Ravensight

See what a decision does to every kind of player

A decision rarely lands on one player. A port, a price change or a new mode helps one group, costs another and leaves a third untouched. A strategy matrix puts that on one page before you commit: your player segments down one side, the directions you are weighing across the top, and a score in every cell for each lens you care about. Open one in Build > Strategy matrix with a title, and optionally the question it answers, how far ahead it looks, and one of your titles and one of your games it is about. The list follows the game switcher, and All games shows every matrix, including one with no game. Start it blank, or seeded with your studio's own starting context, which calls no model and costs nothing. Read it as a sheet, one lens at a time, or as a stack with every lens layered, where each layer has a text summary.

Segments, directions and lenses

A segment is a group of players you name yourself, such as players who finished the campaign or players on a handheld, with an optional description and a weight for how much it matters to this decision. A direction is one option you are weighing, including doing nothing, with an optional description and, set in the dashboard, parameters such as a price. A lens is what you score it on, with an optional description and a weight. Every matrix starts with five: fit, revenue, friction, feasibility and risk, and you can add your own. Fit, revenue and feasibility count for a direction; friction and risk count against it. A matrix holds up to 12 segments, 10 directions and 10 lenses, and a studio can keep up to 100 matrices.

Each cell is a whole score from 1 to 5 with a confidence of low, medium or high and a rationale of up to 1,000 characters. The rollup for a direction is a weighted mean, lenses inside each segment and then segments, and it carries what it rests on: its confidence, how much of it is assumption, how many cells are contested and which cells contributed most. The gaps list what to look at next: cells not scored yet, assumptions, low confidence and contested cells, heaviest first. The rollup and the gaps count accepted cells only. Every cell shows its score as a digit, its confidence as a word and its basis as a letter and a shape, so nothing depends on colour.

Evidence, inference or assumption

Every score says what it rests on. Evidence means it cites at least one record: an insight from your studio context, one of your documents, a touchpoint, a playtest finding, a row in the sales ledger, a Pulse post, a brainstorm card, or a public fact from market research. A score cites up to eight sources, and each citation says whether that source supports the score or contradicts it. Inference means the score was derived from cited evidence rather than read straight off it. Assumption means nobody has evidence yet, and it is allowed to cite nothing. Every citation is checked when it is written, against your own studio or, for a market fact, against the public fact base, and only a reference is kept, never the source text, so a cell cannot point at evidence you do not have. When a market fact a cell cites is later disputed, superseded or retired, the cell is flagged so you can look again; its score is not changed for you.

Scenarios and diffs

A scenario is a branch of the matrix: a what if, such as the port slipping a quarter or the price landing lower. It can rescore cells of its own, and it can carry up to 20 rules that move every cell they name by a whole number of points, up to 4 either way, such as raising risk by one everywhere. A moved score stays between 1 and 5, and everything else stays on the base matrix. Compare a scenario with the base and the diff lists every cell that moved, from what to what and why: rescored in the scenario, moved by one of its rules, or scored only in the scenario. It also shows every direction's rollup and rank before and after. A matrix holds up to 20 scenarios.

Record the decision, then the outcome

When you choose, record the decision: the direction, the scenario it was made under if any, your reasoning, and up to ten predictions. A prediction is a metric, a comparison (at least or at most), a threshold, a date to judge it by and how likely you think it is. Resolve each prediction once. Day 1 and day 7 retention and new players are read from one of your games' events, and a touchpoint's lift from that touchpoint, over the time from the decision to the prediction's date, or to today if that date has not come yet. Any other metric is entered by hand, and so is a prediction that names a segment or has no game, title or touchpoint to read: the decision marks it as one Ravensight cannot pull.

Units sold and net revenue come from one title's sales ledger, the prediction's own or the matrix's, so they cannot name a segment, and they are pulled only once the prediction's date is over. Net revenue names one currency, its threshold and its outcome are whole minor units of that currency, and it is never converted and never typed in. The pull refuses rather than guesses when the ledger rows in the window mix daily and monthly periods, when a monthly row is only partly inside the window, or when some rows did not report the figure. Correct the ledger, or for units sold, enter the actual by hand.

Calibration compares what you predicted with what happened, as a hit rate and a Brier score across your resolved predictions. It needs at least five resolved predictions before it reports a number, and says so rather than showing one from too few. It is a check on your own judgement over time, not a forecast.

Partner suggestions, one AI unit

The partner can suggest segments for a matrix, or suggest scores for one direction at a time, from what your studio already knows. Each suggestion is one shared AI unit, the same unit a Design chat turn uses, it draws the included allowance first, and the dialog names the unit before anything is spent. Suggested segments are not stored: each one joins the matrix only when a person adds it. Suggested scores are stored as the partner's and stay proposed until a person accepts them, and a cell a person has already accepted is never overwritten. A suggested score that does not cite one of the sources it was shown is stored as an assumption. If the partner suggests nothing new, the unit comes back. Filling the whole matrix in one step is not offered.

Over MCP

Your own agent can read matrices, compare a scenario with its base and list decisions with list_strategy_matrices, get_strategy_matrix, diff_strategy_scenario and list_strategy_decisions. With a token an owner has allowed to write, it can also create a matrix, add segments and directions, set a score and record a prediction's outcome with create_strategy_matrix, add_strategy_segment, add_strategy_direction, set_strategy_score and record_prediction_outcome. Partner suggestions, scenarios, recording a decision, a direction's parameters and deletes stay in the dashboard. See the MCP tool cards.

The open core

The scoring core is a separate MIT package, ravensight-matrix: the schema, the rollup, the gaps, the scenario diff and calibration, with no I/O and no dependencies, so the dashboard and the server compute the same numbers. It ships a CommonJS entry and an ES module entry and runs on Node 18 or later. The source is on GitHub and the package on npm (npm install ravensight-matrix). Evidence checks, the partner, metering and MCP stay in the hosted service.