Fund Manager Mode
Fund Manager Mode is the Enterprise way to run an optimization under a written mandate. You state the rules your portfolio must obey, and the optimizer solves inside them rather than producing a portfolio and being corrected afterwards.
What it requires
An Enterprise plan, and at least one constraint. A fund-manager run without a mandate is refused rather than accepted and ignored, because a run that claims to follow rules it was never given is worse than a run that admits it had none.
Two separate entitlements sit behind it. One allows constraints; one allows the mode. They move independently on purpose, so constraints can reach another plan later without the mode moving with them.
The three things a mandate can say
Sector limits and market-cap limits are one family here, not two. They differ only in which classification you attach, and no solver distinguishes them.
| Rule | In plain terms | Methods that solve it directly |
|---|---|---|
| Per-stock bounds | A floor and a ceiling on each holding | 27 of 31 |
| Excluded stocks | Names the portfolio may not hold at all | 26 of 31 |
| Group limits | A cap or floor on a sector or a market-cap segment | 23 of 31 |
The Portfolio Constraints page carries the full per-method table, the mathematics each family uses, and the reason the hierarchical methods handle group limits differently.
There are two answers, not three
For any method and any rule, the answer is either solved directly, meaning the rule entered the program the method actually solves, or unsupported, meaning the rule cannot be expressed for that method at all.
There used to be a third answer. A method could allocate without the rules and then move the result toward them afterwards. Every such path was removed, because a repaired allocation is not a constrained optimum and presenting one as the other is the mistake this whole feature exists to avoid.
An unsupported combination is refused before the run is spent. The optimize form and the API both read the same server-side table, so a method that cannot take your mandate says so before you submit.
What happens before a run is spent
The whole mandate is checked at once. Every rule is compiled into one linear program over the filtered universe. Checking rules one at a time cannot see a conflict between two rules that each pass alone, and those are the conflicts that actually happen.
An impossible mandate names the conflict. The question after a refusal is not “is this impossible” but “which number do I move, and by how much”. The preflight answers the second one.
Reachable ranges are computed from your universe. A 40% technology cap means nothing if the names you selected can only reach 22%. The preflight reports what the universe can actually support, so a cap that can never bind is visible as such.
The preflight spends no run and no quota. You can check a mandate as often as you like.
What the reports attest to
Every PDF and workbook carries a Constraints section stating what was requested, what was enforced, and where the two differ. The stored weights are validated against the mandate they claim to satisfy, so a run cannot report a rule as enforced and store a portfolio that breaks it.
A backtest is the case to read carefully. A group rule needs a classification for each name, and a classification recorded today cannot be applied to a rebalance from years ago without knowing what was true then. Where that would happen the group rules are dropped for the backtest, and the report says so rather than quietly applying today’s sectors to the past.
What it does not add
Stated plainly, because the absence of a number is easier to plan around than a number that turns out to mean something else.
- Active share and active attribution. Both need dated benchmark constituent weights, which the platform does not hold.
- A forward tracking-error distribution, or a probability of underperforming the benchmark. Neither ships. A probability from the simulator is conditional on the model, the frozen inputs, the configuration, the engine version and the seed, and no generator has passed out-of-sample validation.
- SEBI category rules. These key to a sector or size definition the platform cannot yet assert.
- Market impact or execution cost. The liquidity measures report how long an exit would take at a stated participation rate. They do not price the trade.
Information ratio and tracking error are computed, on every plan, and have been for some time. Every Indian benchmark is a price index, so alpha and the information ratio are overstated by approximately the index dividend yield. Each affected value is marked on the results page and in every export. Tracking error, beta and M2 are not affected.
Using it from an assistant
The mode and the mandate are both settable over MCP. preview_constraints runs the same feasibility check the website runs, without spending a run, which is the right call before submitting anything carrying a group rule.
A sector map can be sent inline with the request, or saved and referenced by id. list_classification_maps returns the saved ones. Read AI Assistants for the full tool surface.
Not investment advice. Past performance is not indicative of future results.