The Spellcraft System
The full Spellcraft system for inventing, modifying, and discovering spells through dedicated research. See the overview for what Spellcraft is and the design intent behind it.
Core mechanic
When a caster wants to modify a spell or invent a new one, they make an Intelligence (Arcana) check, with these modifiers:
- Spell level penalty. Subtract the target spell's level from the roll. Cantrip = -0, 5th-level = -5, 9th-level = -9.
- Research time bonus. Each full week of dedicated, uninterrupted study grants +1 to the roll. The maximum bonus is the caster's proficiency bonus.
You may choose to use the spellcaster's spellcasting ability modifier in place of Intelligence for the check. The approach described here intentionally favors wizards and artificers, who are already built around study and research.
Base spell required. Every spellcraft project takes a spell the caster currently knows as its base. The base spell defines the spell-level penalty and the proposed change defines the DC. A caster who does not know the base cannot research a derivative of it. They must acquire and learn it first through normal channels (teacher, scroll, captured spellbook). Spells that are themselves products of prior spellcraft count as known for the purposes of further iteration.
DC tiers
The DM sets the DC based on how ambitious the proposed change is.
| DC | Difficulty | What it covers |
|---|---|---|
| 5 | Trivial | Rename, reflavor, change visual manifestation (color, sound, sigil) |
| 10 | Easy | Change damage type or school |
| 15 | Moderate | Swap save vs. attack roll; reroll dice variance (e.g. 8d6 to 4d12) |
| 20 | Hard | New utility spell loosely modeled on an existing one |
| 25 | Very Hard | Trade single-target damage for AoE or multi-target |
| 30 | Almost Impossible | Genuinely new high-power magic (should probably be plot-gated). |
Results
| Margin | Outcome |
|---|---|
| Beat DC by 5+ | Clean success. The spell works exactly as designed. |
| Beat DC by less than 5 | Success with a quirk. DM picks something minor and flavorful from the Quirk Table: a visible side effect, an unusual component required, an audible signature, an unintended trigger. |
| Miss by 1-9 | Unstable. The spell works, but every casting triggers a roll on a chaos table (default: Unstable Surge; substitute Wild Magic Surge or the PHB / SRD version to taste). Recrafting can stabilize it (or make it worse). |
| Miss by 10+ | Research backfire. Roll immediately on the Mishap Table. The spell is not produced and the materials are consumed. |
Cost of an attempt
- Time. Each week of research is real downtime, so the caster otherwise can't adventure or earn income.
- Material components. Each attempt requires the gp value of the target spell's listed material components, multiplied by the spell's level. For spells without a listed gp value, use a baseline of 25gp × spell level. Cantrips have a minimum of 10gp per attempt. The caster must have access to a market or supplier (apothecary, magical components shop, alchemist's lab, etc.). Spellcraft is not viable in the wilderness.
- Failed attempts consume the materials. A miss does not refund the components.
What success produces
How the new spell enters the caster's repertoire depends on the caster's class:
- Spellbook casters (Wizards, Artificers): the spell is added to the caster's personal spellbook at the level it was crafted at. Prepared and cast like any other spell in the book.
- Prepared-from-list casters (Clerics, Druids, Paladins, Rangers): the spell joins the inventor's personal prepare-from list at the level it was crafted. It is not automatically added to the broader class list; other casters of the same class can pick it up only by being taught (see below).
- Known-spell casters (Sorcerers, Bards, Warlocks): the spell becomes one of their known spells, optionally replacing one currently known if at the class cap.
Teaching the spell
If the inventor learned their magic through study, training, or prayer (Wizards, Artificers, Clerics, Druids, Paladins, Rangers), they can teach their invented spell to another caster of the same class. The student makes their own Arcana check at -10 to the original DC (rediscovery from a teacher or surviving notes is far easier than fresh invention). The inventor decides who learns it, and can hoard, charge for instruction, or share freely.
If the inventor's magic is innate, inspired, or patron-granted (Sorcerers, Bards, Warlocks), there is no teaching mechanism. They can cast the spell, but they cannot transcribe or transmit it. Another caster wanting access would need to observe them cast and reverse-engineer a class-appropriate version through their own spellcraft project.
Cross-class teaching (a cleric's prayer to a wizard's spellbook, etc.) generally isn't possible without the recipient running their own spellcraft project to derive a class-appropriate version.
Recrafting an unstable spell
A caster with an unstable spell can attempt to fix it with a fresh research cycle:
- Same time and cost rules apply.
- DC of the recraft is the original DC + 2 (it is harder once the magic has set in a flawed pattern).
- A clean success removes the instability; the spell becomes reliable.
- A "success with a quirk" replaces the current instability with a flavor quirk picked by the GM.
- A miss leaves the existing instability in place AND triggers an immediate roll on the Mishap Table (the recraft itself backfires). Recrafting twice is risky.
Plot-gated magic
Some spells should not be invented via dice alone:
- Resurrection-class effects (true death-reversal, soul retrieval).
- Setting-banned magic (whatever your campaign treats as forbidden: necromancy, soul-binding, blood magic, mind domination).
- Effects with global consequences (mass mind-control, region-wide weather change, time travel).
These should require a story arc. Spellcraft can advance the arc but should not resolve it.
Variance-swap reference
The DC 15 ("dice variance swap") tier covers trades like 8d6 to 4d12 — same average damage, different swing. Two expressions are variance-swap equivalent if their averages, rounded up, match. Use the calculator to compare across die sizes:
The width of each row's distribution is the visual proxy for swing. A narrow shape clustered around the mean (11d4) is more steady and predictable. A wide shape spanning the whole axis (3d20-4) is more volatile.