picture, not just words

Let’s consider a road accident.

One person states:

"In the evening near the station, I saw a dark car. It was pouring rain, and the driver sharply turned into the parking lot."

Another describes the same situation in German:

"After work, near the train station, a gray van quickly turned into the parking lot. The road was wet."

For a typical search, these appear as different messages. Different languages, different timing references, and different descriptions of objects are used.

FindMeld analyzes the situation holistically:

• What event is being described;

• Who is searching and who provides information;

• Which people, objects, or vehicles are mentioned;

• Where and approximately when the event took place;

• What directions the participants were moving;

• Weather conditions and visibility;

• Unusual details that appear in both descriptions.

The dark car and the gray van can be the same vehicle if it was seen in rain during the evening. "Near the station" and "at the train station" may also refer to the same location.

The system doesn’t require verbatim matches; it tries to understand whether two people are describing the same real situation.

Accounting for human inaccuracies in advance

FindMeld inherently understands that people rarely remember events perfectly.

A witness may misremember the color, mention an approximate time, confuse streets, misjudge distances, or forget certain details.

Common filters exclude such discrepancies. FindMeld considers each detail in the context of the overall picture.

If two descriptions differ in the color of a car but match on location, direction, damage, weather, and time, differences in color don’t necessarily exclude a connection — especially if one observation was made during the day and the other at night or in the rain.

The system doesn’t blindly rely on human memory; it evaluates which features could have been perceived incorrectly and which are more likely to be reliable.

Time matters but isn’t always precise

FindMeld considers the temporal aspect logically, not mechanically.

For example, if someone reports losing an item today, but another states it was found yesterday — such a link should be approached cautiously. The sequence should make sense: loss first, then discovery.

However, not all dates are equally reliable.

If one says, "It happened around the weekend," and the other — "Seems like Monday morning," with other features matching, the approximate date might carry less weight but shouldn’t be excluded.

It’s important to distinguish situations where:

• Confirmed temporal contradiction excludes the connection;

• Approximate or doubtful dates allow for lower match confidence if other features are compelling.

FindMeld doesn’t ignore time altogether but doesn’t require absolute precision.

The role of weather in adjusting geographic information

Weather isn’t a decorative factor in the system.

If the witness notes the place and date, they might remember heavy rain or fog but misidentify the exact point on the map.

FindMeld checks weather conditions for the specified period: if no rain was recorded at the location but heavy precipitation was nearby, this can increase the likelihood of a match.

The search area can be shifted or expanded toward regions with similar conditions.

A person might mistake the district, exit ramp, or misjudge distances, but weather conditions are often more memorable than exact locations.

Weather doesn’t alone prove a match but serves as an additional tool to verify memories and refine search boundaries.

A scratch might be more important than color

Sometimes, the most significant clues are those that seem secondary.

Suppose a car is being searched for, and one witness calls it blue, another — gray. In typical searches, these appear as different cars.

But both mention:

• a long scratch on the right door;

• a broken tail light;

• a sticker on the lower part of the window;

• an unusual engine sound;

• a dent near the wheel arch.

Such details are far more important than differences in color.

The same applies to lost items: a suitcase could be repacked, a bicycle repainted, or a cover or tag removed from a found object. The exterior appearance may change, but scratches, repair traces, and unusual features remain.

FindMeld considers both obvious and indirect signs:

• damage;

• labels and stickers;

• missing elements;

• distinctive sounds;

• contents;

• packing method;

• circumstances of appearance;

• direction of movement.

It’s impossible to know in advance which detail will be decisive. Therefore, the system analyzes the entire description, not just a few fixed fields.

Searching for people and objects beyond specific locations

This is especially relevant in missing persons cases.

Suppose a person disappeared in Berlin. Relatives describe the last locations, clothing, appearance, presumed direction, and behavior.

Standard systems focus on the local area — Berlin and nearby districts.

FindMeld can expand the search worldwide.

The person might have taken a train, flown, or used a ride-share, ending up in another country. Witnesses from Prague, Vienna, or Warsaw can leave descriptions without knowing about the search in Germany.

Although appearance and clothes may have changed, more durable features will likely include:

• distinctive gait;

• a scar;

• an unusual gesture;

• a speech habit;

• noticeable clothing details;

• a pet accompanying them;

• item carried with them;

• particular behaviors.

The first geographic point becomes a starting point, not a boundary for the search.

The same logic applies to objects: a passport lost near a store may be found elsewhere — someone picked it up, transported it, or handed it to another institution.

Multilingualism doesn’t divide participants

Both searchers and witnesses can communicate in different languages.

Each describes the situation as convenient, and FindMeld matches the meanings of the records.

This is especially important in international contexts — train stations, airports, tourist areas, border regions.

For example, someone loses a passport in Germany and creates an announcement in Russian; a tourist who finds it speaks Spanish; an employee reports in German.

The common text search will see different words, but FindMeld analyzes the meaning, circumstances, location, time, damages, and features.

Users don’t need to guess the language in which the witness reports.

No public bulletin board

FindMeld doesn’t have a public wall where all incidents, lost items, and search requests are visible to everyone.

The user only sees their own appeals, witness messages, and matches identified by the system.

There’s no need to scroll through ads for hours or determine manually who might benefit from certain information — everything is handled by the system.

Contacts are revealed only after establishing a link between the request and the witness message and remain private.

FindMeld isn’t a social network or messenger. Its purpose is to find probable matches and connect two sides of a specific event.

Reward without intermediary platform

FindMeld is a paid product. Individual actions and submissions are paid according to the service terms.

But its model doesn’t rely on intermediaries between searcher and witness.

If a witness wishes to receive a reward, they can specify this when submitting information. After a match, sides coordinate the reward and terms of cooperation themselves.

FindMeld does:

• not set the reward amount;

• not accept or transfer money;

• not withhold commissions or act as a guarantor;

• not participate in financial transactions or disputes.

Users only pay for system functions. Mutual agreements are maintained between the parties.

Matches can appear later

If FindMeld doesn’t find anything right away, it doesn’t mean the search is over.

It may simply be that there are no suitable witnesses or items in the system at that moment.

A few days later, someone might leave a description of a witness, lost item, or person. The new record is checked and matched with ongoing searches.

A small detail that seems irrelevant now could become a missing piece of another event over time.

Are there alternatives to FindMeld?

Services for finding missing persons, lost items, and witnesses have existed for a long time — bulletin boards, local communities, map apps, and alert systems.

But among the studied solutions, I haven’t yet encountered a product that combines in one system all the key features of FindMeld:

• two independent participants — seeker and witness;

• matching regardless of the order of record appearance;

• search by meaning, not just exact words;

• working with multiple languages;

• consideration of human inaccuracies;

• checking temporal sequence;

• enhancing the importance of matches with weak temporal links;

• using weather to adjust geography;

• global search for missing persons;

• analysis of indirect signs and details;

• absence of a public feed;

• controlled contact disclosure;

• ability to agree on compensation without an intermediary.

Separate functions can be found in other services, but the uniqueness of FindMeld is in integrating them into a single system.

Standard search engines ask: "Where is a similar text published?"

FindMeld solves the reverse problem: "Are these people describing the same event, even if they speak different languages, make mistakes, or are in different countries?"

Search based on real memory

FindMeld doesn’t require participants to have identical words or perfect recollections.

You can misremember the color, misidentify the location, confuse the time, or even describe the event in another language.

The system matches meaning, temporal and geographic details, weather conditions, and characteristic signs, separating possible contradictions from natural human inaccuracies.

This is why the main object of search becomes not the text but the event itself.