The PtahCast Blog

Every agency has given a client a date like this at least once: "We'll have it done by the 15th." It sounds confident. It's usually a guess — a PM looked at the backlog, added a bit of buffer for the unknown, and picked a number that felt safe enough to say out loud. The problem isn't that the guess is careless. It's that a single date can't represent something that is, by nature, uncertain.

Software and creative work don't take a fixed amount of time. Two tickets that look identical on a board can take two days or two weeks, depending on what they turn up once someone starts. A Gantt chart hides this by assigning each task one duration and adding them up, which produces a date that looks precise and is almost never right. The precision is fake; the uncertainty didn't go away, it just stopped being visible.

Monte Carlo forecasting starts from the opposite assumption: work is variable, so a forecast should describe a range of outcomes, not a point. The method itself is simple. Take a team's actual history — how long tickets have really taken to move from start to done, and how many the team actually finishes in a given week. Then simulate the remaining backlog thousands of times, each time randomly resampling from that real history instead of from a plan. Some simulated runs get lucky and finish fast. Some hit a string of slow tickets. Run it 10,000 times and you don't get one date — you get a distribution of dates, weighted by how often each one actually occurred across all those simulated futures.

That distribution is what turns into the three numbers PtahCast shows on a forecast: an aggressive date (the P50 — the outcome in the middle, what happens in half of all simulated futures), a commercial date (the P85 — safe enough to put in front of a client, since 85% of simulated futures finished by then or earlier), and a safe date (the P95 — the number you can defend if someone asks what the worst case looks like). None of them are guesses. All three come directly from how the team has actually performed.

The practical difference shows up the first time a forecast is wrong in the old system versus the new one. With a single date, being wrong is a surprise and an apology. With a probability range, being outside the P85 some of the time is expected — it's the other 15%, and everyone already knew it was possible. That's not a lower bar. It's an honest one, and clients tend to trust honest more than they trust confident.

A forecast is only useful if someone can check it later. Most agencies never do this in any formal way — a PM says a date out loud or writes it into a Slack message, the project moves on, and by the time it actually ships (or doesn't), nobody goes back to compare the two. The forecast quietly disappears, whether it was right or wrong.

An accountability log is the fix: a permanent, dated record of every forecast a team has made for a piece of work, alongside the commitment date that was actually given to the client at the time, and what happened afterward. In PtahCast, that means every time a forecast is run, the Aggressive (P50), Commercial (P85), and Safe (P95) dates are saved along with the committed delivery date in effect on that day — and the log flags, plainly, whether the engagement is on track or already behind that commitment.

This sounds like a small bookkeeping feature. In practice it changes how forecasting gets used. Without a log, there's no cost to giving an optimistic date — nobody's checking, so an agency can quietly move the goalposts each time a deadline gets close, and the client only sees a single new date each time, never the pattern behind it. With a log, every forecast is on the record. If the commercial date keeps sliding out, that's visible immediately, not months later in a post-mortem. It creates a real incentive to give a defensible number the first time, because the alternative is a visible trail of having been wrong.

It also changes the conversation with clients in a good way. An agency that can show a client its forecast history — including the times it was behind — reads as more credible, not less, because the alternative most clients have experienced is agencies who never show their work at all. A single clean date with nothing behind it is easy to promise and easy to break. A dated, running record of forecasts versus outcomes is much harder to fake, and that's exactly what makes it worth something.

The accountability log isn't a punishment mechanism. It's what makes the rest of the forecast trustworthy — a Monte Carlo simulation is only as credible as the willingness to be checked against it later.